CN116664845A - Smart construction site image segmentation method and system based on inter-block contrastive attention mechanism - Google Patents

Abstract

The application belongs to the technical field of image segmentation and provides an intelligent building image segmentation method and system based on an inter-block contrast attention mechanism. The method comprises the steps of predicting a target segmentation area of a scene image of a construction site by adopting a segmentation model based on the scene image of the construction site to be segmented; extracting a feature map of a scene image training sample of a construction site; based on the split labels, a plurality of label vectors are obtained; performing block processing and maximum pooling processing on the feature map, and obtaining block class classification labels according to the label vectors; dividing the feature map into a plurality of block-level feature maps; under the supervision of block level classification labels, mapping the block level feature map to obtain a plurality of block level CAMs; establishing a block-level correlation matrix based on a plurality of block-level feature graphs and a plurality of block-level CAMs; and optimizing the super parameters of the segmentation model according to the output result of the segmentation model and the segmentation label. The method can carry out target segmentation processing on the scene image of the construction site, and effectively realize intelligent monitoring of safety of the construction site.

Description

Image Segmentation Method and System for Smart Construction Site Based on Inter-Block Contrastive Attention Mechanism

technical field

The invention belongs to the technical field of image segmentation, and in particular relates to an image segmentation method and system for a smart construction site based on an inter-block contrastive attention mechanism.

Background technique

The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

Semantic segmentation task is an important topic in computer vision. Its main task is to assign a pixel-level category label to each pixel of an image. It plays an important role in the fields of autonomous driving, computer vision, medical image analysis, and computer-aided diagnosis. .

Smart construction site refers to the implementation of scientific management and intelligent production on construction sites through information technology, which involves computer technology, artificial intelligence technology, sensor technology and virtual reality technology. Construction projects usually have problems such as tight schedules, heavy tasks, high risks, and difficult management. At present, the management of construction sites mainly includes inspections and random inspections, which have problems such as poor timeliness and high personnel supervision costs, which lead to an increase in the frequency of illegal operations, resulting in the inability to effectively guarantee the safety, quality, and progress of construction sites.

With the development of artificial intelligence, artificial intelligence technology is gradually applied to the construction site auxiliary supervision system, and the image recognition algorithm technology based on deep learning is used to monitor the construction site monitoring and the pictures taken by the tower crane. At present, the segmentation task of the intelligent construction site scene image Some only consider the global long-distance dependence information, and some only consider the short-distance dependence information, both of which affect the accuracy of the results of the image segmentation task.

Contents of the invention

In order to solve the technical problems in the above-mentioned background technology, the present invention provides a smart construction site image segmentation method and system based on inter-block comparative attention mechanism, which can perform target segmentation processing on construction site scene images, and effectively realize intelligent monitoring of construction site safety , provide production management efficiency, and ensure safe construction on the site.

In order to achieve the above object, the present invention adopts the following technical solutions:

The first aspect of the present invention provides a method for image segmentation of smart construction sites based on an inter-block contrastive attention mechanism.

A smart construction site image segmentation method based on the contrastive attention mechanism between blocks, including:

Based on the construction site scene image to be segmented, the trained segmentation model is used to predict the target segmentation area of the construction site scene image;

The training process of the segmentation model includes: obtaining the construction site scene image training samples marked with segmentation labels; extracting the feature map of the construction site scene image training samples; performing one-hot encoding processing on the segmentation labels to obtain several label vectors; Block processing and maximum pooling processing, according to the label vector, obtain block-level classification labels; divide the feature map into several block-level feature maps; under the supervision of block-level classification labels, map the block-level feature maps to obtain several block-level CAM; based on several block-level feature maps and several block-level CAMs, establish a block-level correlation matrix; map the block-level correlation matrix to a global correlation matrix; calculate the positive sample similarity of the block-level correlation matrix, block-level The negative sample similarity of the correlation matrix, the positive sample similarity of the global correlation matrix, and the negative sample similarity of the global correlation matrix are used to obtain the output feature map. According to the output feature map, the output result of the segmentation model is obtained. According to the segmentation model Output the result and segmentation label, use the loss function, optimize the hyperparameters of the segmentation model, and obtain the trained segmentation model.

Further, the process of establishing a block-level correlation matrix based on several block-level feature maps and several block-level CAMs includes: performing matrix transformation on several block-level feature maps and several block-level CAMs respectively, and performing matrix Multiplied together, a block-level correlation matrix of long dependencies between channels and categories is obtained.

Further, the calculation process of the positive sample similarity of the block-level correlation matrix and the negative sample similarity of the block-level correlation matrix includes: the channels in the block-level correlation matrix whose response value is higher than the set value are positive samples , the rest are negative samples, and the weight matrix is introduced for comparative learning to obtain the positive sample similarity of the block-level correlation matrix and the negative sample similarity of the block-level correlation matrix.

Further, the process of mapping the block-level correlation matrix to a global correlation matrix includes: mapping the block-level correlation matrix to a global correlation matrix through a fully connected layer.

Further, the calculation process of the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix includes: the channel in the global correlation matrix whose response value is higher than the set value is a positive sample, and the rest are Negative samples, introduce the weight matrix, and carry out comparative learning to obtain the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix.

Further, after obtaining the output feature map, it also includes: processing the dimension of the output feature map to make it the same size as the feature map, and obtaining a semantic segmentation mask with the same size as the feature map through upsampling, that is, The output of the segmentation model.

Further, the loss function includes:

Under the supervision of block-level classification labels, the block-level feature maps are mapped to obtain several block-level CAM prediction loss functions;

Carry out the semantic segmentation loss function of the target segmentation area;

and a contrastive loss function between inter-block long dependencies and intra-block long dependencies.

The second aspect of the present invention provides a smart construction site image segmentation system based on inter-block contrastive attention mechanism.

Smart construction site image segmentation system based on inter-block contrastive attention mechanism, including:

A prediction module, which is configured to: predict the target segmentation area of the construction site scene image based on the construction site scene image to be segmented, using a trained segmentation model;

The segmentation model training module is configured to: obtain training samples of construction site scene images marked with segmentation labels; extract feature maps of construction site scene image training samples; perform one-hot encoding on the segmentation labels to obtain several label vectors; Block processing and maximum pooling processing, according to the label vector, obtain block-level classification labels; divide the feature map into several block-level feature maps; under the supervision of block-level classification labels, map the block-level feature maps to obtain several blocks level CAM; based on several block-level feature maps and several block-level CAMs, a block-level correlation matrix is established; the block-level correlation matrix is mapped to a global correlation matrix; the positive sample similarity and block-level correlation matrix are calculated. The negative sample similarity of the level correlation matrix, the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix are obtained to obtain the output feature map. According to the output feature map, the output result of the segmentation model is obtained. According to the segmentation model The output result and the segmentation label, using the loss function, optimize the hyperparameters of the segmentation model, and obtain the trained segmentation model.

A third aspect of the present invention provides a computer readable storage medium.

A computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps in the method for image segmentation of a smart construction site based on an inter-block contrastive attention mechanism as described in the first aspect above are implemented.

A fourth aspect of the present invention provides a computer device.

A computer device, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, when the processor executes the program, it realizes the attention based on the comparison between blocks as described in the first aspect above The steps in the image segmentation method of smart construction site based on force mechanism.

Compared with prior art, the beneficial effect of the present invention is:

For site monitoring and construction site monitoring images transmitted by tower cranes, the present invention performs feature extraction through a neural network, and performs pixel-level classification on the input feature images to obtain the segmented areas of each object in the input image, which is conducive to further detection of violations at the construction site Operational and safety hazards.

The present invention introduces contrastive learning into the semantic segmentation task under supervised setting. Through contrastive learning, the pixels with the same label are closer in the feature space, while the pixels with different labels have relatively large distances in the feature space. distance to further enhance the representation ability of features. Since the attention mechanism and contrastive learning perform well in semantic segmentation tasks, the present invention combines the attention mechanism with contrastive learning, forcing the feature map and the channel correlation matrix of the Class Activation Mapping (CAM) to have Higher confidence, so as to obtain robust and accurate target segmentation results of construction site scene images, and improve the segmentation accuracy of construction site scene images.

Description of drawings

The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention.

Fig. 1 is the flow chart of the intelligent construction site image segmentation method based on the inter-block contrast attention mechanism shown in the present invention;

Fig. 2 is a frame diagram of a smart construction site image segmentation method based on an inter-block contrastive attention mechanism shown in the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It should be noted that the terminology used here is only for describing specific embodiments, and is not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions and operations of possible implementations of methods and systems according to various embodiments of the present disclosure. It should be noted that each block in a flowchart or a block diagram may represent a module, a program segment, or a part of a code, and the module, a program segment, or a part of a code may include one or more An executable instruction for a specified logical function. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block in the flowchart and/or block diagrams, and combinations of blocks in the flowchart and/or block diagrams, can be implemented using a dedicated hardware-based system that performs the specified functions or operations , or can be implemented using a combination of dedicated hardware and computer instructions.

Embodiment one

As shown in Figure 1 and Figure 2, this embodiment provides a smart construction site image segmentation method based on the inter-block contrastive attention mechanism. It can also be applied to a terminal, and can also be applied to a terminal, a server and a system, and is realized through interaction between the terminal and the server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or it can provide cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communications, intermediate Cloud servers for basic cloud computing services such as software services, domain name services, security service CDN, and big data and artificial intelligence platforms. The terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server may be connected directly or indirectly through wired or wireless communication, which is not limited in this application. In this embodiment, the method includes the following steps:

Based on the construction site scene image to be segmented, the trained segmentation model is used to predict the target segmentation area of the construction site scene image;

Obtain the training samples of construction site scene images marked with segmentation labels; extract the feature map of the training site image training samples; perform one-hot encoding on the segmentation labels to obtain several label vectors; perform block processing and maximum pooling processing on the feature maps, according to Label vector to obtain block-level classification labels; divide the feature map into several block-level feature maps; under the supervision of block-level classification labels, map the block-level feature maps to obtain several block-level CAMs; based on several block-level feature maps and several block-level CAMs to establish a block-level correlation matrix; map the block-level correlation matrix to a global correlation matrix; calculate the positive sample similarity of the block-level correlation matrix, the negative sample similarity of the block-level correlation matrix, The positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix are used to obtain the output feature map. According to the output feature map, the output result of the segmentation model is obtained. According to the output result of the segmentation model and the segmentation label, the loss function is used , optimize the hyperparameters of the segmentation model, and obtain the trained segmentation model.

The specific scheme of this embodiment is introduced in detail below:

1. Feature extraction

For an input image, the input image is first mapped to The feature map of , where H and W are the length and width of the feature map, respectively, and C is the number of channels of the feature map.

2. Building block level feature map

First, one-hot encoding (One-Hot Encoding) is performed on the segmentation label to obtain K label vectors , where K is the number of categories in the dataset. Let the feature map be divided into /> blocks, /> , followed by /> The maximum pooling of to obtain block-level classification labels /> . Subsequently, the /> Divided into Np blocks, such as Np=4, that is, the feature map /> Divided into 16 block-level feature maps with length and width h and w respectively, and classify labels at the block level /> supervised by /> Convolution, the block-level /> mapped to Np /> block-level CAM. Final block-level feature map /> has a dimension of /> , the dimension of the block-level CAM is /> .

3. Mining long dependencies between classes at the block level

Will a /> The block-level feature map of /> , becomes N _p /> after reshape hw/> the dimension of C, /> a /> The block-level CAM of is changed to N _p /> after reshape K/> dimension of hw, /> N _p /> is obtained after matrix multiplication with CAM K/> A matrix of C that establishes the correlation of K and C, i.e. /> a /> The correlation matrix /> , /> It reflects the correlation between the i-th categories and the j-th channel. Eventually long dependencies between channels and categories can be established within blocks. By constructing a correlation matrix T to mine the long-term dependencies between the channels in the block and the categories of the dataset, while obtaining the long-term dependency information, the pixel-level segmentation task needs fine-grained short-term dependency information.

4. Automatic acquisition of optimal samples

Maintain a learnable weight matrix A for obtaining positive samples for contrastive learning, the dimension is , the matrix contains K normalized weight vectors /> . Specifically, for a certain category /> , in this embodiment the correlation matrix /> The channel with a higher response value in the medium is used as a positive sample, and the correlation matrix /> The channel with a lower response value in the medium is used as a negative sample, and the positive sample adaptively obtains a higher weight coefficient. At the same time, in order to consider the information of all samples in the block, this embodiment uses the coefficient matrix 1-A as an adaptive weight coefficient for selecting negative samples.

5. The construction of inter-block contrastive attention mechanism

By introducing block-level contrastive learning to force It has a stronger representation ability, that is, a certain channel is more relevant to a certain category, so that it can be used in /> Response value in /> bigger. Through the fully connected layer, the /> block-level correlation matrix/> map to global correlation matrix /> , i.e. /> , where /> Indicates that the input dimension is , output a linear layer with dimension 1, and capture long dependencies between blocks. Select both block-level and global correlation matrices for each category through an automatic acquisition strategy of optimal samples /> and /> Channels with high response values are regarded as positive samples, and channels with low response values are regarded as negative samples. And through the weight matrix A, assign higher weights to the similarity of positive samples and weight the summation as the similarity of positive samples, and assign higher weights to the similarity of negative samples through 1-A and perform weighted summation as the similarity of negative samples. Further calculate the contrast loss, through the contrast loss, the distance between positive samples can be reduced, and the distance between negative samples can be enlarged, that is, for a certain category, the channels related to this category will be more similar under the effect of contrast loss, and Irrelevant passages will be alienated.

Since the final positive and negative samples used for the comparison loss calculation come from each block, the above operations can extend the block-level semantic information to the whole world. Further, based on several block-level feature maps and several block-level CAMs, an attention mechanism is introduced. The traditional self-attention mechanism model often adopts the "Query-Key-Value" model (Query-Key-Value, QKV), that is, the input feature map is linearly transformed by W _q , W _k , and W _v to obtain three parameters of Q, K, and V. The feature map, Q and K obtain the correlation matrix by scaling the dot product, etc., and perform matrix multiplication with V to obtain the output feature map. The present invention constructs attention mechanism through input feature map and CAM, and the process of obtaining output feature map includes: passing block-level CAM through After the linear transformation of V as the attention mechanism, calculate the attention mechanism with several block-level feature maps and several block-level CAMs to construct the output feature map.

That is, first, the channel and category correlation matrices are obtained first. Second, after obtaining the inter-class long-distance dependencies within a block, the global correlation matrix of the global inter-class long-distance dependencies is obtained by aggregating multiple block-level correlation matrices. Finally, the contrastive loss is calculated after the positive and negative samples are sampled, and the loss is passed back to the network, thereby forcing the model to learn a more structured channel and category correlation matrix through contrastive learning. In this way, the intra-block long-term dependency and inter-block long-term dependency are established at the same time, which satisfies the semantic segmentation task's dependence on global semantic information and fine-grained information at the same time.

6. Get the semantic segmentation mask

Obtained by the operation of contrastive attention between blocks Feature map of /> , and reshape it to A feature map of dimension , which has the same dimension as the input feature map. Then through the upsampling operation, the semantic segmentation mask with the same size as the original image is obtained, so as to obtain accurate and robust segmentation results, and calculate the segmentation loss with the segmentation label /> .

7. Calculate the loss and perform gradient return

CAM classification prediction loss guided by class-level labels , semantic segmentation loss /> , and the contrastive loss for inter-block contrastive attention /> Aggregate to get the final loss /> , and carry out gradient return, the present invention defines it as follows:

in, , /> , /> are the weight coefficients of each loss, which are set to 1, 0.4, and 1 respectively in the present invention through a large number of experiments.

8. Model testing and application

Input the test set into the trained segmentation model, output the predicted segmentation results, and evaluate the model performance by the average intersection-over-union ratio (mIoU).

The tested model can be used for construction site image segmentation tasks. For construction site monitoring and images taken by tower cranes, the segmented area of each target in the image can be obtained through the segmentation model.

In terms of local information, this embodiment uses the block attention mechanism to establish the correlation between the channels and categories of the original feature map. Compared with the traditional attention mechanism, the block attention mechanism is more conducive to the network to focus on how to Mining for finer-grained information. At the level of global information, this embodiment proposes inter-block comparative learning. First, channels with higher response values in each channel and category correlation matrix are used as positive samples of the corresponding category, and channels with lower response values are used as negative samples of the corresponding category. sample. More specifically, the intra-block positive and negative samples of each category are extended to global positive and negative samples, which enables the model to acquire fine-grained local semantic information while mining more discriminative global semantic information. It is worth noting that this embodiment maintains a learnable weight matrix in terms of how to select positive and negative samples, which makes the selected positive and negative samples more fitting while ensuring that no information about the positive and negative samples is lost.

In the present invention, the feature map and the CAM are divided into blocks, and the channel correlation of the feature map and the CAM is calculated in the block dimension, so as to mine the dependence of the image category on the channel. In order to make the generated feature map and the embedding space of the channel correlation matrix of CAM have more powerful representation capabilities, the present invention proposes a block-level contrastive attention mechanism, which can not only model the long-distance dependencies of images, but also based on Block-level features establish short-distance dependencies between channels and categories, which can simultaneously guarantee the needs of coarse-grained and fine-grained information for semantic segmentation tasks. Use comparative learning to emphasize the correlation between channels and categories, and fuse the block-level positive and negative samples of each category to obtain the global positive and negative samples of each category, establish the correlation between blocks, and make the segmentation model more efficient Good representation ability and robust segmentation performance.

Embodiment two

This embodiment provides a smart construction site image segmentation system based on an inter-block contrastive attention mechanism.

Smart construction site image segmentation system based on inter-block contrastive attention mechanism, including:

A prediction module, which is configured to: predict the target segmentation area of the construction site scene image based on the construction site scene image to be segmented, using a trained segmentation model;

The segmentation model training module is configured to: obtain training samples of construction site scene images marked with segmentation labels; extract feature maps of construction site scene image training samples; perform one-hot encoding on the segmentation labels to obtain several label vectors; Block processing and maximum pooling processing, according to the label vector, obtain block-level classification labels; divide the feature map into several block-level feature maps; under the supervision of block-level classification labels, map the block-level feature maps to obtain several blocks level CAM; based on several block-level feature maps and several block-level CAMs, a block-level correlation matrix is established; the block-level correlation matrix is mapped to a global correlation matrix; the positive sample similarity and block-level correlation matrix are calculated. The negative sample similarity of the level correlation matrix, the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix are obtained to obtain the output feature map. According to the output feature map, the output result of the segmentation model is obtained. According to the segmentation model The output result and the segmentation label, using the loss function, optimize the hyperparameters of the segmentation model, and obtain the trained segmentation model. It should be noted here that the examples and application scenarios implemented by the above-mentioned prediction module and segmentation model training module are the same as those in the first embodiment, but are not limited to the content disclosed in the first embodiment above. It should be noted that, as a part of the system, the above-mentioned modules can be executed in a computer system such as a set of computer-executable instructions.

Embodiment three

This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the method for image segmentation of a smart construction site based on the contrastive attention mechanism between blocks as described in the first embodiment above is implemented. A step of.

Embodiment four

This embodiment provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the program, the above-mentioned first embodiment based on Steps in a Smart Construction Site Image Segmentation Method with Inter-Block Contrastive Attention Mechanism.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. The intelligent map image segmentation method based on the inter-block contrast attention mechanism is characterized by comprising the following steps of:

based on a to-be-segmented construction site scene image, predicting a target segmentation area of the construction site scene image by adopting a trained segmentation model;

the training process of the segmentation model comprises the following steps: acquiring a construction site scene image training sample marked with a segmentation label; extracting a feature map of a scene image training sample of a construction site; performing single-heat encoding treatment on the split labels to obtain a plurality of label vectors; performing block processing and maximum pooling processing on the feature map, and obtaining block class classification labels according to the label vectors; dividing the feature map into a plurality of block-level feature maps; under the supervision of block level classification labels, mapping the block level feature map to obtain a plurality of block level CAMs; establishing a block-level correlation matrix based on a plurality of block-level feature graphs and a plurality of block-level CAMs; mapping the block-level correlation matrix into a global correlation matrix; calculating positive sample similarity of the block-level correlation matrix, negative sample similarity of the block-level correlation matrix, positive sample similarity of the global correlation matrix and negative sample similarity of the global correlation matrix, obtaining an output feature map, obtaining an output result of the segmentation model according to the output feature map, optimizing hyper-parameters of the segmentation model by adopting a loss function according to the output result of the segmentation model and the segmentation label, and obtaining a trained segmentation model.

2. The intelligent job site image segmentation method based on inter-block contrast attention mechanism according to claim 1, wherein the process of establishing a block-level correlation matrix based on a number of block-level feature maps and a number of block-level CAMs comprises: and respectively carrying out matrix transformation on the plurality of block-level feature maps and the plurality of block-level CAM, and then carrying out matrix multiplication to obtain a block-level correlation matrix of long dependency relationship between the channel and the category.

3. The intelligent work image segmentation method based on inter-block contrast attention mechanism of claim 1, wherein the calculation process of the positive sample similarity of the block-level correlation matrix and the negative sample similarity of the block-level correlation matrix comprises: and taking the channel with the response value higher than the set value in the block-level correlation matrix as a positive sample, taking the rest as a negative sample, introducing a weight matrix, and performing contrast learning to obtain the positive sample similarity of the block-level correlation matrix and the negative sample similarity of the block-level correlation matrix.

4. The intelligent job image segmentation method based on inter-block contrast attention mechanism according to claim 1, wherein the process of mapping the block-level correlation matrix into a global correlation matrix comprises: the block-level correlation matrix is mapped to a global correlation matrix through the full connection layer.

5. The intelligent job image segmentation method based on inter-block contrast attention mechanism according to claim 1, wherein the calculation process of the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix comprises: and taking the channel with the response value higher than the set value in the global correlation matrix as a positive sample, taking the rest as a negative sample, introducing a weight matrix, and performing contrast learning to obtain the positive sample similarity of the global correlation matrix and the negative sample similarity of the global correlation matrix.

6. The intelligent job image segmentation method based on inter-block contrast attention mechanism according to claim 1, further comprising, after obtaining the output feature map: and processing the dimension of the output feature map to be the same as the dimension of the feature map, and obtaining a semantic segmentation mask with the same size as the feature map through up-sampling, namely, obtaining an output result of the segmentation model.

7. The intelligent job image segmentation method based on inter-block contrast attention mechanism according to claim 1, wherein the loss function comprises:

under the supervision of block level classification labels, mapping the block level feature map to obtain a plurality of prediction loss functions of the block level CAM;

performing a semantic segmentation loss function of the target segmentation region;

and a contrast loss function between inter-block long dependencies and intra-block long dependencies.

8. An intelligent building image segmentation system based on inter-block contrast attention mechanism, comprising:

a prediction module configured to: based on a to-be-segmented construction site scene image, predicting a target segmentation area of the construction site scene image by adopting a trained segmentation model;

a segmentation model training module configured to: acquiring a construction site scene image training sample marked with a segmentation label; extracting a feature map of a scene image training sample of a construction site; performing single-heat encoding treatment on the split labels to obtain a plurality of label vectors; performing block processing and maximum pooling processing on the feature map, and obtaining block class classification labels according to the label vectors; dividing the feature map into a plurality of block-level feature maps; under the supervision of block level classification labels, mapping the block level feature map to obtain a plurality of block level CAMs; establishing a block-level correlation matrix based on a plurality of block-level feature graphs and a plurality of block-level CAMs; mapping the block-level correlation matrix into a global correlation matrix; calculating positive sample similarity of the block-level correlation matrix, negative sample similarity of the block-level correlation matrix, positive sample similarity of the global correlation matrix and negative sample similarity of the global correlation matrix, obtaining an output feature map, obtaining an output result of the segmentation model according to the output feature map, optimizing hyper-parameters of the segmentation model by adopting a loss function according to the output result of the segmentation model and the segmentation label, and obtaining a trained segmentation model.

9. A computer readable storage medium, on which a computer program is stored, which program, when being executed by a processor, implements the steps of the intelligent worker image segmentation method based on inter-block contrast attention mechanism as claimed in any of claims 1-7.

10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps in the intelligent inter-block contrast attention based method of any of claims 1-7 when the program is executed.