Line insulator image processing method in low-temperature windy and snowy environment

Abstract

The invention provides a line insulator image processing method in a low-temperature windy and snowy environment, which relates to the technical field of graphic image processing, and comprises the following steps: acquiring an insulator image set in original windy and snowy weather, performing snow removal operation on all pictures in the image set by using several selected snow removal models, comparing the snow removal effects of the several models on different pictures, and calculating the snow removal results of the models. Classifying the pictures to a model with the best snow removal effect, and taking the obtained at least two types of pictures as a data set to train a fine-grained classification network DBTNet; and using the trained DBTNet network to carry out feature classification processing on the newly obtained insulator image under the wind and snow to obtain at least two types of to-be-processed image subsets, and respectively inputting the at least two types of to-be-processed image subsets into the corresponding target snow removal models to carry out snow removal processing.

Description

Technical Field

[0001] The present invention relates to the technical field of graphic image processing, and in particular to a method for processing line insulator images in a low-temperature, windy and snowy environment. Background Art

[0002] Insulator strings are important components in high-voltage transmission lines, playing an important role in electrical insulation and mechanical support. Insulator strings are often installed outdoors and are susceptible to natural climate (wind, rain, snow, lightning, etc.) and wild animals. Therefore, they are prone to various defects such as cracks and pollution. The damage and pollution of insulator strings pose a huge safety hazard to the normal operation of the power system. Therefore, the use of drones to monitor insulator strings in real time is an important research topic with practical significance for the safe operation of power systems. On this basis, it is necessary to perform snow removal on the insulator images taken by drones in windy and snowy weather so that the working conditions of the insulator strings can be judged more accurately and intuitively.

[0003] However, the current snow removal models often have good snow removal effects for one scene but poor effects for another scene. The image background, light, distance to the insulator, shape and density of snowflakes may have different effects on different snow removal models. How to efficiently and accurately complete the snow removal operation to obtain a clearer insulator image has become a problem that needs to be solved urgently. Summary of the invention

[0004] In order to solve the technical problems that the snow removal model in the prior art is not generalizable and the snow removal effect is poor, the present invention provides an image processing method for line insulators in a low-temperature windy and snowy environment.

[0005] The technical solution of the embodiment of the present application is implemented as follows: A first aspect of an embodiment of the present application provides a method for processing an image of a line insulator in a low-temperature snowy environment, comprising: Obtain the snow removal feature type corresponding to the image set to be processed and the target snow removal model; Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed; The at least two types of image subsets to be processed are respectively input into target snow removal models matching the snow removal feature types for snow removal processing.

[0006] Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed; The at least two types of image subsets to be processed are respectively input into corresponding target snow removal models for snow removal processing.

[0007] Optionally, the target snow removal model includes: a DesnowNet model, a HDCWNet model and a SMGARN model; and the step of obtaining the snow removal feature type corresponding to the image set to be processed and the target snow removal model also includes: According to the snow removal principles of the DesnowNet model, the HDCWNet model and the SMGARN model, the snow removal feature types corresponding to each model are determined.

[0008] Optionally, the image set to be processed is input into a DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, including: Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain a first category of image subsets to be processed, a second category of image subsets to be processed, and a third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second type of image subset to be processed is adapted to the snow removal feature type of the HDCWNet model; The third type of image subset to be processed is adapted to the snow removal feature type of the SMGARN model.

[0009] Optionally, the step of inputting the at least two types of image subsets to be processed into corresponding target snow removal models for snow removal processing includes: Inputting the first type of image subset to be processed into the DesnowNet model for snow removal; Inputting the second type of image subset to be processed into the HDCWNet model for snow removal; The third category of image subsets to be processed is input into the SMGARN model for snow removal.

[0010] Optionally, before inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, the method further includes: The initial DBTNet network is trained using sample images and at least two different types of snow removal features to obtain the trained DBTNet network.

[0011] A second aspect of the present application provides a line insulator image processing device in a low-temperature snowy environment, comprising: an acquisition module, a classification module and a processing module, wherein: The acquisition module is configured to acquire the snow removal feature type corresponding to the image set to be processed and the target snow removal model; The classification module is configured to input the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed; The processing module is configured to input the at least two types of image subsets to be processed into target snow removal models matching the snow removal feature types for snow removal processing.

[0012] Optionally, the target snow removal model includes: a DesnowNet model, a HDCWNet model and a SMGARN model; the acquisition module is specifically configured as follows: According to the snow removal principles of the DesnowNet model, the HDCWNet model and the SMGARN model, the snow removal feature types corresponding to each model are determined.

[0013] Optionally, the classification module is specifically configured as follows: Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain a first category of image subsets to be processed, a second category of image subsets to be processed, and a third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second type of image subset to be processed is adapted to the snow removal feature type of the HDCWNet model; The third type of image subset to be processed is adapted to the snow removal feature type of the SMGARN model.

[0014] A third aspect of an embodiment of the present application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the line insulator image processing method in a snowy environment described in the first aspect.

[0015] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0016] Compared with the prior art, the technical solution provided by this application has the following beneficial effects: The present invention provides an image processing method for line insulators in a low-temperature windy and snowy environment. The method inputs a set of images to be processed into a DBTNet network for feature classification processing, and inputs the classified subsets of images to be processed into corresponding multiple target snow removal models for snow removal processing, so that each frame of the input image to be processed can be matched with the snow removal model that is most suitable for it, thereby giving full play to the advantages of each snow removal model and improving the snow removal efficiency and quality of the image. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic flow chart of a method for processing line insulator images in a low-temperature, snowy environment provided by an embodiment of the present application; Figure 2 A schematic diagram of the structure of a line insulator image processing device in a low-temperature snowy environment provided by an embodiment of the present application; Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application. The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components. All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0019] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0020] In some embodiments, see Figure 1 , Figure 1 A schematic flow chart of a method for processing line insulator images in a low-temperature, snowy and windy environment provided in an embodiment of the present application. A method for processing line insulator images in a low-temperature, snowy and windy environment provided in an embodiment of the present application includes: S110, obtaining snow removal feature types corresponding to the image set to be processed and the target snow removal model.

[0021] Here, a large number of insulator images in windy and snowy environments can be obtained by taking drone photos on the spot to form a set of images to be processed. The set of images to be processed contains a large number of photographed images. In addition to the main body of the insulator, the photographed images can also contain various possible factors in reality, such as different snowing intensities, different wind intensities, different shooting backgrounds, different shooting distances, different light conditions, etc. The snow removal feature type is related to the image processing focus of the target snow removal model. The target snow removal model can be two or more of the current mainstream snow removal models. These snow removal models can correspond to the same or different snow removal feature types, such as some snow removal models focus on removing snowflake noise in the image, some snow removal models focus on processing distortion in the image, and some snow removal models focus on image generation, enhancement or restoration tasks.

[0022] S120, inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed.

[0023] The principle of DBTNet (Dual-Bilinear Transform Network, fine-grained classification network) is mainly based on the idea of ​​bilinear transform, which combines the feature representation ability of CNN (convolutional neural network) to improve the classification performance of the model. Bilinear transform is a nonlinear feature fusion technology that combines information from different feature channels and generates more discriminative feature representations by calculating pairwise interactions between them. In DBTNet, this bilinear transform is designed as a module that can be embedded in CNN, namely DBT block. Specifically, the DBT block divides the channels of the input feature map into multiple semantic groups. These semantic groups are divided based on the semantic information of the input features (such as different object parts or attributes), so that the channels in the same group have similar semantic meanings. Within each semantic group, the DBT block performs bilinear transform and calculates pairwise interactions between channels. This interaction helps capture the complex relationship between features, thereby enhancing the classification ability of the model. Through the combination of semantic grouping and bilinear transform, DBTNet can improve the classification accuracy of the model while maintaining computational efficiency. In addition, the DBT block can be flexibly embedded into the existing CNN architecture and work in collaboration with other network components (such as convolutional layers, pooling layers, etc.) to build a more powerful deep learning model.

[0024] In this embodiment, the image set to be processed is classified based on the snow removal feature type corresponding to the target snow removal model to obtain at least two types of image subsets to be processed. According to the accuracy requirements, the image set to be processed can be divided into two, three or even more types. It should be noted that the classification features of the at least two types of image subsets to be processed are not obvious, and it is difficult to complete the classification task using a conventional image classification model. Therefore, the DBTNet model in the field of fine-grained image classification is selected here to improve the classification accuracy. The DBTNet network can more easily capture subtle differences in the image, and realize insulator recognition under different light intensities, different shooting angles and occlusion conditions, thereby ensuring the efficiency and accuracy of feature classification. It should be noted that the snow removal feature types mainly include inconspicuous features hidden in the image set to be processed.

[0025] S130, inputting at least two types of image subsets to be processed into target snow removal models matching the snow removal feature types for snow removal processing.

[0026] In this embodiment, two types of image subsets to be processed are used as examples for explanation. The target snow removal model includes a first snow removal model and a second snow removal model. Correspondingly, the snow removal feature type includes a first snow removal feature and a second snow removal feature. The DBTNet network is used to classify the image set to be processed using the first snow removal feature and the second snow removal feature as the classification basis, and the first image subset to be processed is classified into a first image subset to be processed and a second image subset to be processed. The first image subset to be processed is then input into the corresponding first snow removal model for snow removal processing, and the second image subset to be processed is input into the corresponding second snow removal model for snow removal processing, so as to give full play to the advantages of each snow removal model and improve the snow removal efficiency and quality of the image.

[0027] In some embodiments, the target snow removal model includes: a DesnowNet model, a HDCWNet model, and a SMGARN model; S110, obtaining the snow removal feature type corresponding to the image set to be processed and the target snow removal model, further includes: According to the snow removal principles of the DesnowNet model, HDCWNet model and SMGARN model, the snow removal feature types corresponding to each model are determined.

[0028] The DesnowNet model consists of a translucency restoration module and a residual generation module, each of which contains a descriptor and a restorer. The descriptor is designed based on the Inception-v4 backbone and uses multiple dilated convolutions to extract scale-invariant features. The restorer uses convolutional layers with different receptive field sizes to convolve and select the most robust features. Assume that a snow-colored image x of size p×q∈[0,1]p×q×3 consists of a snow-free image y∈[0,1]p×q×3 and an independent snow mask. z∈[0,1]p×q×1, which only introduces the translucency of snow. This relationship can be described as: ; where ⊙ denotes element-wise multiplication and the color difference map a∈ Rp×q×3 introduces the color difference at each coordinate. To derive the estimated snow-free image yˆ from a given x, estimating an accurate snow mask zˆ as well as the color difference map a is crucial to obtain attractive visual quality. To this end, a network with multi-scale receptive fields is designed to model all the variations of snow particles. Specifically, the proposed network consists of two modules for different purposes: a translucent restoration (TR) module that restores the areas occluded by translucent snow particles; and a residual generation (RG) module that generates the residual complement r∈Rp×q×3 of the estimated snow-free image y′ for the parts fully occluded by opaque snow particles based on the transparent (uncovered) areas as well as the areas restored by the TR module. To this end, two modulation descriptors Dt and Dr are used to extract multi-scale features ft and fr with different receptive fields, and two modulation restorers Rt and Rr are designed to produce snow-free estimates. and the residual complement r. Therefore, the estimated snow-free result It can be concluded as follows: .

[0029] Regarding the HDCWNet model, the input image set to be processed is decomposed into high-frequency components and low-frequency components (i.e., {HF11, ..., HF1k} and LF1, where k is the number of subbands) by the first-level DTCWT (Dual-Tree Complex Wavelet Transforms). Then, the low-frequency components are decomposed to the next level, i.e., LF1 is decomposed into {HF21, ..., HF2k} and LF2, respectively. After this operation, the low-frequency components in each frequency can be cyclically decomposed into the i-th frequency. The high-frequency reconstruction network is applied at each scale to remove the residual snow (small and medium size) and restore detailed information, while the low-frequency reconstruction network is applied to restore structural information and remove large snow particles. Then, the reconstructed low-frequency subband (LFc) and high-frequency subband (HFd) are cyclically combined by the inverse DTCWT. Through this decomposition and reconstruction paradigm, each subband may contain a reduced spatial resolution, which means that the images in the image set to be processed are reconstructed at a small scale. Therefore, it may become easier to remove snow patterns and snowflakes of larger sizes. In addition, through better multi-directional geometric representation, snow information can be appropriately retrieved in the high-frequency sub-bands at each scale. Therefore, based on the combination of these two mechanisms, snow particles of larger sizes and various shapes can be effectively processed. The hierarchical dual-tree complex wavelet network architecture is divided into two parts: ii) high-frequency reconstruction (HR) and ii) low-frequency reconstruction (LR). These two sub-networks utilize Res2Net as the backbone. Further, multiple deconvolution, global convolution, and boundary refinement can be introduced into the network to improve the performance.

[0030] The SMGARN model consists of three parts: Mask-Net, GF-Net, and Reconstruct-Net. The backbone of Mask-Net consists of a SnowMaskBlock and two convolutional layers. In SnowMaskBlock, pixel attention units, namely self-pixel attention (SA) units and cross-pixel attention (CA) units, are designed. It should be noted that the attention used in this paper is different from the operation performed to construct the similarity matrix in the self-attention mechanism. The pixel attention used here aims to enhance the activation of snow-covered areas in snowy images to help the model better extract snow features. In addition, residual connections are added to SnowMaskBlock to facilitate feature expression. Among them, SA is used to enhance the representation of important features in the image and suppress unimportant features. CA adopts two different encoding functions to improve the adaptability of the network. The operations of CA and SA can be defined as follows:

[0031] Perform a first-level adaptive encoding on the input mask feature to obtain and , and further obtain and .

[0032] ; Afterwards, the residuals of these two features are used to obtain the first-level residual , and then obtain it through the network and .

[0033] ; Finally, use and Get the secondary residual , and then through coding fusion and Get the final characteristics relative to the snow-free period . Reconstruct-Net consists of three multi-scale aggregated residual blocks (MARBs) and a convolutional layer. Among them, MARB is specially designed to capture multi-scale features with different receptive fields. The haze effect caused by the masking effect and the residual snow removal in the previous part of the network are unevenly distributed in the image. Using the multi-scale receptive field of MARB, the pollution in these areas can be well removed and repaired. Specifically, MARB consists of 1×1, 3×3, and 5×5 parallel convolutional layers, which will pass through a 3×3 convolutional layer respectively. Then, the branch with a kernel size of 3×3 will be connected with the branches with kernel sizes of 1×1 and 5×5 to form two new branches. After performing a 3×3 convolutional layer on these two new branches, they will also be connected as new branches and perform a 3×3 convolutional layer again. Finally, the input and output features of MARB are added to enhance the information flow. The operation of Reconstruct-Net can be defined as: .

[0034] In this embodiment, according to the snow removal principles of the DesnowNet model, the HDCWNet model and the SMGARN model, the snow removal feature types corresponding to each model are determined. It should be noted that the types of target snow removal models are not limited to the DesnowNet model, the HDCWNet model and the SMGARN model, but can also be other types of snow removal models. The more types of target snow removal models there are, the higher the quality of the final image is, and correspondingly, the greater the difficulty of model training and the classification workload is. Therefore, the type and quantity of the target classification model can be reasonably selected according to actual needs, and there is no restriction on this.

[0035] In some embodiments, S120, the image set to be processed is input into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, including: The image set to be processed is input into the DBTNet network for feature classification processing to obtain the first category of image subsets to be processed, the second category of image subsets to be processed, and the third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second subset of images to be processed is adapted to the snow removal feature type of the HDCWNet model; The third subset of images to be processed is adapted to the snow removal feature type of the SMGARN model.

[0036] In this embodiment, the snow removal feature types of the DesnowNet model, HDCWNet model and SMGARN model are obtained respectively, based on which the DBTNet network is used to perform feature classification on the image set to be processed, thereby obtaining the first category of image subsets to be processed, the second category of image subsets to be processed and the third category of image subsets to be processed. It can be understood that the DesnowNet model, HDCWNet model and SMGARN model have different focuses on image processing, so there are differences between the corresponding snow removal feature types.

[0037] In some embodiments, S130, inputting at least two types of image subsets to be processed into corresponding target snow removal models for snow removal processing respectively includes: The first category of image subsets to be processed is input into the DesnowNet model for snow removal; The second type of image subset to be processed is input into the HDCWNet model for snow removal; The third category of image subsets to be processed is input into the SMGARN model for snow removal.

[0038] In this embodiment, the DBTNet network is used to classify the image set to be processed to obtain the first category, the second category and the third category of image subsets to be processed. The first category of image subsets to be processed are input into the corresponding DesnowNet model for snow removal, the second category of image subsets to be processed are input into the corresponding HDCWNet model for snow removal, and the third category of image subsets to be processed are input into the corresponding SMGARN model for snow removal. The advantages of each snow removal model can be fully utilized to improve the snow removal efficiency and quality of the image.

[0039] In some embodiments, before inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, the method further includes: The initial DBTNet network is trained using sample images and at least two different types of snow removal features to obtain a trained DBTNet network.

[0040] The specific principles of the DBTNet network are as follows: DBT consists of a semantic grouping layer and a group bilinear layer. The semantic grouping layer maps the input channels into uniformly divided groups according to their semantic information, so the most distinctive representations of a specific semantic are concentrated in one group. This representation can be further enhanced by pairwise interactions in the group bilinear layer, which performs a bilinear transformation within each semantic group. Since the bilinear transformation increases the feature dimension in each group, the output of the group bilinear layer is obtained by performing inter-group aggregation, which can significantly improve the efficiency of the output channel. In order to retain the group order information during the aggregation process, group index encoding is adopted, and the shortcut connection method is used to perform residual learning on the original features and bilinear features.

[0041] In the semantic grouping layer, high-level convolutional channels are often able to respond to specific semantic forms, so the convolutional channels can be divided into G groups according to their semantic information. S(mi) ∈[1,G] is a mapping function used to evenly distribute convolutional channels to G semantic groups, and the condition is that if ⌊i / G⌋=⌊j / G⌋, then S(mi)=S(mj), where ⌊ ⌋ is rounded down and ⌈ ⌉ is rounded up.

[0042] To obtain the bilinear features of semantic components, the channels are first arranged in the order of semantic groups: ; in, is the semantic mapping matrix, which needs to be optimized.

[0043] Since different semantic parts are located in different regions of a given image, corresponding to different positions in the convolutional features, this spatial information can be used for semantic grouping. Specifically, channels in the same semantic group are optimized with large overlapping space, while channels in different semantic groups are optimized with small overlapping space, which is formulated as the semantic grouping loss :

[0044] .

[0045] This semantic grouping can be achieved through a 1×1 convolutional layer1. Specifically, the convolutional layer can be expressed as x=Wz, where z∈ RM is the input feature, x∈ RM is the output feature, and W∈ RN×M is the weight matrix. Let U=AW, the mapping feature can be obtained by Ax=AWz=Uz, so U is the weight matrix of the semantic grouping layer, and the output feature x is grouped by semantics. Note that U is not only used for semantic grouping, but also for convolutional feature learning, so the output channels can be uniformly divided into a preset number of groups and the grouping features can be obtained during CNN training.

[0046] The role of the group bilinear layer is to efficiently obtain bilinear features without increasing the feature dimension based on the semantically grouped convolutional features.

[0047] The group bilinear layer calculates the bilinear transformation between channels within the group, which can enhance the representation of corresponding semantics through pairwise interaction. At the same time, considering that the bilinear transformation within the group increases the feature dimension of each group, in order to improve the efficiency of the output channel, these features are further aggregated: ; in , N is the number of channels. This ensures that the feature dimension remains unchanged, and DBT can be easily integrated into CNNs without modifying the network structure.

[0048] Aggregation on groups can reduce the feature dimension by a factor of G, but in this process, the group order information is lost. Therefore, position encoding is introduced in the convolution channel representation, and group index encoding terms are added to preserve group order information and improve the distinction between features in different groups: ; Where t represents the frequency of the function sin(·). Before the bilinear transformation, this group index encoding is added element by element to the j-th group of features to obtain the formula: ; in to preserve group index information during aggregation.

[0049] It should be noted that during the training process, semantic grouping constraints play an important role in both the pre-training stage and the fine-tuning stage. Semantic grouping constraints encourage channels with similar semantics to be clustered together, which is an important preprocessing for group bilinearity. In the pre-training stage, appropriate semantic grouping constraints can improve the accuracy compared to randomly grouping channels. In the fine-tuning stage, the impact of semantic grouping is smaller, but still important. In addition, group index encoding also affects the performance of the model. By adjusting the weights of group index encoding, the accuracy of the model can be affected.

[0050] The loss function of the entire model consists of softmax cross entropy loss and semantic grouping loss, where the semantic grouping loss is the sum of the losses for each block. DBT does not change the feature dimension, so it can be easily integrated into the convolutional block. DBT is performed before the 3×3 convolutional layer, and G represents the number of semantic groups. DBT has a very low computational cost, no additional parameters, and only 5M FLOPs.

[0051] During the training process, nonlinear activation functions and shortcut connections are also used to enhance the model’s representation ability and optimization process. Nonlinear activation functions (such as the tanh function) are used to activate bilinear features, while shortcut connections are used to assist in optimization, fuse the original features with the bilinear features, and build a back-propagation channel for the original features.

[0052] In this embodiment, the initial DBTNet network can be trained using sample images and at least two different types of snow removal features, thereby improving the image classification ability of the initial DBTNet network to meet image classification requirements. The more sample images and snow removal feature types, the stronger the image classification ability of the DBTNet network. It should be noted that if there are obviously abnormal images or problem images in the sample images, such as the insulator details in the image cannot be determined, such images can be classified as invalid images.

[0053] The embodiment of the present application inputs the image set to be processed into the DBTNet network for feature classification processing, and inputs the classified subsets of the image to be processed into corresponding multiple target snow removal models for snow removal processing, so that each frame of the input image to be processed can be matched with the most suitable snow removal model, thereby giving full play to the advantages of each snow removal model and improving the snow removal efficiency and quality of the image.

[0054] In some embodiments, see Figure 2 , Figure 2 The schematic diagram of the structure of a line insulator image processing device in a low-temperature snowy environment provided by an embodiment of the present application. The line insulator image processing device 200 in a low-temperature snowy environment provided by an embodiment of the present application comprises: an acquisition module 210, a classification module 220 and a processing module 230, wherein: The acquisition module 210 is configured to acquire the snow removal feature type corresponding to the image set to be processed and the target snow removal model.

[0055] The classification module 220 is configured to input the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed.

[0056] The processing module 230 is configured to input at least two types of image subsets to be processed into target snow removal models matching the snow removal feature types for snow removal processing.

[0057] In some embodiments, the target snow removal model includes: DesnowNet model, HDCWNet model and SMGARN model; the acquisition module is specifically configured as follows: According to the snow removal principles of the DesnowNet model, HDCWNet model and SMGARN model, the snow removal feature types corresponding to each model are determined.

[0058] In some embodiments, the classification module 220 is specifically configured as follows: The image set to be processed is input into the DBTNet network for feature classification processing to obtain the first category of image subsets to be processed, the second category of image subsets to be processed, and the third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second subset of images to be processed is adapted to the snow removal feature type of the HDCWNet model; The third subset of images to be processed is adapted to the snow removal feature type of the SMGARN model.

[0059] In some embodiments, the processing module 230 is specifically configured as follows: The first category of image subsets to be processed is input into the DesnowNet model for snow removal; The second type of image subset to be processed is input into the HDCWNet model for snow removal; The third category of image subsets to be processed is input into the SMGARN model for snow removal.

[0060] In some embodiments, the low temperature wind and snow environment line insulator image processing device 200 further includes a training module; the training module is specifically configured as follows: The initial DBTNet network is trained using sample images and at least two different types of snow removal features to obtain a trained DBTNet network.

[0061] The image processing device for line insulators in a low-temperature, windy and snowy environment provided in the embodiment of the present application can implement each process in the embodiment corresponding to the above-mentioned image processing method for line insulators in a low-temperature, windy and snowy environment. To avoid repetition, they will not be described here.

[0062] It should be noted that the image processing device for line insulators in a low-temperature, windy and snowy environment provided in the embodiment of the present application and the image processing method for line insulators in a low-temperature, windy and snowy environment provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned image processing method for line insulators in a low-temperature, windy and snowy environment, and the repeated parts will not be repeated.

[0063] In some embodiments, see Figure 3 , Figure 3 The electronic device 300 provided in the embodiment of the present application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program implements the above-mentioned line insulator image processing method in a low temperature wind and snow environment when executed by the processor. Specifically, the processor 310 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 310 may also include an onboard memory for cache purposes. The processor 310 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application. The memory 320 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 320 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 320 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or a flash memory; and / or a wired / wireless communication link. The present application also provides a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned line insulator image processing method in a snowy environment is implemented. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiment; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented. According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above. Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. A method for processing line insulator images in a low-temperature snowy environment, characterized in that: include: Obtain the snow removal feature type corresponding to the image set to be processed and the target snow removal model; Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed; The at least two types of image subsets to be processed are respectively input into target snow removal models matching the snow removal feature types for snow removal processing.

2. The method for processing line insulator images in a low-temperature snowy environment according to claim 1, characterized in that: The target snow removal model includes: DesnowNet model, HDCWNet model and SMGARN model; the method of obtaining the snow removal feature type corresponding to the image set to be processed and the target snow removal model also includes: According to the snow removal principles of the DesnowNet model, the HDCWNet model and the SMGARN model, the snow removal feature types corresponding to each model are determined.

3. The line insulator image processing method in a low temperature and snowy environment according to claim 2 is characterized in that: The image set to be processed is input into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, including: Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain a first category of image subsets to be processed, a second category of image subsets to be processed, and a third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second type of image subset to be processed is adapted to the snow removal feature type of the HDCWNet model; The third type of image subset to be processed is adapted to the snow removal feature type of the SMGARN model.

4. The method for processing line insulator images in a low-temperature snowy environment according to claim 3, characterized in that: The step of inputting the at least two types of image subsets to be processed into corresponding target snow removal models for snow removal processing comprises: Inputting the first type of image subset to be processed into the DesnowNet model for snow removal; Inputting the second type of image subset to be processed into the HDCWNet model for snow removal; The third category of image subsets to be processed is input into the SMGARN model for snow removal.

5. The method for processing line insulator images in a low-temperature snowy environment according to claim 1, characterized in that: Before inputting the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed, the method further includes: The initial DBTNet network is trained using sample images and at least two different types of snow removal features to obtain the trained DBTNet network.

6. A line insulator image processing device in a low temperature and snowy environment, characterized in that: include: Acquisition module, classification module and processing module, among which, The acquisition module is configured to acquire the snow removal feature type corresponding to the image set to be processed and the target snow removal model; The classification module is configured to input the image set to be processed into the DBTNet network for feature classification processing to obtain at least two types of image subsets to be processed; The processing module is configured to input the at least two types of image subsets to be processed into target snow removal models matching the snow removal feature types for snow removal processing.

7. The line insulator image processing device in a low temperature and snowy environment according to claim 6, characterized in that: The target snow removal model includes: DesnowNet model, HDCWNet model and SMGARN model; the acquisition module is specifically configured as follows: According to the snow removal principles of the DesnowNet model, the HDCWNet model and the SMGARN model, the snow removal feature types corresponding to each model are determined.

8. The line insulator image processing device in a low temperature and snowy environment according to claim 7, characterized in that: The classification module is specifically configured as follows: Inputting the image set to be processed into the DBTNet network for feature classification processing to obtain a first category of image subsets to be processed, a second category of image subsets to be processed, and a third category of image subsets to be processed; The first type of image subset to be processed is adapted to the snow removal feature type of the DesnowNet model; The second type of image subset to be processed is adapted to the snow removal feature type of the HDCWNet model; The third type of image subset to be processed is adapted to the snow removal feature type of the SMGARN model.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the line insulator image processing method in a low-temperature snowy environment according to any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images