A method and system for segmenting electric power equipment based on fusion modal contrast learning

Abstract

The present invention belongs to the field of image segmentation technology, and provides a method and system for segmenting electric equipment based on fused modal contrast learning. The method includes obtaining an infrared image and a visible light image pair of the electric equipment to be segmented; obtaining the electric equipment segmentation result based on the obtained image pair and the trained electric equipment segmentation network; during the training process, the electric equipment segmentation network is composed of a fused modal feature extractor, a semantic segmentation decoder and a contrast learning module; after the training is completed, the electric equipment segmentation network is composed of a fused modal feature extractor and a semantic segmentation decoder; the fused modal feature extractor is used to extract the fused modal features of the infrared image and the visible light image pair, and the semantic segmentation decoder is used to decode the fused modal features; during the training process, the semantic segmentation decoder and the contrast learning module are used to calculate the fused modal segmentation loss and the fused modal contrast loss respectively, and then obtain the fused modal total loss to train the electric equipment segmentation network.

Description

Technical Field

[0001] The present invention belongs to the technical field of image segmentation, and in particular relates to a method and system for segmenting electric power equipment based on fusion modal contrast learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Power equipment segmentation collects images and segments the equipment to predict faults in advance. The goal of semantic segmentation is to assign a corresponding category label to each pixel in the image, so as to identify and segment different objects, thereby enabling the segmentation of different power equipment in the power grid. With the development of sensors, multimodal semantic segmentation tasks have attracted more and more research interests due to their high adaptability to the environment, but have not yet been applied to infrared and visible light image power equipment segmentation. The inventors found that existing power equipment segmentation methods mostly rely on traditional methods, which are inefficient and have inaccurate segmentation results, or only use infrared images for segmentation, which is prone to false positives or false negatives. Summary of the invention

[0004] In order to solve at least one technical problem existing in the above background technology, the present invention provides a method and system for segmenting power equipment based on fused modal contrast learning, which uses a multi-level fused modal feature encoder to extract and fuse visible light images and infrared images to obtain high-dimensional features of power equipment with good characterization capabilities; and uses a contrast learning module to mine contextual semantic information across images to better distinguish different categories of power equipment. The combined effect of the two improves the reliability and accuracy of power equipment segmentation, providing technical support for power equipment fault analysis.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A first aspect of the present invention provides a method for segmenting electric power equipment based on fusion modal contrast learning, comprising:

[0007] Acquire an infrared image and a visible light image pair of the electric power equipment to be segmented;

[0008] Based on the acquired image pairs and the trained power equipment segmentation network, the power equipment segmentation result is obtained;

[0009] Among them, during the training process, the power equipment segmentation network is composed of a fusion modal feature extractor, a semantic segmentation decoder and a contrastive learning module; after the training is completed, the power equipment segmentation network is composed of a fusion modal feature extractor and a semantic segmentation decoder; the fusion modal feature extractor is used to extract the fusion modal features of the infrared image and the visible light image pair, and the semantic segmentation decoder is used to decode the fusion modal features; during the training process, the semantic segmentation decoder and the contrastive learning module are used to calculate the fusion modal segmentation loss and the fusion modal contrast loss respectively, and then obtain the fusion modal total loss to train the power equipment segmentation network.

[0010] As an implementation mode, the fused modal feature encoder includes an infrared feature encoder, a visible light feature encoder and multiple fusion modules. The fused modal feature encoder is provided with several levels, each level module is a fusion module, and the fusion module is used to fuse the features encoded by the infrared feature encoder and the visible light feature encoder of the corresponding level.

[0011] As an implementation manner, the infrared feature encoder has the same structure as the visible light feature encoder.

[0012] As an implementation method, the contrastive learning module includes a projection module, a feature storage unit and a calculation module; the projection module is used to project the fused modal features output by the fused feature encoder into multiple fused modal pixel features; the feature storage unit is used to store the fused modal center pixel features and update them in each batch; the calculation module is used to sample and output an anchor sample set from the fused modal pixel features of the current batch, sample a positive and negative difficult sample set from the fused modal center pixel features stored in the feature storage unit, and calculate the fused modal contrast loss by comparing the negative difficult sample set and the anchor sample set.

[0013] As an implementation method, based on the fused modality pixel features, the fused modality center pixel features of each category in the current batch are calculated, and the feature storage unit is updated.

[0014] As an implementation method, for each infrared and visible light image pair in the current batch, the fused modality center pixel features of the infrared and visible light image pair in each category are obtained by pooling multiple fused modality pixel features belonging to the same category corresponding to the infrared and visible light image pair.

[0015] As an implementation mode, the computing module uses a semantically-aware anchor sample sampling method to select anchor samples for multiple fused modality pixel features of the current batch to obtain an anchor sample set of the current batch.

[0016] A second aspect of the present invention provides a power equipment segmentation system based on fusion modal contrast learning, comprising:

[0017] An image pair acquisition module, which is used to acquire an infrared image and a visible light image pair of the electric power equipment to be segmented;

[0018] A power equipment segmentation module, which is used to obtain a power equipment segmentation result based on the acquired image pairs and the trained power equipment segmentation network;

[0019] Among them, during the training process, the power equipment segmentation network is composed of a fusion modal feature extractor, a semantic segmentation decoder and a contrastive learning module; after the training is completed, the power equipment segmentation network is composed of a fusion modal feature extractor and a semantic segmentation decoder; the fusion modal feature extractor is used to extract the fusion modal features of the infrared image and the visible light image pair, and the semantic segmentation decoder is used to decode the fusion modal features; during the training process, the semantic segmentation decoder and the contrastive learning module are used to calculate the fusion modal segmentation loss and the fusion modal contrast loss respectively, and then obtain the fusion modal total loss to train the power equipment segmentation network.

[0020] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for segmenting electric power equipment based on fusion modal contrast learning as described above.

[0021] The fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for segmenting power equipment based on fusion modal contrast learning as described above are implemented.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) The fused modal feature encoder of the present invention comprises an infrared feature encoder, a visible light feature encoder and a plurality of fusion modules. The fused modal feature encoder is provided with a plurality of levels, each level module is a fusion module, and the fusion module is used to fuse the features encoded by the infrared feature encoder and the visible light feature encoder of the corresponding level. Through the multi-level design of the fused modal feature encoder, better extraction and fusion of infrared and visible light modal features are achieved, thereby utilizing the different information reflected by the infrared and visible light images to obtain more reliable and accurate segmentation results.

[0024] (2) During the training process, a contrastive learning module is provided for the electric power equipment segmentation network of the present invention. The semantic segmentation decoder and the contrastive learning module are used to respectively calculate the fused modal segmentation loss and the fused modal contrast loss, and then the fused modal total loss is obtained to train the electric power equipment segmentation network. This improves the intra-class compactness and inter-class diversity of the fused modal pixel features in the feature embedding space, and further improves the segmentation accuracy of the electric power equipment.

[0025] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 is a schematic diagram of a power equipment segmentation network based on fusion modal contrast learning according to an embodiment of the present invention;

[0028] Figure 2 4 is a flow chart of a comparison module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0031] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0032] Embodiment 1

[0033] This embodiment provides a method for segmenting electric power equipment based on fusion modal contrast learning, which specifically includes the following steps:

[0034] S101: Acquire an infrared image and a visible light image pair of the electric power equipment to be segmented.

[0035] S102: Obtaining a power equipment segmentation result based on the acquired image pairs and the trained power equipment segmentation network.

[0036] Among them, Figure 1 As shown, the power equipment segmentation network consists of a fusion modal feature extractor, a semantic segmentation decoder and a contrastive learning module. The semantic segmentation decoder and the contrastive learning module are both connected to the fusion modal feature extractor, and are used to calculate the fusion modal segmentation loss and the fusion modal contrast loss, respectively, to obtain the fusion modal total loss, and the power equipment segmentation network is trained based on the fusion modal total loss to obtain the trained power equipment segmentation network.

[0037] In the specific implementation, the SGD algorithm with momentum is used for optimization training. The specific settings of the optimizer are: the initial learning rate is set to 0.0025, the momentum is set to 0.9, and the weight decay coefficient is set to 0.0005. In addition, the poly learning rate strategy is adopted to automatically adjust the learning rate size as the number of iterations increases, and the batch size is set to 8.

[0038] Before training the power equipment segmentation network, it also includes: collecting multiple infrared images and corresponding visible light images that need to be segmented for power equipment, and the required training set is composed of multiple infrared images and corresponding visible light images and the category labels of the power equipment in each infrared image; after the required training set is processed in batches, the infrared images and their corresponding visible light images are input into the power equipment segmentation network in batches to train the power equipment segmentation network.

[0039] In the specific implementation process, the fused modal feature encoder includes an infrared feature encoder, a visible light feature encoder and multiple fusion modules. The fused modal feature encoder is provided with several levels, each level module is a fusion module, and the fusion module is used to fuse the features encoded by the infrared feature encoder and the visible light feature encoder of the corresponding level.

[0040] Wherein, the infrared feature encoder has the same structure as the visible light feature encoder.

[0041] exist Figure 1In the figure, the infrared feature encoder and the visible light feature encoder both contain K levels, and the number of fusion modules is consistent with the number of levels; the kth (k=1, 2, ...K-1)th level of the infrared feature encoder outputs the kth level of infrared original coding features, and the kth level of the visible light feature encoder outputs the kth level of visible light original coding features. Both the kth level of infrared original coding features and the visible light original coding features are input into the kth fusion module; the kth fusion module outputs the kth level of infrared-visible light fusion features, which are added to the kth level of infrared original coding features and then input into the k+1th level of the infrared feature encoder, and added to the kth level of visible light original coding features and then input into the k+1th level of the visible light feature encoder; the kth fusion module outputs the fused modal features and inputs them into the semantic segmentation decoder and contrastive learning module.

[0042] In the specific implementation process, the contrastive learning module includes a projection module, a feature storage unit and a calculation module; the projection module is used to project the fused modal features output by the fused feature encoder into multiple fused modal pixel features; the feature storage unit is used to store the fused modal center pixel features and update them in each batch; the calculation module is used to sample and output the anchor sample set from the fused modal pixel features of the current batch, sample the positive and negative difficult sample sets from the fused modal center pixel features stored in the feature storage unit, and calculate the fused modal contrast loss by comparing the negative difficult sample set and the anchor sample set.

[0043] The specific structure of the projection module is: it is composed of a first convolution layer with a convolution kernel size of 1×1, a batch normalization layer, a RELU activation function layer, and a second convolution layer with a convolution kernel size of 1×1. The projection module converts the input fusion modal features into Dimensionality reduction to Where H and W represent the height and width of the feature, respectively, and D and D′ represent the dimensions of the feature before and after dimensionality reduction, respectively. i =M′ (i) It represents the fused modality pixel feature corresponding to the pixel position i. In the specific implementation, the feature dimension after dimensionality reduction is D′=256.

[0044] The feature storage unit is specifically implemented by C queues, each of which has a size of Q, and is used to store the central pixel features of the fused modalities not limited to the current batch and is updated in each batch, where C represents the number of categories of power equipment. In a specific implementation, Q=1000.

[0045] The calculation module samples the anchor sample set from the current batch of fused modality pixel features output by the projection module and outputs it, samples the positive and negative difficult sample sets from the fused modality center pixel features stored in the feature storage unit, and calculates the fused modality contrast loss, such as Figure 2 As shown, specifically:

[0046] Step a: The projection module projects the fused modality features output by the fusion feature encoder into multiple fused modality pixel features.

[0047] Step b: Based on the fused modality pixel features, calculate the fused modality center pixel features of each category in the current batch and update the feature storage unit.

[0048] Among them, the specific process of step b is:

[0049] Step b1: For each infrared and visible light image pair of the current batch, multiple fusion modality pixel features belonging to the same category corresponding to the infrared and visible light image pair are pooled to obtain the fusion modality center pixel features of each category of the infrared and visible light image pair:

[0050] Step b2: Repeat step b1, traverse the remaining infrared and visible light image pairs in the current batch, and obtain the fusion modality center pixel features of each category in the current batch;

[0051] Step b3: For each fused modality center pixel feature obtained based on step b2, according to the category of the fused modality center pixel feature, insert it into the back end of the queue of the same category in the feature storage unit. If the queue size has exceeded the limit capacity, then delete a fused modality center pixel feature from the front end of the queue.

[0052] Step c: Use the semantic-aware anchor sample sampling method to select anchor samples for multiple fusion modality pixel features of the current batch to obtain the anchor sample set of the current batch.

[0053] Among them, the specific process of step c is:

[0054] Step c1: Randomly sample pixel positions of multiple fused modality pixel features of the current batch, and use the sampled pixel positions as half of the anchor points; sample pixel positions that are misclassified in the semantic segmentation decoder, and use the sampled pixel positions as the other half of the anchor points;

[0055] Step c2: extract pixel features of all anchor points in step c1 from multiple fused modality pixel features as anchor point samples, and the anchor point samples of the current batch are composed of the anchor point samples.

[0056] Step d: According to the anchor sample set of the current batch, for each anchor sample, use the positive and negative sample sampling method of difficult sample mining to select positive and negative samples of the fused modality center pixel features stored in the feature storage unit, obtain the positive and negative difficult sample set corresponding to the anchor sample, and calculate the fused modality contrast loss of the anchor sample.

[0057] The specific process of step d is as follows:

[0058] Step d1: Among the fusion modality center pixel features stored in the feature storage unit, the pixel features that belong to the same category as the current anchor point sample are used as the positive sample set of the current anchor point sample, and the pixel features that do not belong to the same category as the current anchor point sample are used as the negative sample set of the current anchor point sample;

[0059] Step d2: Calculate the similarity between the current anchor sample and each positive sample in the positive sample set of the current anchor sample and sort the similarities in ascending order, take the positive samples with the top 10% similarity as the difficult sample set of the current positive sample, and then randomly sample K1 samples from the difficult sample set of the current positive sample set as the positive difficult sample set of the current anchor sample, where K1=1024 in the specific implementation;

[0060] Step d3: Calculate the similarity between the current anchor sample and each negative sample in the negative sample set of the current anchor sample and sort the similarities in descending order, take the negative samples with the top 10% similarity as the difficult sample set of the current negative sample, and then randomly sample K2 samples from the difficult sample set of the current negative sample set as the negative difficult sample set of the current anchor sample, where K2=1024 in the specific implementation;

[0061] Step d4: Based on the positive difficult sample set and negative difficult sample set obtained in steps d1 to d3, the fusion modality contrast loss of the current anchor point sample is calculated by the following formula:

[0062]

[0063] Among them, m i Represents the current anchor point sample, that is, the fusion modality pixel feature corresponding to the pixel position i, and uses and Represents the anchor point samples m respectively i Positive and negative samples of hard samples, and They represent the positive and negative difficult sample sets where pixel i is located, τ represents the temperature coefficient, It represents the size of the positive difficult sample set where the pixel i is located. In the specific implementation, the temperature coefficient τ=0.1.

[0064] The multimodal total loss is calculated by the following formula:

[0065]

[0066] in, represents the total loss of the fused modality for a single batch, represents the fusion modality semantic segmentation loss, λ represents the loss balance weight, and P represents the number of anchor samples in a single batch of anchor sample sets. The fusion modality semantic segmentation loss adopts pixel-by-pixel cross entropy loss. In a specific implementation, for example, λ=1, P=500.

[0067] Step e: Repeat step d, traverse the remaining anchor samples in the current anchor sample set, obtain the positive and negative difficult sample sets of each anchor sample, and calculate the fused modality contrast loss of each anchor sample, and take the sum of the fused modality contrast loss of each anchor sample as the fused modality contrast loss of the batch.

[0068] Embodiment 2

[0069] This embodiment provides a power equipment segmentation system based on fusion modal contrast learning, which specifically includes the following modules:

[0070] An image pair acquisition module, which is used to acquire an infrared image and a visible light image pair of the electric power equipment to be segmented;

[0071] A power equipment segmentation module, which is used to obtain a power equipment segmentation result based on the acquired image pairs and the trained power equipment segmentation network;

[0072] Among them, during the training process, the power equipment segmentation network is composed of a fusion modal feature extractor, a semantic segmentation decoder and a contrastive learning module; after the training is completed, the power equipment segmentation network is composed of a fusion modal feature extractor and a semantic segmentation decoder; the fusion modal feature extractor is used to extract the fusion modal features of the infrared image-visible light image pair, and the semantic segmentation decoder is used to decode the fusion modal features; during the training process, the semantic segmentation decoder and the contrastive learning module are used to calculate the fusion modal segmentation loss and the fusion modal contrast loss respectively, and then obtain the fusion modal total loss to train the power equipment segmentation network.

[0073] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0074] Embodiment 3

[0075] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the method for segmenting electric power equipment based on fusion modal contrast learning as described above are implemented.

[0076] Embodiment 4

[0077] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for segmenting electric power equipment based on fusion modal contrast learning as described above are implemented.

[0078] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of the processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for segmenting power equipment based on fusion modal contrast learning, characterized in that: include: Acquire an infrared image and a visible light image pair of the electric power equipment to be segmented; Based on the acquired image pairs and the trained power equipment segmentation network, the power equipment segmentation result is obtained; Among them, during the training process, the power equipment segmentation network consists of a fusion modal feature encoder, a semantic segmentation decoder and a contrastive learning module; after the training is completed, the power equipment segmentation network consists of a fusion modal feature encoder and a semantic segmentation decoder; the fusion modal feature encoder is used to extract the fusion modal features of the infrared image and the visible light image pair, and the semantic segmentation decoder is used to decode the fusion modal features; During the training process, the semantic segmentation decoder and contrastive learning module are used to calculate the fusion modality segmentation loss and the fusion modality contrast loss respectively, and then obtain the fusion modality total loss to train the power equipment segmentation network; The contrastive learning module includes a projection module, a feature storage unit and a calculation module; the projection module is used to project the fused modality features output by the fused feature encoder into multiple fused modality pixel features; the feature storage unit is used to store the fused modality center pixel features and update them in each batch; the calculation module is used to sample and output the anchor sample set from the fused modality pixel features of the current batch, sample the positive and negative difficult sample sets from the fused modality center pixel features stored in the feature storage unit, and calculate the fused modality contrast loss by comparing the negative difficult sample set and the anchor sample set; Based on the fused modality pixel features, calculate the fused modality center pixel features of each category in the current batch and update the feature storage unit; The semantically-aware anchor sample sampling method is used to select anchor samples for multiple fused modal pixel features of the current batch to obtain the anchor sample set of the current batch; based on the anchor sample set of the current batch, for each anchor sample, the positive and negative sample sampling method of difficult sample mining is used to select positive and negative samples for the fused modal central pixel features stored in the feature storage unit to obtain the positive and negative difficult sample sets corresponding to the anchor sample, and the fused modality contrast loss of the anchor sample is calculated.

2. The power equipment segmentation method based on fusion modal contrast learning according to claim 1 is characterized in that: The fused modal feature encoder includes an infrared feature encoder, a visible light feature encoder and multiple fusion modules. The fused modal feature encoder is provided with several levels, each level module is a fusion module, and the fusion module is used to fuse the features encoded by the infrared feature encoder and the visible light feature encoder of the corresponding level.

3. The power equipment segmentation method based on fusion modal contrast learning according to claim 2 is characterized in that: The infrared feature encoder has the same structure as the visible light feature encoder.

4. The power equipment segmentation method based on fusion modal contrast learning according to claim 1 is characterized in that: For each infrared and visible light image pair in the current batch, the fusion modality center pixel features of each category of the infrared and visible light image pair are obtained by pooling multiple fusion modality pixel features belonging to the same category corresponding to the infrared and visible light image pair.

5. The power equipment segmentation method based on fusion modal contrast learning according to claim 1, characterized in that: The calculation module uses a semantically-aware anchor sample sampling method to select anchor samples for multiple fusion modality pixel features of the current batch to obtain an anchor sample set of the current batch.

6. A power equipment segmentation system based on fusion modal contrast learning, characterized in that: include: An image pair acquisition module, which is used to acquire an infrared image and a visible light image pair of the electric power equipment to be segmented; A power equipment segmentation module, which is used to obtain a power equipment segmentation result based on the acquired image pairs and the trained power equipment segmentation network; Among them, during the training process, the power equipment segmentation network consists of a fusion modal feature encoder, a semantic segmentation decoder and a contrastive learning module; after the training is completed, the power equipment segmentation network consists of a fusion modal feature encoder and a semantic segmentation decoder; the fusion modal feature encoder is used to extract the fusion modal features of the infrared image and the visible light image pair, and the semantic segmentation decoder is used to decode the fusion modal features; During the training process, the semantic segmentation decoder and contrastive learning module are used to calculate the fusion modality segmentation loss and the fusion modality contrast loss respectively, and then obtain the fusion modality total loss to train the power equipment segmentation network; The contrastive learning module includes a projection module, a feature storage unit and a calculation module; the projection module is used to project the fused modality features output by the fused feature encoder into multiple fused modality pixel features; the feature storage unit is used to store the fused modality center pixel features and update them in each batch; the calculation module is used to sample and output the anchor sample set from the fused modality pixel features of the current batch, sample the positive and negative difficult sample sets from the fused modality center pixel features stored in the feature storage unit, and calculate the fused modality contrast loss by comparing the negative difficult sample set and the anchor sample set; Based on the fused modality pixel features, calculate the fused modality center pixel features of each category in the current batch and update the feature storage unit; The semantically-aware anchor sample sampling method is used to select anchor samples for multiple fused modal pixel features of the current batch to obtain the anchor sample set of the current batch; based on the anchor sample set of the current batch, for each anchor sample, the positive and negative sample sampling method of difficult sample mining is used to select positive and negative samples for the fused modal central pixel features stored in the feature storage unit to obtain the positive and negative difficult sample sets corresponding to the anchor sample, and the fused modality contrast loss of the anchor sample is calculated.

7. The power equipment segmentation system based on fusion modal contrast learning according to claim 6, characterized in that: The fused modal feature encoder includes an infrared feature encoder, a visible light feature encoder and multiple fusion modules. The fused modal feature encoder is provided with several levels, each level module is a fusion module, and the fusion module is used to fuse the features encoded by the infrared feature encoder and the visible light feature encoder of the corresponding level.

8. The power equipment segmentation system based on fusion modal contrast learning according to claim 7, characterized in that: The infrared feature encoder has the same structure as the visible light feature encoder.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the power equipment segmentation method based on fusion modal contrast learning as described in any one of claims 1 to 5 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the power equipment segmentation method based on fusion modal contrast learning as described in any one of claims 1 to 5 are implemented.

Images