CN118628841B - Remote sensing image data processing method, device, equipment and storage medium - Google Patents

Abstract

The embodiment of the application provides a remote sensing image data processing method, a remote sensing image data processing device, remote sensing image data processing equipment and a storage medium, and relates to the technical field of remote sensing image processing. Selecting a target probability image from probability images of the remote sensing image, acquiring a time front probability image and a time rear probability image, selecting a target pixel from the target probability image, selecting a current category, calculating a spatial neighborhood similarity weight of the target pixel according to the target probability image, calculating a temporal neighborhood similarity weight of the target pixel according to the time front probability image and the time rear probability image, calculating a correction weight of the current category based on a probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, and updating a category result corresponding to the target pixel in the remote sensing image according to the correction weight. And the classification result is corrected by utilizing the rationality that the classification result of the long-time sequence remote sensing data changes back and forth in time and space and combining the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, so that the accuracy of the classification result is improved.

Description

Remote sensing image data processing method, device, equipment and storage medium

Technical Field

The present application relates to the technical field of remote sensing image processing, and in particular to a remote sensing image data processing method, device, equipment and storage medium.

Background Art

Long-term land cover classification in remote sensing image processing refers to the use of satellites or aircraft to repeatedly observe the ground for a long time, and obtain the type information of each pixel at multiple time points through supervised classification methods. For example, a pixel point is grassland in the first year, becomes a water body in the second year, becomes bare soil in the third year, and becomes a residential area from the fourth to the tenth year.

In related technologies, methods based on probability space modeling, such as Markov random fields, are used to improve classification accuracy. Markov random fields can reasonably constrain the consistency and changes of time series data to a certain extent. However, the estimation process of its global and local transition probabilities is relatively rough, resulting in limitations in solution accuracy and prior knowledge constraints.

Summary of the invention

The main purpose of the embodiments of the present application is to propose a remote sensing image data processing method, device, equipment and storage medium to improve the classification accuracy of remote sensing data.

To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a remote sensing image data processing method, comprising:

Classifying multiple time-series remote sensing images to obtain a probability image corresponding to each remote sensing image, wherein each pixel in the probability image includes a probability value of the remote sensing image at a position corresponding to the pixel under each surface category;

Selecting a target probability image corresponding to the target time from the probability images, and obtaining a first preset number of probability images before the target time as a pre-time probability image, and obtaining a second preset number of probability images after the target time as a post-time probability image;

Selecting a target pixel from the target probability image and selecting a current category from the surface category, calculating a spatial neighborhood similarity weight of the target pixel under the current category according to the target probability image, and calculating a temporal neighborhood similarity weight of the target pixel under the current category according to the pre-time probability image and the post-time probability image;

The correction weight of the target pixel under the current category is calculated based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, and the correction weight corresponding to each surface category in the target pixel is calculated, and the category result corresponding to the target pixel in the remote sensing image is updated according to the correction weight.

In some embodiments, calculating the spatial neighborhood similarity weight of the target pixel under the current category according to the target probability image includes:

Determining spatial neighborhood pixels corresponding to the target pixel from the target probability image according to a preset neighborhood pixel range;

The probability values corresponding to the spatial neighborhood pixels under the current category are added to obtain the spatial neighborhood similarity weight corresponding to the current category.

In some embodiments, the temporal neighborhood similarity weight includes a temporal similarity weight and a temporal shift weight, and the step of calculating the temporal neighborhood similarity weight of the target pixel under the current category according to the temporal pre-probability image and the temporal post-probability image includes:

Calculate the time similarity weight of the target pixel under the current category according to the time-previous probability image and the time-postvious probability image;

Select the first probability image before the target time from the before-time probability image as the first transition probability image, and select the first probability image after the target time from the after-time probability image as the second transition probability image;

The temporal transfer weight of the target pixel in the current category is calculated according to the first transfer probability image and the second transfer probability image.

In some embodiments, calculating the temporal similarity weight of the target pixel according to the pre-temporal probability image and the post-temporal probability image includes:

Selecting neighboring pixels at the same position as the target pixel in the time-previous probability image and the time-postvious probability image, and combining the target pixel and the neighboring pixels to obtain a time-neighboring pixel;

The probability values of the temporal neighborhood pixels under the current category are added to obtain the temporal similarity weight corresponding to the current category.

In some embodiments, calculating the temporal transfer weight of the target pixel according to the first transfer probability image and the second transfer probability image includes:

Based on the spatial neighborhood pixels of the target pixel, acquiring a first transition pixel of the first transition probability image, and acquiring a second transition pixel of the second transition probability image;

For each surface category, based on the same pixel position, multiply the probability value of the surface category of the first shifted pixel by the probability value corresponding to the current category of the spatial neighboring pixel to obtain a first product value, sum the first product values within the range corresponding to the spatial neighboring pixels to obtain a first candidate value under the surface category, and obtain a second candidate value corresponding to the surface category according to the second shifted pixel;

The maximum value of the first candidate value and the maximum value of the second candidate value are added to obtain the time transfer weight corresponding to the current category.

In some embodiments, the calculating the modified weight of the target pixel in the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight includes:

Obtaining a first weight coefficient of the spatial neighborhood similarity weight, a second weight coefficient of the temporal similarity weight, and a third weight coefficient of the temporal transfer weight;

Obtaining a first revised value of the first weight coefficient and the spatial neighborhood similarity weight, a second revised value of the second weight coefficient and the temporal similarity weight, and a third revised value of the third weight coefficient and the temporal transfer weight;

The probability value, the first correction value, the second correction value, and the third correction value of the target pixel in the current category are accumulated to obtain the correction weight of the current category.

In some embodiments, the calculating the modified weight of the target pixel in the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight further includes:

Obtaining an invalid transition probability matrix corresponding to the surface category, wherein the invalid transition probability matrix includes invalid transition probability values between each of the surface categories;

Obtaining an invalid transfer weight of the target pixel under the current category according to the first transfer probability image, the second transfer probability image and the invalid transfer probability matrix;

obtaining a fourth weight coefficient of the invalid transfer weight, and obtaining a fourth correction value of the fourth weight coefficient and the invalid transfer weight;

The fourth correction value is subtracted from the correction weight to update the correction weight.

In some embodiments, obtaining the invalid transfer weight of the target pixel under the current category according to the first transfer probability image, the second transfer probability image and the invalid transfer probability matrix includes:

Based on the spatial neighborhood pixels of the target pixel, selecting a first probability transition pixel from the first transition probability image, and selecting a second probability transition pixel from the second transition probability image;

The surface category corresponding to the first probability transfer pixel is used as the pre-transfer type, and the surface category corresponding to the second probability transfer pixel is used as the post-transfer type;

Acquire the pre-transfer type and the pre-transfer probability of the current type according to the invalid transition probability matrix, and acquire the post-transfer type and the post-transfer probability of the current type;

For each surface category, based on the same pixel position, the probability value of the surface category of the pre-transfer probability is multiplied by the probability value corresponding to the current category of the spatial neighboring pixels to obtain a third product value, the third product value is summed within the range corresponding to the spatial neighboring pixels to obtain a third candidate value under the surface category, and a fourth candidate value corresponding to the surface category is obtained according to the post-transfer probability;

The maximum value of the third candidate value and the maximum value of the fourth candidate value are added together to obtain the invalid transfer weight under the current category.

To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a remote sensing image data processing device, comprising:

A surface category classification module is used to classify multiple time-series remote sensing images to obtain a probability image corresponding to each remote sensing image, wherein each pixel in the probability image includes a probability value of the remote sensing image at the corresponding position of the pixel under each surface category;

A probability image selection module is used to select a target probability image corresponding to a target time from the probability image, and obtain a first preset number of probability images before the target time as a pre-time probability image, and obtain a second preset number of probability images after the target time as a post-time probability image;

A similarity weight calculation module: used to select a target pixel from the target probability image, select a current category from the surface category, calculate the spatial neighborhood similarity weight of the target pixel under the current category according to the target probability image, and calculate the temporal neighborhood similarity weight of the target pixel under the current category according to the pre-time probability image and the post-time probability image;

Surface category correction module: used to calculate the correction weight of the target pixel under the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, calculate the correction weight corresponding to each surface category in the target pixel, and update the category result corresponding to the target pixel in the remote sensing image according to the correction weight.

To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a storage medium, which is a storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

The remote sensing image data processing method, device, equipment and storage medium proposed in the embodiment of the present application obtain a probability image corresponding to each remote sensing image by classifying multiple time series remote sensing images, wherein each pixel in the probability image includes the probability value of the remote sensing image at the corresponding position image of the pixel under each surface category. Next, the target probability image corresponding to the target time is selected from the probability image, and the probability images of the first preset number before the target time are obtained as the probability image before time, and the probability images of the second preset number after the target time are obtained as the probability image after time. The target pixel is selected from the target probability image, and the current category is selected from the surface category, and the spatial neighborhood similarity weight of the target pixel under the current category is calculated according to the target probability image, and the temporal neighborhood similarity weight of the target pixel under the current category is calculated according to the probability image before time and the probability image after time. Finally, the correction weight of the target pixel under the current category is calculated based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, and the correction weight corresponding to each surface category in the target pixel is calculated, and the category result corresponding to the target pixel in the remote sensing image is updated according to the correction weight. After obtaining the classification results of the remote sensing image, the embodiment of the present application utilizes the rationality of the classification results of the long-term remote sensing data in time and space, and combines the spatial neighborhood similarity weight and the temporal neighborhood similarity weight to correct the classification results, thereby reducing pseudo-changes and improving the accuracy of the classification results.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a remote sensing image data processing method provided in an embodiment of the present application.

FIG. 2 is a schematic diagram of multiple time-series remote sensing images provided in an embodiment of the present application.

FIG. 3 is a schematic diagram of a probability image provided in an embodiment of the present application.

FIG4 is a schematic diagram of an embodiment of the present application for calculating the spatial neighborhood similarity weight of a target pixel in the current category based on a target probability image.

FIG. 5 is a schematic diagram of spatial neighborhood pixels provided in an embodiment of the present application.

FIG6 is a flowchart of calculating the temporal neighborhood similarity weight of a target pixel in the current category based on a temporal pre-probability image and a temporal post-probability image provided by an embodiment of the present application.

FIG7 is a schematic diagram of calculating the temporal similarity weight of a target pixel based on a temporal pre-probability image and a temporal post-probability image provided by an embodiment of the present application.

FIG8 is a flowchart of calculating the temporal transfer weight of a target pixel according to the first transfer probability image and the second transfer probability image provided by an embodiment of the present application.

9 is a flowchart of calculating the modified weight of the target pixel in the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight provided by an embodiment of the present application.

FIG10 is another flowchart provided in an embodiment of the present application for calculating the modified weight of the target pixel in the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight.

11 is a flowchart of obtaining the invalid transfer weight of a target pixel in the current category according to the first transfer probability image, the second transfer probability image and the invalid transfer probability matrix provided by an embodiment of the present application.

FIG. 12 is a schematic diagram showing the implementation effect of the remote sensing image data processing method provided in an embodiment of the present application.

FIG13 is a structural block diagram of a remote sensing image data processing device provided in yet another embodiment of the present application.

FIG. 14 is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

It should be noted that although the functional modules are divided in the device schematic and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

Long-term land cover classification in remote sensing image processing refers to the use of satellites or aircraft to repeatedly observe the ground for a long time to obtain the type information of each pixel at multiple time points through supervised classification methods, where the size of the pixel is related to the resolution. For example, a 30-meter resolution corresponds to a surface area of approximately 30 meters * 30 meters at the imaging location. Each pixel of the result data corresponding to the type information corresponds to a time series, representing a quarterly or yearly type sequence of the corresponding imaging range. For example, a pixel point is grassland in the first year, becomes a water body in the second year, becomes bare soil in the third year, and becomes a residential area from the fourth to the tenth year.

In the related art, methods based on probability space modeling, such as Markov random fields, are used to improve classification accuracy. The probability space modeling method uses a classifier that can output a probability vector. The dimension of the probability vector is the same as the number of types. The elements of each dimension correspond to a type, and its value corresponds to the probability that the current point is classified as that type. In the probability space modeling method, the input features and classification methods at each time can be different, and the classification results correspond to pixels to form a time series. Although the Markov random field can reasonably constrain the consistency and changes of time series data to a certain extent. However, the estimation process of its global and local transition probabilities is relatively rough, resulting in limitations in solution accuracy and prior knowledge constraints.

Specifically, the accuracy of the solution is mainly reflected in the consistency of the time series results. For example, a forest and a bare land appear in the remote sensing image of the previous time period. In the remote sensing image of the next time period, the forest actually remains unchanged, while the bare land becomes grassland. However, in the classification results, the forest part of the next time period is mistakenly classified as bushes, while the sparse grassland is still misclassified as bare land. At this time, the bushes classified in the latter time period are pseudo-changes, and it fails to identify the real change of bare land turning into grassland. Considering that the actual change ratio of the surface is usually not high, most of the changes in the time series in the classification results may be pseudo-changes. Therefore, it is necessary to correct the classification results to reduce the occurrence of pseudo-changes and miss as few real changes as possible to obtain accurate classification results.

Based on this, the embodiments of the present application provide a remote sensing image data processing method, apparatus, equipment and storage medium. After obtaining the classification results of the remote sensing image, the rationality of the classification results of long-term remote sensing data in time and space is utilized, and the classification results are corrected in combination with the spatial neighborhood similarity weights and the temporal neighborhood similarity weights, thereby reducing pseudo-changes and improving the accuracy of the classification results.

The embodiments of the present application provide a remote sensing image data processing method, apparatus, device and storage medium, which are specifically described through the following embodiments. First, the remote sensing image data processing method in the embodiments of the present application is described.

The remote sensing image data processing method provided in the embodiment of the present application relates to the technical field of remote sensing image processing. The remote sensing image data processing method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, and can also be a computer program running in a terminal or a server side. For example, a computer program can be a native program or software module in an operating system; it can be a native application (APP, Application), that is, a program that needs to be installed in an operating system to run, such as a client that supports remote sensing image data processing, or it can be a small program, that is, a program that can be run only by downloading it to a browser environment; it can also be a small program that can be embedded in any APP. In short, the above-mentioned computer program can be an application, module or plug-in in any form. Among them, the terminal communicates with the server through a network. The remote sensing image data processing method can be executed by a terminal or a server, or by a terminal and a server in collaboration.

In some embodiments, the terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, or a smart watch. In addition, the terminal may also be an intelligent vehicle-mounted device. The intelligent vehicle-mounted device applies the remote sensing image data processing method of this embodiment to provide related services and enhance the driving experience. The server may be an independent server, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms; or a service node in a blockchain system, wherein each service node in the blockchain system forms a peer-to-peer (P2P, Peer To Peer, P2P) network, and the P2P protocol is an application layer protocol running on the Transmission Control Protocol (TCP) protocol. The terminal and the server may be connected via Bluetooth, Universal Serial Bus (USB) or a network or other communication connection methods, which are not limited in this embodiment.

The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

The remote sensing image data processing method in the embodiment of the present application is described below.

FIG1 is an optional flow chart of a remote sensing image data processing method provided in an embodiment of the present application, and the method in FIG1 may include but is not limited to steps 110 to 140. It is also understood that the present embodiment does not specifically limit the order of steps 110 to 140 in FIG1, and the order of steps may be adjusted or some steps may be reduced or added according to actual needs.

Step 110: Classify multiple time-series remote sensing images to obtain a probability image corresponding to each remote sensing image.

In one embodiment, the time sequence refers to a plurality of continuous acquisition times, and each acquisition time acquires a remote sensing image corresponding to that time, wherein each pixel in the remote sensing image corresponds to an image area divided according to actual needs. Referring to Figure 2, Figure 2 is a schematic diagram of remote sensing images of multiple time sequences provided in an embodiment of the present application. In Figure 2, T1, T2, T3 and T4 are used as four continuous acquisition times, and corresponding remote sensing images are obtained respectively, and the remote sensing images obtain pixels of the same size according to the same area division method. Taking the pixel in the upper right corner as an example for explanation, the pixel in phase T1 is a grassland, which is represented as "grass" in the figure, the pixel in phase T2 is a water body, which is represented as "water" in the figure, the pixel in phase T3 is bare land, which is represented as "soil" in the figure, and the pixel in phase T4 is a residential area, which is represented as "residential" in the figure.

It is understandable that the embodiments of the present application do not limit the method of collecting remote sensing images of different phases. The remote sensing data can come from different sensor types, such as optical sensors, synthetic aperture radar (SAR), etc., and they only need to be aligned pixel by pixel.

In one embodiment, the remote sensing images of each phase are then sent to a classifier for data supervised classification to obtain a probability image corresponding to the remote sensing images of each phase. Referring to Figure 3, Figure 3 is a schematic diagram of a probability image provided by an embodiment of the present application. Among them, each pixel in the probability image corresponds to a probability vector, the dimension of the probability vector is consistent with the number of pre-defined surface categories, and the element value on each dimension corresponds to the probability value of a surface category. This probability value is the probability that the remote sensing image is divided into a certain surface category at the corresponding position of the pixel, so the larger the probability value, the higher the possibility that the pixel is divided into the corresponding category. Taking the lower right corner pixel X as an example, assuming that there are 8 surface categories, this position corresponds to an 8-dimensional probability vector.

In one embodiment, based on the classification of multiple surface categories, such as 10 categories or more, usually only 2 to 3 probability values in the probability vector of each pixel are large, and the other probability values are very close to 0. Therefore, the embodiment of the present application can only output the first few probability values in the probability vector and the corresponding surface categories, and set the probability values of other surface categories to 0, thereby reducing the intermediate results that need to be stored and improving data processing efficiency.

It can be understood that for probability images of multiple time series, if they are divided in time order, the probability value of the pixel in the probability image of each time phase can directly correspond to the classification result in the remote sensing image at the current acquisition time. If divided by pixel, the probability value of the pixel at the same position in the probability images of different time series can reflect the change of the terrain at that position over time.

Step 120: Select a target probability image corresponding to the target time from the probability images, obtain a first preset number of probability images before the target time as the before-time probability image, and obtain a second preset number of probability images after the target time as the after-time probability image.

In one embodiment, assuming that time t is selected as the target time, the probability image corresponding to the remote sensing image at time t is used as the target probability image. Assuming that the first preset number and the second preset number are both 5, it means that with time t as the center, 5 probability images are selected forward and backward along the time direction, and the 5 probability images before time t are used as the probability images before time, and the 5 probability images after time t are used as the probability images after time. It can be understood that, first of all, the first preset number and the second preset number may not be equal. Secondly, the corresponding numerical value can be set according to the actual image processing requirements.

Step 130: Select a target pixel from the target probability image and select a current category from the surface category, calculate the spatial neighborhood similarity weight of the target pixel in the current category based on the target probability image, and calculate the temporal neighborhood similarity weight of the target pixel in the current category based on the pre-time probability image and the post-time probability image.

In one embodiment, when correcting the result, if the computing power of the computing device is insufficient, the result of selecting the pixels of the key part in the remote sensing image for correction can be considered, and if the computing power of the computing device is sufficient and the actual accuracy requirement is high, correction can be performed pixel by pixel. Therefore, according to the actual situation, the pixels that need to be corrected can be selected one by one from the target probability image as the target pixels, and this embodiment does not limit this.

In one embodiment, after the target pixel is selected, each or some of the surface categories may be corrected based on actual accuracy requirements, and the surface categories that need to be corrected are selected one by one from the surface categories as the current category.

Next, we describe how to modify the current category of the target limit.

In one embodiment, considering that a type with a probability value slightly smaller than the maximum probability value may also be a correct result, and that spatially adjacent pixels are more likely to be of the same type, a certain proportion of spatially adjacent pixels changing at the same time may be a true change, so the consistency of the spatially adjacent results is used for auxiliary correction. Referring to FIG4 , FIG4 is a schematic diagram of an embodiment of the present application calculating the spatial neighborhood similarity weight of a target pixel under the current category based on a target probability image, specifically comprising the following steps:

Step 410: Determine spatial neighborhood pixels corresponding to the target pixel from the target probability image according to a preset neighborhood pixel range.

In one embodiment, the preset neighborhood pixel range may be a rectangular pixel range of preset length centered on the target pixel. The size of the preset length Ns here can be set according to actual needs. For example, if Ns is 5 or 7, the corresponding preset neighborhood pixel range is or .

Then, the spatial neighborhood pixels of the target pixel are delineated from the target probability image according to the preset neighborhood pixel range. It can be considered that the category of the spatial neighborhood pixels is associated with the category of the target pixel to a certain extent. Referring to FIG. 5 , FIG. 5 is a schematic diagram of the spatial neighborhood pixels provided by an embodiment of the present application. In FIG. 5 , Ns is 5, and pixel X is taken as the target pixel. At this time, the target pixel is located at The central area of the target pixel is taken as the spatial neighborhood pixels of the target pixel, including the target pixel.

Step 420: Add the probability values corresponding to the spatial neighborhood pixels under the current category to obtain the spatial neighborhood similarity weight corresponding to the current category.

In one embodiment, the probability values of all spatial neighborhood pixels under the current category are added together to obtain the spatial neighborhood similarity weight corresponding to the target pixel under the current category. . Combined with Figure 5, that is, The probability values of all pixels in the area in the current category are added together to obtain the spatial neighborhood similarity weight.

In one embodiment, considering that a small amount of changes before and after time are mostly pseudo changes caused by classifier discrimination errors, the consistency of the temporal proximity results is used for auxiliary correction. Referring to FIG6 , FIG6 is a flowchart of calculating the temporal neighborhood similarity weight of the target pixel under the current category based on the temporal probability image before and the temporal probability image after provided by an embodiment of the present application, specifically including the following steps:

Step 610: Calculate the temporal similarity weight of the target pixel in the current category according to the temporal pre-probability image and the temporal post-probability image.

In one embodiment, the classification results of the same pixel position under continuous time series usually do not change significantly, so the impact of continuous time series on the consistency of results can be considered. In other words, when there is a strong similarity before and after time, it means that there is a high probability of consistency and no change, and it should be given priority over other types of conversions. For example, the overall forest probability in the previous remote sensing image is 0.9, the overall grassland probability in the current remote sensing image is 0.5 and the forest is 0.4, and the overall forest in the next remote sensing image is also 0.9. In the related art, the Markov random field uses the type with the largest probability to calculate the transfer matrix, which will strengthen the probability of forest conversion to grassland, and grassland conversion to forest, so that the current remote sensing image is strengthened as grassland in the final classification result. In fact, the remote sensing images before and after the time are all forests. Even if the probability of forest in the prediction result of the current remote sensing image is only slightly lower than that of grassland, it should be inferred that the current type is more likely to be forest. Referring to Figure 7, Figure 7 is a schematic diagram of calculating the time similarity weight of the target pixel based on the probability image before time and the probability image after time provided by the embodiment of the present application, which specifically includes the following steps:

Step 710: Selecting neighboring pixels at the same position as the target pixel in the pre-temporal probability image and the post-temporal probability image, and combining the target pixel and the neighboring pixels to obtain a temporal neighborhood pixel.

In one embodiment, based on the position of the target pixel, pixels at the same position are selected from multiple time-previous probability images and time-postvious probability images as neighborhood pixels of the target pixel, and then the target pixel and the neighborhood pixels are combined to form the temporal neighborhood pixels of the target pixel. It can be understood that, for example, in the above example, the number of the time-previous probability image and the time-postvious probability image is 5, and the neighborhood pixels obtained at this time also include the corresponding 10, and the number of temporal neighborhood pixels is 11.

Step 720: Add the probability values of the temporal neighborhood pixels under the current category to obtain the temporal similarity weight corresponding to the current category.

In one embodiment, the probability values of all pixels in the temporal neighborhood pixels under the current category are added to obtain the temporal similarity weight corresponding to the target pixel under the current category. For example, if there are 11 temporal neighborhood pixels and the current category is the third surface category, the probability values corresponding to the third element value in the probability vector of the 11 pixels are added together to obtain the temporal similarity weight.

Step 620: Select the first probability image before the target time from the pre-time probability image as the first transition probability image, and select the first probability image after the target time from the post-time probability image as the second transition probability image.

Step 630: Calculate the temporal transfer weight of the target pixel in the current category according to the first transfer probability image and the second transfer probability image.

In one embodiment, considering that the consistency of the classification results of the adjacent time of the target time has a greater impact on the target time, the embodiment of the present application also quantifies the classification transfer possibility of the pixels in the spatial domain under the condition of adjacent time sequence. First, based on the target time, the first probability image before the target time is selected as the first transfer probability image, and the first probability image after the target time is selected as the second transfer probability image. It can be understood that, according to the time sequence relationship, the order of the probability images is: the second transfer probability image, the target probability image and the first transfer probability image.

After obtaining the first transition probability image and the second transition probability image, refer to FIG. 8 , which is a flowchart of calculating the time transition weight of the target pixel according to the first transition probability image and the second transition probability image provided by an embodiment of the present application, specifically including the following steps:

Step 810: Based on the spatial neighborhood pixels of the target pixel, obtain a first transition pixel of the first transition probability image, and obtain a second transition pixel of the second transition probability image.

The pixels in the first transition probability image corresponding to the spatial neighborhood pixels of the target pixel are used as the first transition pixels, and the pixels in the second transition probability image corresponding to the spatial neighborhood pixels of the target pixel are used as the second transition pixels. It can be seen that the number of the first transition pixels and the second transition pixels is consistent with the number of the spatial neighborhood pixels.

Step 820: For each surface category, based on the same pixel position, multiply the probability value of the surface category of the first transferred pixel by the probability value corresponding to the current category of the spatial neighboring pixel to obtain a first product value, sum the first product values within the range corresponding to the spatial neighboring pixels to obtain a first candidate value under the surface category, and obtain a second candidate value corresponding to the surface category according to the second transferred pixel.

In one embodiment, a first candidate value and a second candidate value need to be calculated for each surface category. Taking the surface category CLS as an example, if the neighborhood coordinate range of the spatial neighborhood pixel is represented as S, the probability values of all pixels (including the target pixel) in the neighborhood coordinate range S in the current type are first obtained. If the neighborhood coordinate range S is as shown in Figure 5, the current type is the first surface category, and the first element value in the probability vector of each pixel is selected to obtain a The probability value matrix Ps.

Next, taking the i-th pixel in the preset neighborhood pixel range as an example, assuming that the i-th spatial neighborhood pixel is in the probability value matrix The corresponding element value in is , the probability value corresponding to the surface category CLS of the i-th first transfer pixel at the corresponding position is expressed as , the probability value corresponding to the surface category CLS of the i-th second transfer pixel is expressed as .

The first product value of the i-th first transfer pixel under the ground surface category is expressed as:

Therefore, the first candidate value under the surface category is expressed as:

in, It means to sum i within the neighborhood coordinate range S.

Similarly, the corresponding second product value is calculated according to the second transfer pixel, and then the second candidate value corresponding to the surface category is obtained according to the second product value, which is expressed as:

in, Represents the second product value of the i-th second transferred pixel under the ground surface category.

It is understandable that the first product value and the second product value may be repeatedly calculated multiple times during the calculation process of the first candidate value and the second candidate value. They can be stored and the corresponding multiplication results can be directly read in the subsequent calculation process, thereby reducing the calculation complexity and improving the calculation efficiency.

Step 830: Add the maximum value of the first candidate value and the maximum value of the second candidate value to obtain the time transfer weight corresponding to the current category.

Among them, each surface category corresponds to a first candidate value and a second candidate value. Therefore, the maximum value is selected from the first candidate value and the second candidate value respectively, and the two maximum values are added together to obtain the influence of other surface categories on the current category in the continuous time series, which is used as the time transfer weight and expressed as:

in, represents the time transfer weight.

Step 140: Calculate the correction weight of the target pixel under the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, calculate the correction weight corresponding to each surface category in the target pixel, and update the category result corresponding to the target pixel in the remote sensing image according to the correction weight.

In one embodiment, after obtaining the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, the correction process can be performed. Referring to FIG. 9 , FIG. 9 is a flow chart of calculating the correction weight of the target pixel under the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight provided by an embodiment of the present application, specifically including the following steps:

Step 910: Obtain a first weight coefficient of a spatial neighborhood similarity weight, a second weight coefficient of a temporal similarity weight, and a third weight coefficient of a temporal transfer weight.

Among them, the first weight coefficient It is used to indicate the importance of spatial neighborhood similarity weight in the correction process, and the second weight coefficient The third weight coefficient is used to indicate the importance of time similarity weight in the correction process. It is used to indicate the importance of the time transfer weight in the correction process. It can be understood that the first weight coefficient , the second weight coefficient and the third weight coefficient They are all non-negative parameters and can be set according to actual conditions.

Step 920: Obtain a first revised value of the first weight coefficient and the spatial neighborhood similarity weight, a second revised value of the second weight coefficient and the temporal similarity weight, and a third revised value of the third weight coefficient and the temporal transfer weight.

Step 930: Accumulate the probability value, the first correction value, the second correction value and the third correction value of the target pixel in the current category to obtain the correction weight of the current category.

In one embodiment, the modified weight is expressed as:

Among them, W represents the correction weight, P represents the probability value of the target pixel under the current type, represents the first correction value, represents the second correction value, Indicates the third correction value.

In one embodiment, after obtaining the correction weight of the target pixel under the current category, the correction weights corresponding to other surface categories are calculated in the above manner, and the surface category corresponding to the maximum correction weight is selected as the category result corresponding to the target pixel in the remote sensing image at the target time. In this way, the category results of other target pixels in the remote sensing image are updated, thereby correcting the classification result of the remote sensing image.

In one embodiment, the probability vector of the target pixel can be updated using the correction weights after multiple iterations, and the above process can be repeated until the iteration end condition is obtained, and the classification result of the remote sensing image can be corrected according to the correction weights obtained in the last iteration.

In one embodiment, taking into account the use of some prior information, for example, if the land was bare land in the previous year, the possibility of it becoming a forest in the next year is very low, such a priori knowledge constraint can be added to increase the accuracy of the above-mentioned modified weights. Referring to FIG10 , FIG10 is another flow chart of calculating the modified weight of the target pixel in the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight provided by the embodiment of the present application, which specifically includes the following steps:

Step 1010: Obtain the invalid transition probability matrix corresponding to the surface category.

In one embodiment, the invalid transition probability matrix is obtained based on prior knowledge, which includes invalid transition probability values between two surface categories at two consecutive acquisition times. For example, if the number of surface categories is N, the dimension of the invalid transition probability matrix is , whose element values are used to indicate the possibility of transition between two surface categories. The purpose of the embodiment of the present application is to reduce the impact of category conversions with very small transfer possibilities in the results, and therefore the invalid transition probability values of the two surface categories that may be converted are set to be greater than the invalid transition probability values of the two surface categories that are unlikely to be converted. For example, sandy land may be converted into grassland, and the invalid transition probability value of the corresponding position is set to 0. It is basically impossible for sandy land to be converted into forest, and the invalid transition probability value of the corresponding position is set to 1. By analogy, the invalid transition probability matrix corresponding to the surface category is obtained based on prior knowledge. It can be understood that the invalid transition probability matrices obtained at different collection time intervals are different.

Step 1020: Obtain the invalid transfer weight of the target pixel in the current category according to the first transfer probability image, the second transfer probability image and the invalid transfer probability matrix.

In one embodiment, a probability image is selected before and after the target probability image in the time sequence, that is, a first transition probability image and a second transition probability image. Referring to FIG. 11 , FIG. 11 is a flowchart of obtaining the invalid transfer weight of the target pixel under the current category according to the first transition probability image, the second transition probability image and the invalid transition probability matrix provided by an embodiment of the present application, specifically including the following steps:

Step 1110: Based on the spatial neighborhood pixels of the target pixel, select a first probability transition pixel from the first transition probability image, and select a second probability transition pixel from the second transition probability image.

In one embodiment, the spatial neighborhood pixels are Take the first transition probability image as an example. The pixel at the corresponding position is taken as the first probability transfer pixel, and the pixel in the second transfer probability image is selected. The pixel at the corresponding position is used as the second probability transfer pixel.

Step 1120: The ground surface category corresponding to the first probability transfer pixel is used as the pre-transfer category, and the ground surface category corresponding to the second probability transfer pixel is used as the post-transfer category.

In one embodiment, for the first probability transfer pixel, the surface category corresponding to the maximum value of the probability vector of each pixel is used as its type result, and all the type results are used as the type before transfer. Similarly, the type after transfer corresponding to the second probability transfer pixel is obtained.

Step 1130: Obtain the pre-transfer type and the pre-transfer probability of the current type according to the invalid transition probability matrix, and obtain the post-transfer type and the post-transfer probability of the current type.

In one embodiment, according to the time sequence relationship, the pre-transfer type may become the current type in the next time sequence, and the current type may become the post-transfer type in the next time sequence. Therefore, the pre-transfer type and the pre-transfer probability of the current type are obtained according to the invalid transition probability matrix, and the post-transfer type and the post-transfer probability of the current type are obtained. For example, if the current type is mountainous, then for The first probability transition pixel corresponding to the range is determined based on the invalid transition probability matrix to determine the pre-transition probability of its type changing from pre-transition type to mountainous area. The second probability transition pixel corresponding to the range is used to determine the post-transition probability of the mountain changing to the post-transition type according to the invalid transition probability matrix. It can be understood that if the invalid transition probability matrix uses numbers 0 and 1 to represent the transition possibility, the obtained pre-transition probability and post-transition probability are also composed of numbers 0 and 1. matrix.

Step 1140: For each surface category, based on the same pixel position, multiply the probability value of the surface category of the probability before transfer by the probability value corresponding to the current category of the spatial neighboring pixels to obtain a third product value, sum the third product values within the range corresponding to the spatial neighboring pixels to obtain the third candidate value under the surface category, and obtain the fourth candidate value corresponding to the surface category according to the probability after transfer.

In one embodiment, it is assumed that the preset neighborhood pixel range is , the probability value of the temporal neighborhood pixel in the current category is The matrix of the pre-transfer probability and the post-transfer probability is also Matrix. Taking the i-th pixel in the preset neighborhood pixel range as an example, assuming that the i-th spatial neighborhood pixel is in the probability value matrix The corresponding element value in is , the probability before the i-th transfer of the corresponding position is expressed as , the probability after the i-th transfer is expressed as .

The third product value of the i-th pre-transfer probability under the surface category is expressed as:

Therefore, the third candidate value under the surface category is expressed as:

in, It means to sum i within the neighborhood coordinate range S.

Similarly, the corresponding fourth product value is calculated according to the probability after transfer, and then the fourth candidate value corresponding to the surface category is obtained according to the fourth product value, which is expressed as:

in, Represents the fourth product value of the i-th second transferred pixel under the ground surface category.

Step 1150: Add the maximum value of the first intermediate value and the maximum value of the second intermediate value to obtain the invalid transfer weight under the current category.

In one embodiment, each surface category corresponds to a third candidate value and a fourth candidate value, so the maximum value is selected from the third candidate value and the fourth candidate value respectively, and the two maximum values are added to obtain the invalid transfer weight under the current category, which is expressed as:

in, Indicates invalid transfer weight under the current category.

It is understandable that if the current type and all other surface categories can be converted to each other, the invalid transfer weight is 0.

Step 1030: Obtain a fourth weight coefficient of the invalid transfer weight, and obtain the fourth weight coefficient and a fourth correction value of the invalid transfer weight.

In one embodiment, the fourth weight coefficient for indicating the importance of the invalid transfer weight in the correction process is selected according to the actual situation. .

Step 1040: Subtract the fourth correction value from the correction weight to update the correction weight.

In one embodiment, the updating process of the modified weight is expressed as:

in, represents the updated correction weight, Indicates the fourth correction value.

The following describes the implementation effect of the remote sensing image data processing method provided in the embodiment of the present application.

In one embodiment, the classification results obtained by multi-year classification of the same area are taken as an example. Referring to Figure 12, Figure 12 is a schematic diagram of the implementation effect of the remote sensing image data processing method provided by the embodiment of the present application. The first row in Figure 12 is respectively the legend, the corrected classification results of the first year, and the corrected classification results of the second year, and the second row is respectively: the corrected classification results of the third year, the corrected classification results of the fourth year, and the corrected classification results of the fifth year. The classification results of these five years are matched pixel by pixel, and each pixel corresponds to the type of each year at that position. Among them, it is unreasonable for the types of forests, grasslands, and farmland in the initial classification results to change back and forth in the sequential years. After the processing enhancement of the embodiment of the present application, most of the unreasonable changes have been corrected.

In one embodiment, the implementation effect of the remote sensing image data processing method provided by the embodiment of the present application is related to the specific remote sensing data, the measurement area and the set surface category. The embodiment of the present application uses the indicators of the accuracy of each time classification result and the accuracy of the before and after changes to evaluate the implementation effect. After processing the measured data, it is found that the embodiment of the present application can improve the accuracy of the single time classification result and the accuracy of the before and after changes, among which the accuracy of the before and after changes is greatly improved.

In one embodiment, the classification accuracy is evaluated by a method of calculating a confusion matrix by sampling. An example of a confusion matrix is shown in Table 1 below. Taking the bush column as an example, among A1+B1+C1+D1 bush samples, A1 is correctly classified; the bush row indicates that among A1+A2+A3+A4 samples classified as bush, A1 is correct; overall accuracy = (A1+B2+C3+D4)/total number of samples).

Table 1. Example of confusion matrix

The accuracy of surface changes is mainly verified by sampling comparison method. For example, sampling the changing and unchanged points respectively, and calculating the four indicators described in Table 2 below.

Table 2 Examples of land surface change indicators

The percentage of correctly changed and unchanged samples to the total number of samples in Table 2 is the total accuracy. In the actual multi-temporal land cover classification results, the proportion of false changes is high.

For example, in a certain area, Landsat data was used to classify the 30-meter surface cover for 5 years, and 7 surface categories were defined, including farmland, forest, grassland, shrub, water body, impermeable layer, and bare land. Among them, the overall accuracy of the original classification accuracy in the first year was 75.2%, and the classification accuracy after the remote sensing image data processing of the embodiment of the present application was 81.9%. When performing sampling verification of the change type, sampling was performed on the basis of the classification corresponding to the original probability image. 50 pixels were extracted for each of the pixels that changed and the pixels that remained unchanged in the first and second years, including 41 pixels of pseudo-change, 9 pixels of correct change, 10 pixels of missed detection, and 40 pixels of correct unchanged. The total accuracy is (9+40)/100=49%. After the remote sensing image data is processed according to the embodiment of the present application (these transfers are defined as impossible transfers: impermeable layer to forest, farmland to forest, water body to forest, grassland to forest), the pseudo-change is corrected to 5 pixels, the correct change pixels are corrected to 45 pixels, the missed detection is corrected to 9 pixels, the correct unchanged pixels are corrected to 41 pixels, and the total accuracy is improved to (45+41)/100=86%.

The technical solution provided by the embodiment of the present application is to obtain a probability image corresponding to each remote sensing image by classifying multiple time-series remote sensing images, wherein each pixel in the probability image includes the probability value of the remote sensing image at the corresponding position image of the pixel under each surface category. Next, the target probability image corresponding to the target time is selected from the probability image, and the probability images of the first preset number before the target time are obtained as the probability image before time, and the probability images of the second preset number after the target time are obtained as the probability image after time. The target pixel is selected from the target probability image, and the current category is selected from the surface category, and the spatial neighborhood similarity weight of the target pixel under the current category is calculated according to the target probability image, and the temporal neighborhood similarity weight of the target pixel under the current category is calculated according to the probability image before time and the probability image after time. Finally, the correction weight of the target pixel under the current category is calculated based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, and the correction weight corresponding to each surface category in the target pixel is calculated, and the category result corresponding to the target pixel in the remote sensing image is updated according to the correction weight. After obtaining the classification results of the remote sensing image, the embodiment of the present application utilizes the rationality of the classification results of the long-term remote sensing data in terms of time and space, and corrects the classification results in combination with the spatial neighborhood similarity weights and the temporal neighborhood similarity weights, thereby reducing pseudo-changes, making the changes of the long-term classification results consistent and reasonable, and improving the accuracy of the classification results.

The present application also provides a remote sensing image data processing device, which can implement the remote sensing image data processing method. Referring to FIG. 13 , the device includes:

The surface category classification module 1310 is used to classify multiple time series remote sensing images to obtain a probability image corresponding to each remote sensing image. Each pixel in the probability image includes the probability value of the remote sensing image at the corresponding position of the pixel under each surface category.

Probability image selection module 1320: used to select the target probability image corresponding to the target time from the probability image, and obtain the first preset number of probability images before the target time as the before-time probability image, and obtain the second preset number of probability images after the target time as the after-time probability image.

Similarity weight calculation module 1330: used to select target pixels from the target probability image and select the current category from the surface category, calculate the spatial neighborhood similarity weight of the target pixel under the current category based on the target probability image, and calculate the temporal neighborhood similarity weight of the target pixel under the current category based on the pre-time probability image and the post-time probability image.

Surface category correction module 1340: used to calculate the correction weight of the target pixel under the current category based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, calculate the correction weight corresponding to each surface category in the target pixel, and update the category result corresponding to the target pixel in the remote sensing image according to the correction weight.

The specific implementation of the remote sensing image data processing device of this embodiment is basically the same as the specific implementation of the remote sensing image data processing method described above, and will not be described in detail here.

The present application also provides an electronic device, including:

at least one memory;

at least one processor;

at least one program;

The program is stored in the memory, and the processor executes the at least one program to implement the remote sensing image data processing method implemented in the present application. The electronic device can be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a car computer, etc.

Please refer to FIG. 14 , which schematically shows the hardware structure of an electronic device according to another embodiment. The electronic device includes:

The processor 1401 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

The memory 1402 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1402 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1402, and the processor 1401 calls and executes the remote sensing image data processing method of the embodiment of this application;

Input/output interface 1403, used to implement information input and output;

Communication interface 1404, used to realize communication interaction between the device and other devices, which can be realized by wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.); and

A bus 1405 that transmits information between various components of the device (e.g., the processor 1401, the memory 1402, the input/output interface 1403, and the communication interface 1404);

The processor 1401 , the memory 1402 , the input/output interface 1403 and the communication interface 1404 are connected to each other in communication within the device via a bus 1405 .

An embodiment of the present application further provides a storage medium, which is a storage medium storing a computer program, and the computer program implements the above-mentioned remote sensing image data processing method when executed by a processor.

As a non-transitory storage medium, the memory can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

The remote sensing image data processing method, device, equipment and storage medium proposed in the embodiment of the present application obtain a probability image corresponding to each remote sensing image by classifying multiple time series remote sensing images, wherein each pixel in the probability image includes the probability value of the remote sensing image at the corresponding position image of the pixel under each surface category. Next, the target probability image corresponding to the target time is selected from the probability image, and the probability images of the first preset number before the target time are obtained as the probability image before time, and the probability images of the second preset number after the target time are obtained as the probability image after time. The target pixel is selected from the target probability image, and the current category is selected from the surface category, and the spatial neighborhood similarity weight of the target pixel under the current category is calculated according to the target probability image, and the temporal neighborhood similarity weight of the target pixel under the current category is calculated according to the probability image before time and the probability image after time. Finally, the correction weight of the target pixel under the current category is calculated based on the probability value of the target pixel, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight, and the correction weight corresponding to each surface category in the target pixel is calculated, and the category result corresponding to the target pixel in the remote sensing image is updated according to the correction weight. After obtaining the classification results of the remote sensing image, the embodiment of the present application utilizes the rationality of the classification results of the long-term remote sensing data in time and space, and combines the spatial neighborhood similarity weight and the temporal neighborhood similarity weight to correct the classification results, thereby reducing pseudo-changes and improving the accuracy of the classification results.

The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules/units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And/or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and/or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character "/" generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store programs.

The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.

Claims (11)

1. A remote sensing image data processing method, comprising:

Classifying the remote sensing images with a plurality of time sequences to obtain probability images corresponding to each remote sensing image, wherein each pixel in the probability images comprises a probability value of the corresponding position image of the remote sensing image in each earth surface category;

selecting target probability images corresponding to target time from the probability images, acquiring a first preset number of probability images before the target time as time front probability images, and acquiring a second preset number of probability images after the target time as time rear probability images;

Selecting a target pixel from the target probability image, selecting a current category from the earth surface category, calculating a spatial neighborhood similarity weight of the target pixel under the current category according to the target probability image, and calculating a temporal neighborhood similarity weight of the target pixel under the current category according to the time-before probability image and the time-after probability image;

And calculating the correction weight of the target pixel under the current category based on the probability value, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight of the target pixel, calculating to obtain the correction weight corresponding to each earth surface category in the target pixel, and updating the category result corresponding to the target pixel in the remote sensing image according to the correction weight.

2. The method according to claim 1, wherein said calculating spatial neighborhood similarity weights of the target pixels under the current class from the target probability image comprises:

determining a spatial neighborhood pixel corresponding to the target pixel from the target probability image according to a preset neighborhood pixel range;

and adding the probability values corresponding to the spatial neighborhood pixels under the current category to obtain the spatial neighborhood similarity weight corresponding to the current category.

3. The method according to claim 2, wherein the temporal neighborhood similarity weights include a temporal similarity weight and a temporal transition weight, and the calculating the temporal neighborhood similarity weight of the target pixel under the current category from the pre-temporal probability image and the post-temporal probability image includes:

Calculating the time similarity weight of the target pixel under the current category according to the time front probability image and the time rear probability image;

selecting a first probability image before the target time from the time-before probability images as a first transition probability image, and selecting a first probability image after the target time from the time-after probability images as a second transition probability image;

The time transition weight of the target pixel under the current category is calculated according to the first transition probability image and the second transition probability image.

4.A method of processing remote sensing image data according to claim 3, wherein said calculating said temporal similarity weight of said target pixel under said current category from said pre-temporal probability image and said post-temporal probability image comprises:

selecting a neighborhood pixel at the same position as the target pixel in the time front probability image and the time rear probability image, and combining the target pixel and the neighborhood pixel to obtain a time neighborhood pixel;

and adding the probability values of the time neighborhood pixels under the current category to obtain the time similarity weight corresponding to the current category.

5. A remote sensing image data processing method according to claim 3, wherein said calculating the temporal transition weight of the target pixel under the current class from the first transition probability image and the second transition probability image comprises:

acquiring a first transition pixel of the first transition probability image and a second transition pixel of the second transition probability image based on the spatial neighborhood pixel of the target pixel;

For each earth surface category, multiplying the probability value of the earth surface category of the first transfer pixel by the probability value corresponding to the current category of the spatial neighborhood pixel based on the same pixel position to obtain a first product value, summing the first product value in the range corresponding to the spatial neighborhood pixel to obtain a first candidate value under the earth surface category, and obtaining a second candidate value corresponding to the earth surface category according to the second transfer pixel;

And adding the maximum value of the first candidate value and the maximum value of the second candidate value to obtain the time transfer weight corresponding to the current category.

6. The method according to claim 3, wherein calculating the correction weight of the target pixel in the current category based on the probability value, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight of the target pixel comprises:

Acquiring a first weight coefficient of the space neighborhood similarity weight, a second weight coefficient of the time similarity weight and a third weight coefficient of the time transfer weight;

Acquiring a first correction value of the first weight coefficient and the spatial neighborhood similarity weight, a second correction value of the second weight coefficient and the time similarity weight, and a third correction value of the third weight coefficient and the time transfer weight;

And accumulating the probability value, the first correction value, the second correction value and the third correction value of the target pixel under the current category to obtain the correction weight of the current category.

7. The method of claim 6, wherein the calculating the correction weight of the target pixel in the current class based on the probability value, the spatial neighborhood similarity weight, and the temporal neighborhood similarity weight of the target pixel, further comprises:

obtaining an invalid transition probability matrix corresponding to the earth surface categories, wherein the invalid transition probability matrix comprises invalid transition probability values between every two earth surface categories;

obtaining an invalid transition weight of the target pixel under the current category according to the first transition probability image, the second transition probability image and the invalid transition probability matrix;

acquiring a fourth weight coefficient of the invalid transfer weight, and acquiring the fourth weight coefficient and a fourth correction value of the invalid transfer weight;

And subtracting the fourth correction value from the correction weight to update the correction weight.

8. The method according to claim 7, wherein the obtaining the invalid transition weight of the target pixel in the current class according to the first transition probability image, the second transition probability image, and the invalid transition probability matrix includes:

Selecting a first probability transition pixel from the first transition probability image and a second probability transition pixel from the second transition probability image based on the spatial neighborhood pixel of the target pixel;

taking the earth surface category corresponding to the first probability transfer pixel as a pre-transfer type, and taking the earth surface category corresponding to the second probability transfer pixel as a post-transfer type;

Acquiring the pre-transition probability of the pre-transition type and the current category according to the invalid transition probability matrix, and acquiring the post-transition probability of the post-transition type and the current category;

For each earth surface category, multiplying the probability value of the earth surface category of the probability before transition by the probability value corresponding to the current category of the spatial neighborhood pixel based on the same pixel position to obtain a third multiplication value, summing the third multiplication value in the range corresponding to the spatial neighborhood pixel to obtain a third candidate value under the earth surface category, and obtaining a fourth candidate value corresponding to the earth surface category according to the probability after transition;

and adding the maximum value of the third candidate value and the maximum value of the fourth candidate value to obtain the invalid transfer weight under the current category.

9. A remote sensing image data processing apparatus, comprising:

The earth surface category classification module: the method comprises the steps of classifying a plurality of time-sequence remote sensing images to obtain probability images corresponding to each remote sensing image, wherein each pixel in each probability image comprises a probability value of the corresponding position image of the remote sensing image under each earth surface category;

The probability image selecting module: the method comprises the steps of selecting target probability images corresponding to target time from the probability images, acquiring a first preset number of probability images before the target time as time front probability images, and acquiring a second preset number of probability images after the target time as time rear probability images;

And the similarity weight calculation module is used for: the method comprises the steps of selecting a target pixel from the target probability image, selecting a current category from the earth surface category, calculating a spatial neighborhood similarity weight of the target pixel under the current category according to the target probability image, and calculating a temporal neighborhood similarity weight of the target pixel under the current category according to the temporal pre-probability image and the temporal post-probability image;

The earth surface category correction module: and the correction weight of the target pixel under the current category is calculated based on the probability value, the spatial neighborhood similarity weight and the temporal neighborhood similarity weight of the target pixel, the correction weight corresponding to each earth surface category in the target pixel is calculated, and a category result corresponding to the target pixel in the remote sensing image is updated according to the correction weights.

10. An electronic device comprising a memory storing a computer program and a processor that when executing the computer program implements the remote sensing image data processing method of any one of claims 1 to 8.

11. A storage medium storing a computer program, wherein the computer program when executed by a processor implements the remote sensing image data processing method of any one of claims 1 to 8.