CN109872331A - A kind of remote sensing image data automatic recognition classification method based on deep learning - Google Patents
A kind of remote sensing image data automatic recognition classification method based on deep learning Download PDFInfo
- Publication number
- CN109872331A CN109872331A CN201910088823.7A CN201910088823A CN109872331A CN 109872331 A CN109872331 A CN 109872331A CN 201910088823 A CN201910088823 A CN 201910088823A CN 109872331 A CN109872331 A CN 109872331A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing images
- image
- pretreatment
- images
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 230000002708 enhancing effect Effects 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000012360 testing method Methods 0.000 claims abstract description 13
- 238000012546 transfer Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 5
- 238000003708 edge detection Methods 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 4
- 230000001629 suppression Effects 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000009966 trimming Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 11
- 230000011218 segmentation Effects 0.000 abstract description 7
- 238000013527 convolutional neural network Methods 0.000 description 17
- 210000002569 neuron Anatomy 0.000 description 14
- 238000003709 image segmentation Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000005192 partition Methods 0.000 description 2
- 230000001902 propagating effect Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- RWYFURDDADFSHT-RBBHPAOJSA-N diane Chemical compound OC1=CC=C2[C@H]3CC[C@](C)([C@](CC4)(O)C#C)[C@@H]4[C@@H]3CCC2=C1.C1=C(Cl)C2=CC(=O)[C@@H]3CC3[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@@](C(C)=O)(OC(=O)C)[C@@]1(C)CC2 RWYFURDDADFSHT-RBBHPAOJSA-N 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000002922 simulated annealing Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The remote sensing image data automatic recognition classification method based on deep learning that the invention discloses a kind of, the described method comprises the following steps: carrying out color transfer processing to remote sensing images, obtains remote sensing images after the first pretreatment；Remote sensing images carry out feature enhancing processing after pre-processing to first time, obtain remote sensing images after the second pretreatment；Data enhancing processing is carried out to remote sensing image after the second pretreatment, obtains remote sensing images after third pretreatment；Remote sensing images are divided into training set and test set after third is pre-processed, and then use training set training pattern, are finally tested with trained model test set.The present invention can effectively overcome the problems, such as that the limited applicable surface of traditional images cutting techniques, over-segmentation, edge smoothness is poor, segmentation accuracy rate is not high；The problem that neural network causes accuracy rate not high because remote sensing images bright-dark degree is different, resolution ratio is different, shooting angle is different is overcome simultaneously.
Description
Technical field
The present invention relates to image segmentation, image classification and deep learning field, it is related to full convolutional neural networks technology and distant
Feel technical field of image processing more particularly to a kind of remote sensing image data automatic recognition classification method based on deep learning.
Background technique
Traditional image partition method has very much, mainly includes following several: image segmentation based on edge is based on region
The image segmentation of growth method, the image segmentation based on clustering procedure, the image segmentation based on threshold value.Since remote sensing images have information
The problems such as amount is very big, object construction is complex, obscure boundary is clear, gray level is various, if carried out using the method for exhaustion
The selection of threshold value then will appear the lower problem of efficiency, therefore can be used that maximum likelihood is similar or genetic algorithm improves threshold
It is worth the efficiency chosen.But still to will appear limited segmentation result applicable surface, over-segmentation, edge flat for traditional image partition method
Slippery is poor, divides the not high problems of accuracy rate.
With the fast development of computer technology, a large amount of new method is constantly known applied to the parting of remote sensing images
Not, such as the methods of artificial neural network, support vector machines, fuzzy classification.Wherein Awad, M and DIANE M.MILLER et al.
Parting identification is carried out to remote sensing image data using the method for artificial neural network, and the texture information of image is combined to analyze.And
Mercier, G et al. using support vector machines to remote sensing images carried out parting identification research, the experimental results showed that support to
Amount machine is more preferable than classical supervised classification algorithm, and proposes the kernel function after some improvement on this basis to consider to prop up
The spectral similarity between vector is held, and reduces and reports phenomenon by mistake as caused by traditional kernel.In addition to this, fuzzy clustering is distant
Feel the important tool of satellite image unsupervised segmentation.Mukhopadhyay et al. proposes a kind of fuzzy poly- based on simulated annealing
Class method, and also combined in this way with support vector machines, to achieve the effect that improve the performance of fuzzy clustering.
In addition to this, due to good generalization ability, and image is rotated, translate and local deformation after will not shadow
The characteristics of ringing experimental result, convolutional neural networks (Convolution Neural Network, CNN) are also gradually widely used
It has arrived among the tasks such as image classification and segmentation.Convolutional neural networks are a kind of multi-level deep learning moulds based on biology
Type usually has better recognition effect and wider practicability.2015, Papandreou, George et al. were using deep
Degree convolutional neural networks (Deep Convolution Neural Network, DCNN) has developed to be divided for semantic image
Expectation-maximization (Expectation-Maximization, EM) method of model training.But due to by convolutional Neural net
Each pixel needs the original for the input for using surrounding block of pixels as convolutional neural networks when network is applied to image segmentation
Cause, the appearance of computational efficiency low the problems such as big so as to cause storage overhead.In addition to this, the input figure of convolutional neural networks
As size cannot be variation, all input image sizes will be consistent.
In order to solve this problem, Jonathan Long et al. proposed full convolutional neural networks (Fully in 2015
Convolutional Networks, FCN), this network structure replaces institute in Standard convolution neural network using convolutional layer
Some is fully connected layer, while can also keep the two-dimensional structure of image.The advantages of full convolutional neural networks are had clearly,
For example the input picture that uses of full convolutional neural networks is not necessarily to the size of arrowhead, does not need the size and test of training image
The size of image is consistent, and since full convolutional neural networks do not use block of pixels, so being not in calculating convolution sum
The problem of storing is repeated, so that full convolutional neural networks is more efficient.With more and more deep, the Gang Fu of research
Et al. propose a kind of based on full convolutional neural networks model is improved come the method classified to remote sensing images, pass through introducing
Atrous convolution improves the density of output class figure, and can accomplish the precise classification to high-resolution remote sensing image.But it leads
It is limited in that it needs Ground Truth (authentic and valid) label of a large amount of high quality to carry out model training, this very great Cheng
Explanation experience and a large amount of manual work on degree dependent on profession.
Summary of the invention
The remote sensing image data automatic recognition classification method based on deep learning that the present invention provides a kind of, the present invention can
Effectively overcome that the limited applicable surface of traditional images cutting techniques, over-segmentation, edge smoothness are poor, divide that accuracy rate is not high to ask
Topic；Overcome neural network simultaneously causes accurately because remote sensing images bright-dark degree is different, resolution ratio is different, shooting angle is different
The not high problem of rate, described below:
A kind of remote sensing image data automatic recognition classification method based on deep learning, the described method comprises the following steps:
Color transfer processing is carried out to remote sensing images, obtains remote sensing images after the first pretreatment；
Remote sensing images carry out feature enhancing processing after pre-processing to first time, obtain remote sensing images after the second pretreatment；
Data enhancing processing is carried out to remote sensing image after the second pretreatment, obtains remote sensing images after third pretreatment；
Remote sensing images are divided into training set and test set after third is pre-processed, and then use training set training pattern, finally use
Trained model tests test set.
It is wherein, described that color transfer processing is carried out to remote sensing images specifically:
Wherein, L, A, B indicate image channel value after transformation, L', L ", L " ', A', A ", A " ', B', B ", B " ' be intermediate become
Amount, l, a, b expression original image channel value, ml, ma, mb and ml', ma', mb' expression image channel mean value, nl, na, nb and nl',
Na', nb' indicate image channel standard variance.
Further, remote sensing images carry out feature enhancing processing after the pretreatment to first time specifically:
First using the image of remote sensing image after smooth first pretreatment of Gaussian filter, and noise is filtered out, then calculated
The gradient intensity of each pixel and direction in image out；
Using non-maxima suppression method eliminate edge detection generate spurious response, with dual threshold detection really and
Potential edge finally inhibits isolated weak edge.
It is wherein, described that data enhancing processing is carried out to remote sensing images after the second pretreatment specifically:
Conventional images are rotated, are translated, are scaled and trimming operation come realize data enhance, training when, by same
Data under varying environment and different data the enhancing operation of image, which are put together, to be trained.
Further, the method also includes:
The edge detail information of treated remote sensing images is added among image as an important information, it will be former distant
Image spreading is felt into the remote sensing images of one 5 dimension as input picture.
The beneficial effect of the technical scheme provided by the present invention is that: the present invention is by introducing depth learning technology, so that remote sensing
The accuracy rate that image data is classified automatically is higher, convergence is more preferable, the scope of application is wider；And it is pre-processed by volume of data
Operation, error caused by overcoming traditional neural network due to data are inconsistent.
Detailed description of the invention
Fig. 1 is a kind of flow chart of remote sensing image data automatic recognition classification method based on deep learning；
Fig. 2 is the structural schematic diagram of full convolutional neural networks；
Fig. 3 is characterized the schematic diagram of enhancing front and back image comparison；
Fig. 4 is experiment effect figure.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, embodiment of the present invention is made below further
Ground detailed description.
Embodiment 1
The embodiment of the present invention proposes a kind of remote sensing image data automatic recognition classification method based on deep learning, referring to
Fig. 1, the method includes the steps of:
101: color transfer processing being carried out to remote sensing images, obtains remote sensing images after the first pretreatment；
Because can exist between different remote sensing image datas bigger difference (such as bright-dark degree is different, resolution ratio not
Same, shooting angle difference etc.), so this method can be carried out using what Reinhard et al. was proposed for each color component
The method of color transfer.First two different images are all transformed under LAB (colour model), then calculate separately out two figures
As the standard deviation nl, nl ' and mean value ml, ml ' in LAB.Standard deviation nl, nl ' and mean value ml are obtained, after ml ', with target figure
The value l of each pixel of picture makes the difference with the mean value ml of target image, then with the standard of resulting difference L' and reference picture
Poor nl ' quadrature, then divided by the standard deviation nl of target image, finally again plus a reference picture mean value ml ' obtain one most
Result L afterwards calculates result and target image is transformed into rgb space again later.
102: remote sensing images carry out feature enhancing processing after pre-processing to first time, obtain remote sensing figure after the second pretreatment
Picture；
Since full convolutional neural networks cause using pond layer the loss of some inevitable image informations, to lead
Cause segmented image when will appear not obvious enough the problem of marginal information, so this method using Canny edge detection algorithm come into
The enhancing of row feature.
That is, first using the image of remote sensing images after smooth first pretreatment of Gaussian filter, and noise is filtered out, then count
The gradient intensity of each pixel and direction in image are calculated, edge detection is eliminated using the method for non-maxima suppression later and produces
Raw spurious response.Then finally isolated weak edge is inhibited really with potential edge with dual threshold detection.
103: data enhancing processing being carried out to remote sensing images after the second pretreatment, obtains remote sensing images after third pretreatment；
First conventional images rotated, translated, scaled and are cut etc. with operations to realize that data enhance.It is being trained
When, the data under the varying environment and different data enhancing operation of same image are put together and are trained.
104: remote sensing images are divided into training set and test set after third is pre-processed, and then use training set training pattern, most
Test set is tested with trained model afterwards.
Wherein, the step 104 is specific as follows:
When propagating forward, remote sensing images are put into input layer after first pre-processing third, operate to obtain by multiple convolution
The smaller and smaller characteristic pattern of a sheet by a sheet size.After by all convolutional layers, the thermal map of a minimum dimension will eventually get.
Then backpropagation is carried out, current convolution kernel can be restored in the picture by carrying out up-sampling operation to the thermal map
Feature.By the way that constantly iteration, each convolution operation all carry out deconvolution to the image obtained after up-sampling before forward, this
Sample can retain the detailed information in original image data, finally can entirely restore image and obtain final result.
In conclusion the embodiment of the present invention, which passes through, introduces depth learning technology, so that remote sensing image data was classified automatically
Accuracy rate is higher, convergence is more preferable, the scope of application is wider.
Embodiment 2
The scheme in embodiment 1 is further introduced below with reference to specific example, calculation formula, the present invention is real
The full convolutional neural networks structure of example use is applied as shown in Fig. 2, described below:
201: color transfer
First two different images are all transformed under LAB color space, by taking L as an example, two images is calculated separately out and exists
Standard deviation nl, nl ' and mean value ml, ml ' in LAB color space.It obtains standard deviation nl, nl ' and mean value ml, after ml ', uses mesh
The value l of each pixel of logo image makes the difference with the mean value ml of target image, then with resulting difference L' and reference picture
Standard deviation nl ' quadrature finally adds the mean value ml ' of a reference picture to obtain one again then divided by the standard deviation nl of target image
A last result L is calculated after result and target image to be transformed into rgb space again, specific formula such as formula (1), (2),
(3),(4)；
Wherein, L, A, B indicate image channel value after transformation, L', L ", L " ', A', A ", A " ', B', B ", B " ' be intermediate become
Amount, l, a, b expression original image channel value, ml, ma, mb and ml', ma', mb' expression image channel mean value, nl, na, nb and nl',
Na', nb' indicate image channel standard variance.
202: feature enhancing；
In order to become readily apparent from the edge of remote sensing images, Canny edge detection algorithm can be used to carry out feature
Enhancing, main flow are as follows: first carrying out smoothed image using Gaussian filter and filter out noise.Calculate each pixel in image
The gradient intensity of point and direction.Shown in specific formula such as formula (5), (6):
θ=arctan (Gy/Gx) (6)
Wherein, G is gradient intensity, and θ is gradient direction, GxFor pixel x-axis direction value, GyFor pixel y-axis direction value.
The spurious response that edge detection generates is eliminated using the method for non-maxima suppression later, is then detected with dual threshold
Really with potential edge, finally isolated weak edge is inhibited.It is characterized enhancing front and back effect pair as shown in Figure 3
Than, it can be seen that after carrying out feature enhancing using Canny edge detection algorithm, the edge letter of available original remote sensing images
Breath, and result is than more visible.
203: data enhancing；
Realize that data enhance by the way that conventional images are rotated, translate, scale and cut etc. with operations.It can make in this way
The type for obtaining remote sensing image data is more complete and perfect.When being trained, by the varying environment of same image and not
It puts together and is trained with the data under data enhancement operations, so that it may allow neural network learning to varying environment, different mould
The feature of formula, different types of remote sensing image data, the feature generality learnt in this way is stronger, and confidence level is higher, so as to
To obtain more accurate experimental result.
204: by treated, remote sensing image data is divided into training set and test set, then uses training set training pattern, most
Test set is tested with trained model afterwards.
Specific method is first to add the edge detail information of treated the remote sensing images important information additional as one
Enter among image, former remote sensing images can be thus extended to the remote sensing images of one 5 dimension as input picture.
When propagating forward, first input image data is put into input layer, is obtained by multiple convolution operation a sheet by a sheet
The smaller and smaller characteristic pattern of size.After by all convolutional layers, the thermal map of a minimum dimension will eventually get.
Then backpropagation is carried out, current convolution kernel can be restored in the picture by carrying out up-sampling operation to the thermal map
Feature.The main flow of backpropagation is as follows:
1) training set data is inputted to the input layer of artificial neural network, by obtaining output knot after multiple hidden layers
Fruit；
2) error of output result and actual result is calculated；
3) by calculated error amount backpropagation, hidden layer first is traveled to from output layer, then again by a series of hidden
It hides Es-region propagations and parameter is modified according to error amount to input layer, and in back-propagation process；
4) three above that iterates step, until to the last numerical value convergence.
Wherein, the calculating of error is constantly unfolded to input layer again by output layer to hidden layer, and specific formula is such as
Under: output layer:
Wherein, E1Indicate output layer error, dkIndicate k-th of neuron desired output of output layer, ykIndicate output layer kth
A neuron reality output, l are output layer neuron number.
Hidden layer:
Wherein, E2Indicate hidden layer error, dkIndicate k-th of neuron desired output of output layer, ωjkIndicate hidden layer the
Connection weight between k-th of neuron of j neuron and output layer, yjIndicate the value of j-th of neuron of hidden layer, f is
Sigmoid function, m are hidden layer neuron number.
Input layer:
Wherein, E3Indicate input layer error, dkIndicate k-th of neuron desired output of output layer, ωjkIndicate hidden layer the
Connection weight between k-th of neuron of j neuron and output layer, vijIndicate i-th of neuron of input layer and j-th of hidden layer
Connection weight between neuron, xiIndicate the input value of i-th of neuron of input layer, n is input layer number.
It calculates each layer error and then each layer error is carried out that partial derivative is asked to can be obtained by optimal weight ginseng respectively
Number, formula are as follows:
Wherein, η indicates learning rate.
By the way that constantly iteration, each convolution operation all carry out warp to the image obtained after up-sampling before forward
Product, can thus retain the detailed information in original image data, finally entirely can restore image to tie to the end
Fruit.
In conclusion the embodiment of the invention provides a kind of remote sensing image data automatic recognition classification based on deep learning
Method provides a kind of new approaches for the identification sorting technique in remote sensing images, and the present invention overcomes traditional images cutting techniques
The problem that applicable surface is limited, over-segmentation, edge smoothness is poor, segmentation accuracy rate is not high；Solves neural network because distant simultaneously
Feel the bottleneck that image bright-dark degree is different, resolution ratio is different, shooting angle is different and accuracy rate is not high.
Embodiment 3
Feasibility verifying is carried out to the scheme in Examples 1 and 2 below with reference to specific experimental data, it is as detailed below to retouch
It states:
The accuracy of experimental result is indicated with the accuracy of average classification and Kappa coefficient.Use the final segmentation of experiment
Result images are compared with the image manually marked, are then calculated the consistent pixel quantity of classification results and are accounted for whole picture pixel
The i.e. classification accuracy of the ratio of quantity, the accuracy averagely classified is exactly the accurate average value of all image classifications.Kappa
Coefficient is the measurement standard that a kind of pair of classification results precision calculated based on confusion matrix carries out statistical measurement.Kappa coefficient
Specific definition, as shown in formula (12), (13):
Wherein, k indicates Kappa coefficient.p0Indicate whole nicety of grading, i.e., the sum of the sample size that every one kind is correctly classified
The total sample number correctly classified divided by total sample number.peRepresentation theory nicety of grading.nk1And nk2Respectively represent the pre- of each classification
Survey the quantity of pixel and the quantity of actual pixels.N is the sum of sample, i.e. the total quantity of pixel.Specific experiment result such as 1 institute of table
Show:
1 experimental result data of table
Can be seen that by the experimental result data in table 1 is enhancing a series of numbers by color transfer, feature enhancing, data
The FCN network of Data preprocess, and addition detailed information and preprocess method improvement FCN network after, the standard of experimental result
True rate has a degree of promotion.
Pass through above-mentioned experimental result, it can be seen that this method is to the pretreatment of remote sensing image data and addition detailed information
Full convolutional neural networks contribute to improve the order of accuarcy of remote sensing images identification parting.From Fig. 4, (first is classified as former remote sensing figure
Picture, second is classified as Ground Truth, and third is classified as final segmentation result) in as can be seen that the effect integrally divided is preferable.
It will be appreciated by those skilled in the art that attached drawing is the schematic diagram of a preferred embodiment, the embodiments of the present invention
Serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (5)
1. a kind of remote sensing image data automatic recognition classification method based on deep learning, which is characterized in that the method includes
Following steps:
Color transfer processing is carried out to remote sensing images, obtains remote sensing images after the first pretreatment；
Remote sensing images carry out feature enhancing processing after pre-processing to first time, obtain remote sensing images after the second pretreatment；
Data enhancing processing is carried out to remote sensing image after the second pretreatment, obtains remote sensing images after third pretreatment；
Remote sensing images are divided into training set and test set after third is pre-processed, and then use training set training pattern, finally with training
Good model tests test set.
2. a kind of remote sensing image data automatic recognition classification method based on deep learning according to claim 1, special
Sign is, described to carry out color transfer processing to remote sensing images specifically:
Wherein, L, A, B indicate image channel value after transformation, L', L ", L " ', A', A ", A " ', B', B ", B " ' be intermediate variable, l,
A, b indicate original image channel value, ml, ma, mb and ml', ma', mb' indicate image channel mean value, nl, na, nb and nl', na',
Nb' indicates image channel standard variance.
3. a kind of remote sensing image data automatic recognition classification method based on deep learning according to claim 1, special
Sign is that remote sensing images carry out feature enhancing processing after the pretreatment to first time specifically:
First using the image of remote sensing image after smooth first pretreatment of Gaussian filter, and noise is filtered out, then calculates figure
The gradient intensity of each pixel and direction as in；
The spurious response that edge detection generates is eliminated using the method for non-maxima suppression, it is really and potential with dual threshold detection
Edge, finally isolated weak edge is inhibited.
4. a kind of remote sensing image data automatic recognition classification method based on deep learning according to claim 1, special
Sign is, described to carry out data enhancing processing to remote sensing images after the second pretreatment specifically:
Conventional images are rotated, are translated, are scaled and trimming operation come realize data enhance, training when, by same image
Varying environment and different data enhancing operation under data put together and be trained.
5. a kind of remote sensing image data based on deep learning described in any claim is known automatically in -4 according to claim 1
Other classification method, which is characterized in that the method also includes:
The edge detail information of treated remote sensing images is added among image as an important information, by former remote sensing figure
As being extended to the remote sensing images of one 5 dimension as input picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910088823.7A CN109872331A (en) | 2019-01-30 | 2019-01-30 | A kind of remote sensing image data automatic recognition classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910088823.7A CN109872331A (en) | 2019-01-30 | 2019-01-30 | A kind of remote sensing image data automatic recognition classification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109872331A true CN109872331A (en) | 2019-06-11 |
Family
ID=66918362
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910088823.7A Pending CN109872331A (en) | 2019-01-30 | 2019-01-30 | A kind of remote sensing image data automatic recognition classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109872331A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111008986A (en) * | 2019-11-20 | 2020-04-14 | 天津大学 | Remote sensing image segmentation method based on multitask semi-convolution |
| CN111091532A (en) * | 2019-10-30 | 2020-05-01 | 中国资源卫星应用中心 | Remote sensing image color evaluation method and system based on multilayer perceptron |
| CN112163549A (en) * | 2020-10-14 | 2021-01-01 | 中南大学 | Remote sensing image scene classification method based on automatic machine learning |
| CN112348823A (en) * | 2020-09-22 | 2021-02-09 | 陕西土豆数据科技有限公司 | Object-oriented high-resolution remote sensing image segmentation algorithm |
| CN113077405A (en) * | 2021-03-27 | 2021-07-06 | 荆门汇易佳信息科技有限公司 | Color transfer and quality evaluation system for two-segment block |
| CN113963249A (en) * | 2021-10-29 | 2022-01-21 | 山东大学 | Detection method and system of galaxy images |
| CN115082807A (en) * | 2022-06-10 | 2022-09-20 | 厦门精图信息技术有限公司 | Method, device, medium and equipment for training and interpreting remote sensing image encoder |
| CN118247280A (en) * | 2024-05-29 | 2024-06-25 | 中国海洋大学 | River section form migration simulation method based on multisource remote sensing data |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955902A (en) * | 2014-05-08 | 2014-07-30 | 国网上海市电力公司 | Weak light image enhancing method based on Retinex and Reinhard color migration |
| CN103955901A (en) * | 2014-05-08 | 2014-07-30 | 国网上海市电力公司 | Enhancing method of weak-illumination video image |
| CN107610141A (en) * | 2017-09-05 | 2018-01-19 | 华南理工大学 | A kind of remote sensing images semantic segmentation method based on deep learning |
| CN108596213A (en) * | 2018-04-03 | 2018-09-28 | 中国地质大学（武汉） | A kind of Classification of hyperspectral remote sensing image method and system based on convolutional neural networks |
| CN108765347A (en) * | 2018-05-30 | 2018-11-06 | 长光卫星技术有限公司 | A kind of color enhancement method of suitable remote sensing image |
| CN108805861A (en) * | 2018-04-28 | 2018-11-13 | 中国人民解放军国防科技大学 | Remote sensing image cloud detection method based on deep learning |
| CN108846380A (en) * | 2018-04-09 | 2018-11-20 | 北京理工大学 | A kind of facial expression recognizing method based on cost-sensitive convolutional neural networks |
| CN109255334A (en) * | 2018-09-27 | 2019-01-22 | 中国电子科技集团公司第五十四研究所 | Remote sensing image terrain classification method based on deep learning semantic segmentation network |
-
2019
- 2019-01-30 CN CN201910088823.7A patent/CN109872331A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955902A (en) * | 2014-05-08 | 2014-07-30 | 国网上海市电力公司 | Weak light image enhancing method based on Retinex and Reinhard color migration |
| CN103955901A (en) * | 2014-05-08 | 2014-07-30 | 国网上海市电力公司 | Enhancing method of weak-illumination video image |
| CN107610141A (en) * | 2017-09-05 | 2018-01-19 | 华南理工大学 | A kind of remote sensing images semantic segmentation method based on deep learning |
| CN108596213A (en) * | 2018-04-03 | 2018-09-28 | 中国地质大学（武汉） | A kind of Classification of hyperspectral remote sensing image method and system based on convolutional neural networks |
| CN108846380A (en) * | 2018-04-09 | 2018-11-20 | 北京理工大学 | A kind of facial expression recognizing method based on cost-sensitive convolutional neural networks |
| CN108805861A (en) * | 2018-04-28 | 2018-11-13 | 中国人民解放军国防科技大学 | Remote sensing image cloud detection method based on deep learning |
| CN108765347A (en) * | 2018-05-30 | 2018-11-06 | 长光卫星技术有限公司 | A kind of color enhancement method of suitable remote sensing image |
| CN109255334A (en) * | 2018-09-27 | 2019-01-22 | 中国电子科技集团公司第五十四研究所 | Remote sensing image terrain classification method based on deep learning semantic segmentation network |
Non-Patent Citations (3)
| Title |
|---|
| ADIA: "color transfer between images", 《COLOR TRANSFER BETWEEN IMAGES - 知乎 (ZHIHU.COM)》 * |
| ERIK REINHARD, MICHAEL ASHIKHMIN, BRUCE GOOCH, PETER SHIRLEY: "Color Transfer between Images", 《IEEE COMPUTER GRAPHICS AND APPLICATIONS》 * |
| 霍冠英等: "《侧扫声呐图像目标分割》", 31 December 2017 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091532A (en) * | 2019-10-30 | 2020-05-01 | 中国资源卫星应用中心 | Remote sensing image color evaluation method and system based on multilayer perceptron |
| CN111008986A (en) * | 2019-11-20 | 2020-04-14 | 天津大学 | Remote sensing image segmentation method based on multitask semi-convolution |
| CN111008986B (en) * | 2019-11-20 | 2023-09-05 | 天津大学 | Remote sensing image segmentation method based on multitasking semi-convolution |
| CN112348823A (en) * | 2020-09-22 | 2021-02-09 | 陕西土豆数据科技有限公司 | Object-oriented high-resolution remote sensing image segmentation algorithm |
| CN112163549A (en) * | 2020-10-14 | 2021-01-01 | 中南大学 | Remote sensing image scene classification method based on automatic machine learning |
| CN112163549B (en) * | 2020-10-14 | 2022-06-10 | 中南大学 | Remote sensing image scene classification method based on automatic machine learning |
| CN113077405A (en) * | 2021-03-27 | 2021-07-06 | 荆门汇易佳信息科技有限公司 | Color transfer and quality evaluation system for two-segment block |
| CN113963249A (en) * | 2021-10-29 | 2022-01-21 | 山东大学 | Detection method and system of galaxy images |
| CN113963249B (en) * | 2021-10-29 | 2024-04-09 | 山东大学 | Detection method and system for star image |
| CN115082807A (en) * | 2022-06-10 | 2022-09-20 | 厦门精图信息技术有限公司 | Method, device, medium and equipment for training and interpreting remote sensing image encoder |
| CN118247280A (en) * | 2024-05-29 | 2024-06-25 | 中国海洋大学 | River section form migration simulation method based on multisource remote sensing data |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109872331A (en) | A kind of remote sensing image data automatic recognition classification method based on deep learning | |
| CN113408605B (en) | Hyperspectral image semi-supervised classification method based on small sample learning | |
| CN106920243B (en) | Improved ceramic material part sequence image segmentation method of full convolution neural network | |
| CN111445488B (en) | Method for automatically identifying and dividing salt body by weak supervision learning | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN111753828B (en) | Natural scene horizontal character detection method based on deep convolutional neural network | |
| CN111553837B (en) | Artistic text image generation method based on neural style migration | |
| CN111986125B (en) | Method for multi-target task instance segmentation | |
| CN111178432A (en) | Weak supervision fine-grained image classification method of multi-branch neural network model | |
| CN109583425A (en) | A kind of integrated recognition methods of the remote sensing images ship based on deep learning | |
| CN111191718B (en) | Small sample SAR target identification method based on graph attention network | |
| CN109583483A (en) | A kind of object detection method and system based on convolutional neural networks | |
| CN112862792B (en) | Wheat powdery mildew spore segmentation method for small sample image dataset | |
| CN109285162A (en) | A kind of image, semantic dividing method based on regional area conditional random field models | |
| CN109448015A (en) | Image based on notable figure fusion cooperates with dividing method | |
| CN114863263B (en) | Snakehead fish detection method for blocking in class based on cross-scale hierarchical feature fusion | |
| CN111986126B (en) | Multi-target detection method based on improved VGG16 network | |
| CN113870157A (en) | SAR image synthesis method based on cycleGAN | |
| CN104657980A (en) | Improved multi-channel image partitioning algorithm based on Meanshift | |
| CN110853070A (en) | Underwater sea cucumber image segmentation method based on significance and Grabcut | |
| CN115049952A (en) | Juvenile fish limb identification method based on multi-scale cascade perception deep learning network | |
| CN112733614A (en) | Pest image detection method with similar size enhanced identification | |
| CN108230330B (en) | Method for quickly segmenting highway pavement and positioning camera | |
| CN110633727A (en) | Deep neural network ship target fine-grained identification method based on selective search | |
| CN117079098A (en) | Space small target detection method based on position coding |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190611 |
|
| RJ01 | Rejection of invention patent application after publication |