Power transmission line inspection motion blurred image super-resolution reconstruction method and system

Abstract

The invention provides a power transmission line inspection motion blurred image super-resolution reconstruction method and system, and the method comprises the steps: obtaining a blurred image related to an insulator, carrying out the time sequence processing of the blurred image, obtaining a time sequence image, and carrying out the time reconstruction operation of the time sequence image, and obtaining a reconstructed image; processing the reconstructed image through a multi-scale space extraction network module to obtain a first feature map; processing the reconstructed image through a multi-dimensional detail enhancement module to obtain a second feature map; the width dimension of the second feature map is improved, the first feature map and the improved second feature map are spliced on the channel dimension to obtain a spliced feature map, channel dimension reduction and feature fusion are carried out on the spliced feature map through convolution to obtain a fused feature map, and weighting is carried out on the fused feature map to obtain a super-resolution image. According to the embodiment of the invention, the definition of image details can be enhanced, so that the target detection technology can identify and analyze the discharge and heating state of the insulator more accurately.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and system for super-resolution reconstruction of motion blurred images for power transmission line inspection. Background Art

[0002] Insulators are an indispensable and important component in the power system, mainly used to support and isolate high-voltage wires to prevent current leakage through the air or other media, thereby ensuring the stability and safety of power transmission. With the continuous growth of power demand and the increasing complexity of power systems, higher requirements are placed on the maintenance and detection of insulators.

[0003] Traditional insulator inspection methods mainly include manual inspection and regular maintenance. Manual inspection relies on experienced technicians to inspect by naked eye and hand tools. This method is not only inefficient, but also easily affected by human factors, making it difficult to detect potential faults in time. Although regular maintenance can prevent faults to a certain extent, it often has a large lag due to the lack of real-time performance and cannot meet the high standards of modern power systems.

[0004] With the rapid development of computer vision and deep learning technologies, image-based target detection technology has gradually become a new trend in insulator detection. Target detection technology can automatically identify and locate insulators in images, and monitor their discharge and heating status in real time through ultraviolet imagers, greatly improving the accuracy and efficiency of detection. Especially in the daily inspection of transmission lines, target detection technology can quickly detect defects in insulators, such as cracks, breakage, pollution, etc., and take timely measures to avoid power outages and safety accidents caused by insulator failures.

[0005] However, despite the great potential of target detection technology in insulator detection, it still faces some challenges. For example, the complex environment of transmission lines, the many background interferences, the different shapes and colors of insulators, and the often-affected quality of infrared and ultraviolet images all increase the difficulty of detection. Especially when using drones for shooting, motion blur or insufficient image resolution may occur, making it difficult for target detection technology to accurately identify the status of insulators. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for super-resolution reconstruction of motion blurred images during transmission line inspection, aiming to solve the problem that traditional technologies will produce motion blur and insufficient image resolution during insulator shooting, which makes it difficult for target detection technology to accurately identify the insulator status.

[0007] In a first aspect, the present invention provides a method for super-resolution reconstruction of motion blurred images of power transmission line inspection, the method comprising: Acquire a blurred image of the insulator, perform time series processing on the blurred image to obtain a time series image, and perform time attention, time aggregation and time reconstruction operations on the time series image in sequence to obtain a reconstructed image; The reconstructed image is sequentially subjected to multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing to obtain a first feature map; The reconstructed image is sequentially subjected to 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing to obtain a second feature map; The width dimension of the second feature map is enhanced, and the first feature map and the enhanced second feature map are spliced ​​in the channel dimension to obtain a spliced ​​feature map, and the spliced ​​feature map is subjected to channel dimension reduction and feature fusion through a 1x1x1 convolution to obtain a fused feature map, and the fused feature map is weighted to obtain a super-resolution image.

[0008] In a second aspect, the present invention provides a transmission line inspection motion blurred image super-resolution reconstruction system, the system comprising: A fuzzy image acquisition module is used to acquire a fuzzy image of the insulator, perform time series processing on the fuzzy image to obtain a time series image, and perform time attention, time aggregation and time reconstruction operations on the time series image in sequence to obtain a reconstructed image; A first convolution processing module, used for sequentially performing multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing on the reconstructed image to obtain a first feature map; A second convolution processing module is used to sequentially perform 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing on the reconstructed image to obtain a second feature map; A feature fusion module is used to enhance the width dimension of the second feature map, and to splice the first feature map and the enhanced second feature map in the channel dimension to obtain a spliced ​​feature map, and to perform channel dimension reduction and feature fusion on the spliced ​​feature map through a 1x1x1 convolution to obtain a fused feature map, and to weight the fused feature map to obtain a super-resolution image.

[0009] In a third aspect, the present invention provides a storage medium storing one or more programs, which, when executed by a processor, implement the above-mentioned method for super-resolution reconstruction of motion blurred images of power transmission line inspection.

[0010] In a fourth aspect, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, the above-mentioned transmission line inspection motion blurred image super-resolution reconstruction method is implemented.

[0011] Compared with the prior art, the embodiments of the present invention have the following advantages: 1. Through the super-resolution reconstruction technology proposed in the present invention, low-quality, blurred images can be upgraded to high-resolution images, thereby enhancing the clarity of image details, so that the target detection technology can more accurately identify and analyze the state of insulator discharge and heating. This method effectively solves the problem of insufficient quality of traditional images, improves the accuracy and reliability of detection, and ensures the real-time and accuracy of insulator maintenance in the power system. Specifically, for the continuous frame images taken by the drone, the blurred images are first arranged in sequence to prevent the image entering the backbone network from having a time sequence disorder problem, and then the time sequence images with time sequence features are processed to combine the time sequence convolution with the time attention mechanism, and the areas blurred by motion in the image are analyzed. By extracting information from the unblurred frames, the information of the blurred frames is repaired, so as to improve the image blurred by motion and obtain a reconstructed image; then the improved reconstructed image is divided into two parallel processes, one of which is to extract comprehensive spatial features from the input feature map (extracting information features about space from low-resolution and spatially detail-deficient images); the other branch processes scenes that require fine-grained feature enhancement and complex high-dimensional feature processing (tasks that require enhanced feature details from significant blur in the input data). The first and second feature maps after processing are fused. The fusion method is to first splice the two feature maps along the channel dimension, and then perform channel dimension reduction and feature fusion through 1x1x1 convolution (recombine channel information at each pixel position through learning weights). This can enhance the information exchange between the two streams while reducing the number of channels. Then, the attention mechanism is used for weighting, and important channels are amplified, while unimportant channels are suppressed, thereby improving the model's ability to focus on key features. Then, upsampling is used to make the output image consistent with the input image size to obtain a high-resolution image. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A flowchart of a method for super-resolution reconstruction of motion blurred images of power transmission line inspections proposed in one embodiment of the present invention; Figure 2 The present invention is a schematic diagram of the structure of a system for super-resolution reconstruction of motion blurred images for power transmission line inspection proposed in one embodiment of the present invention.

[0013] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0015] like Figure 1 As shown, an embodiment of the present invention provides a method for super-resolution reconstruction of motion blurred images of power transmission line inspection, the method comprising steps S101 to S104, wherein: Step S101: acquiring a blurred image of an insulator, performing time series processing on the blurred image to obtain a time series image, and sequentially performing time attention, time aggregation and time reconstruction operations on the time series image to obtain a reconstructed image; It should be noted that the blurred image is taken by the drone through the imager, and then the low-resolution image (blurred image) taken by the drone is input into the temporal convolution module. The temporal convolution module is a convolution layer used to process time series data. Its main function is to keep the order of the image in the time dimension (that is, the dimension of the sequence) through convolution operation, and perform convolution operation on the spatial dimension. This module uses 3D convolution to process time series image data, maintaining the consistency of the time sequence without disrupting the time sequence of the image. The shape of the input image is (batch_size, 5, 416, 416, 3) batch_size represents the batch, each sample has 5 frames of images, each frame is 416x416, and each frame has 3 channels. After permute, the X shape is converted to (2, 3, 5, 416, 416). After the convolution operation, the output shape is (batch_size, 64, 5, 416, 416), and then the permute output image is (batch_size, 5, 416, 416, 64), so that each sample still has 5 frames of images, each frame size is 416x416, but the number of channels has been converted from 3 to 64.

[0016] Specifically, the goal of the temporal convolution module is to process multidimensional data with a time dimension. No convolution operation is performed on the time dimension, but only on the spatial dimension (height and width). In this way, the model can capture temporal data and spatial features while maintaining the continuity of temporal information. The input of this module is (batch_size, 5, 416, 416, 3) (batch, time step, height, width, number of channels); the output is (batch_size, 5, 416, 416, 64) (batch, time step, height, width, number of channels).

[0017] By inputting (batch_size, 5, 416, 416, 3), 5 is the number of picture frames, which can be adjusted according to the specific situation, 416, 416 corresponds to the image size, and 3 is the number of channels. After permute, the dimension is transformed to generate (batch_size, 3, 5, 416, 416); after Conv3D convolution, it is generated (batch_size, 64, 5, 416, 416); after permute, the dimension is transformed to generate (batch_size, 5, 416, 416, 64).

[0018] Then it is input into the temporal feature network module, which extracts features from the input reconstructed image, reconstructs the temporal dimension, and restores the temporal information. It also processes the temporal and spatial features and finally outputs a feature map with rich temporal information. That is, it learns the features of several similar steps and then reconstructs the best feature map. The specific working method is: (batch_size, 5, 416, 416, 64) is input into the Conv3D and pooling layers and activation functions (this process does not change the image dimension) and the output (batch_size, 5, 416, 416, 64) is input into the temporal attention mechanism module.

[0019] The temporal attention mechanism module assigns weights to each time step in the temporal dimension, thereby enhancing the model's attention to important time steps. Its input is (batch_size, 5, 416, 416, 64) (batch, time step, height, width, number of channels); its output is (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); the working method is: first permute (dimensional rearrangement) output (batch_size, 64, 5, 416, 416); then flatten the spatial dimension to generate .

[0020] Compute query, key, value:

[0021]

[0022]

[0023] Among them: Q (Query) is used to query the weight vectors of other time steps, represents the combination of all samples and channels, 5 represents the time dimension, Represents the spatial features of the time step; K (Key) represents the features of each time step, which is used to calculate the similarity with Q. Through the inner product operation with Q, K can help determine which time steps should be paid more attention to, and the shape is the same as Q; V (Value) is the actual value calculated at each time step. It carries the feature information of each time step, and is weighted with the attention weight (the similarity between Q and K). The weighted features are finally output, which mainly provide the information that "aggregation is needed". Then the attention weight output (batch_size, 64, 5, 416, 416) is calculated, and then (batch_size, 64, 5, 416, 416) is input into the time aggregation module.

[0024] The time aggregation module aggregates the input features in the time dimension, aggregates the time features of each position through a fully connected layer, and obtains a compressed representation of information about time. Its input is: (batch_size, 64, 5, 416, 416) (batch, number of channels, time step, height, width); its output is: (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); its working method is: (1) permute (rearrange dimensions and move the time dimension to the end to facilitate operations on the time series at each spatial location) output (batch_size, 64, 416, 416, 5).

[0025] (2) Flatten the dimensions by using x.view to flatten the height, width, and channel dimensions of the spatial dimension, flattening the first four quantities into one quantity , extract the time step feature vector of each spatial position to form a two-dimensional matrix, where each row represents the features of a spatial position at different time steps, and each column corresponds to the features of all spatial positions at the same time step.

[0026] Specifically, a two-dimensional matrix is ​​formed according to the following formula: ; in, represents the flattened matrix, represents the eigenvalue of the Hth row and the Wth column at time step T, and B represents the feature Figure 1The number of samples input, C represents the number of channels of the feature map, H represents the height of the input feature map, W represents the width of the input feature map, and T represents the number of time steps.

[0027] (3) Through the fully connected layer, Aggregation Output.

[0028] Time aggregation is performed according to the following formula: ; in, represents an aggregate image, is a weight matrix, are the weight values ​​learned by the fully connected layer. Represents the transpose of the weight matrix.

[0029] (4) Yes By rearranging the tensor dimension operation, the shape (i.e., dimension) of the tensor is changed, but the data in it is not changed. The output (batch_size, 64, 1, 416, 416) is input into the temporal feature reconstruction module.

[0030] The temporal feature reconstruction module mainly performs a "deconvolution" operation on the input feature map to help the model better learn and represent features in the temporal dimension, enhance the ability to express temporal information, and optimize the ability to restore feature details through deconvolution operations. Its input is: (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); the output is: (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); its working method is: The calculation formula for the deconvolution operation is as follows:

[0031] in, Indicates the size of the output feature map; Represents the size of the input feature map; Represents the offset of the convolution operation, which is determined by the size and stride of the convolution kernel; represents the step size, which is the movement step of the convolution kernel on the input feature map each time; 2 represents the effect of padding, because padding is performed on both ends of the input at the same time, so all must be multiplied by 2; Indicates padding, the number of zeros added to each dimension (depth, height, width) of the input data; Represents the size of the convolution kernel, the size of the convolution kernel in each dimension. Represents output padding, which is used to add additional zero padding to the output feature map. Here input , we can get the output depth and width by substituting it into the above formula. So the final output is (batch_size, 64, 1, 416, 416), which is the reconstructed image.

[0032] Step S102: performing multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimension conversion processing on the reconstructed image in sequence to obtain a first feature map; It should be pointed out that this step is to obtain the first feature map by inputting the reconstructed image into the multi-scale spatial extraction network module. The multi-scale spatial extraction network module mainly performs deep feature extraction and optimization on the output feature map of the temporal feature reconstruction module to extract deeper spatial and temporal features; increases the receptive field and optimizes feature learning through the hole convolution kernel residual block; weights important features through the attention mechanism to enhance the model's attention to key features; reduces the dimension and outputs the optimized feature map. Its input is: (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); output is: (batch_size, 1, 416, 416, 64) (batch, time step, height, width, number of channels). Its working method is: First, the number of channels of the input feature map (batch_size, 64, 1, 416, 416) is increased to (batch_size, 256, 1, 416, 416) through two consecutive basic 3D convolution operations; then the receptive field is expanded through the dilated convolution and input into the dilatedConv.3D. The calculation formula is: ; in, Indicates the output size, including the output depth, width and height. represents the input size, Indicates filling, represents the depth of the convolution kernel, represents the size of the convolution kernel, Indicates the step length.

[0033] Exemplarily, the depth dimension is calculated: ; in, ; Calculate the height and width dimensions:

[0034] in, ; Similarly, the feature size of the output of the pooling layer and activation function after the dilated convolution is: (batch_size, 512, 1, 416, 416).

[0035] The result of the dilated convolution is then processed by a residual block (consisting of two convolutional layers, each of which keeps the number of channels and spatial dimensions unchanged), and then the output of the residual block is added to the result of the dilated convolution through a residual connection to output the feature map (batch_size, 512, 1, 416, 416).

[0036] Then the attention mechanism is processed: (1) Average pooling: average pooling is performed on the feature map to generate a weight for each channel to enhance important features and suppress unimportant features. The formula for global pooling is as follows:

[0037] in, and are the height and width of the input feature map respectively; is the bth sample, cth channel, and cth time step, hth height position, The value of the width position; after global pooling, a weighted feature value of a feature map (batch_size, channel, 1, 1) is generated. Here, assuming that the input b=1 and the first channel c=1, the average pooling calculation process is as follows: ; in, is the weighted feature value of channel 1 in sample 1; 416 and 416 are the height and width of the input feature map; It is the value of the 1st sample, 1st channel, 1st time step, 1st height position, 1st width position of the input tensor.

[0038] (2) Convolution generates attention weights. First, through a 1x1 convolution layer, the feature map after global pooling can be mapped to a new tensor, and this mapping will be restricted by the activation function. The convolution operation is as follows: ; in: It is Samples, The attention weight of each channel; Indicates that through a The convolutional layer is mapped to generate a tensor of (batch_size, channels, 1, 1); Represents the activation function (Sigmoid), which is used to ensure that the weight outputs a reasonable range of values ​​between 0 and 1.

[0039] (3) The attention weights are extended to each spatial position (d, h, w, i.e. depth, height, and width) through a broadcast mechanism.

[0040]

[0041] in, are the attention weights expanded by the broadcast mechanism.

[0042] (4) Attention weights after applying broadcast: ; in, is the output tensor, where the features of each spatial position are adjusted by the attention weight of the corresponding channel, is the bth sample, cth channel, dth time step, hth height position, The value of the width position.

[0043] Then the final convolution is performed to reduce the dimension back to the number of input channels: the Conv.3D convolution kernel is (1, 1, 1), padding=0, the number of image channels is reduced to be consistent with the input image dimension, and the output feature map is (batch_size, 64, 1, 416, 416); then permute is performed again to transform the dimension; the feature map is converted from (batch_size, 64, 1, 416, 416) to (batch_size, 1, 416, 416, 64); the output here (batch_size, 1, 416, 416, 64) is the final output of the multi-scale space extraction network module, that is, the first feature map is obtained.

[0044] Step S103: the reconstructed image is sequentially subjected to 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing to obtain a second feature map; It should be noted that this step is to process complex data containing spatiotemporal information by inputting the reconstructed image into the multidimensional detail enhancement module, and to improve the accuracy and efficiency of data processing through multidimensional feature extraction, feature fusion and time series analysis. The input of the multidimensional detail enhancement module is: (batch_size, 64, 1, 416, 416) (batch, number of channels, time step, height, width); the output is: (batch_size. 1000, 416, 416, 1) (batch, number of channels, time step, height, width); its working method is: First, use 3D convolution to process time series data. Specifically, use three consecutive 3D convolutions. The parameters of the three convolutions are the convolution kernel size (3, 3, 3) and padding = (1, 1, 1). The number of channels of (batch_size, 64, 1, 416, 416) is increased to (batch_size, 1024, 1, 416, 416), and then input into the spatial attention mechanism. The spatial attention mechanism module dynamically adjusts the position of each weight in the feature map by inputting different spatial positions of the feature map. Specifically, the spatial attention mechanism module generates a spatial attention map and applies it to the input data, so that the model pays more attention to important spatial positions and ignores unimportant areas. The spatial attention mechanism module works as follows: (1) Convolutional layer: The number of channels of the input feature map (batch_size, 1024, 1, 416, 416) is reduced to (batch_size, 1, 1, 416, 416) through three 3D convolutions (the convolution kernel is 1 and the padding is 0), and nonlinearity is introduced by applying the activation function. The sigmoid function then compresses the value of the convolution output to the range of [0, 1], generates the attention weight and applies it.

[0045] (2) Weighted input feature map: The generated spatial attention map is multiplied element-by-element with the input feature map to achieve weighting. The features of each spatial position will be enlarged or reduced according to its corresponding attention weight. This process can enhance the features of the focused area and suppress the features of unimportant areas.

[0046] (3) The output is a weighted feature map with the same shape as the input feature map. Through the spatial attention mechanism, the model can dynamically adjust the features of each position according to the attention value of the spatial position.

[0047] Right now: ; Here the output shape is: (batch_size, 1024, 1, 416, 416).

[0048] Then reshape (batch_size, 1024, 1, 416, 416) into ; Then perform the permute operation, and then process it through the 3D multi-head attention mechanism. The input and output shapes of the 3D multi-head attention mechanism are (L, N, E), where L is the sequence length; N is the batch size; E is the embedding dimension. In this embodiment, L=416x416=173056; ; E=1024.

[0049] In addition, it should be noted that the feature map processed by the 3D multi-head attention mechanism is passed through three linear layers to generate query, key, and value matrices respectively; Divide the query, key, and value matrices into multiple heads, each head corresponds to an embedding dimension, and the embedding dimension of each head is: , represents the total embedding dimension in the 3D multi-head attention mechanism, Represents the total number of heads in the 3D multi-head attention mechanism. Specifically, the parameters of the first multi-head attention layer (attn1) are: embedding dimension (embed_dim) = 1024; number of heads (num_heads) = 8; the embedding dimension of each head is 128.

[0050] In addition, the input image is passed through three linear layers to generate query, key, and value matrices respectively.

[0051]

[0052] in, is the input image; is the weight matrix obtained through learning, corresponding to the mapping of query, key and value respectively; the calculated represents the embedding dimension at each position.

[0053] Specifically, for the input :

[0054] in, The shape of is (173056, Batch_size, 1024), which means the embedding dimension of each position is 1024.

[0055] Next, we perform head splitting: split the attention mechanism query, key, and value into multiple heads to process information independently, and each head has a lower dimension. Assuming there are h heads, the embedding dimension of each head is:

[0056] Where E is the embedding dimension of the input vector; h is the number of heads, which is used in multi-head attention to increase the parallelism of the model and capture information from different aspects; represents the dimension of the feature subset processed by each head, allowing each head to capture different aspects of the features in a smaller dimensional space. For each head i:

[0057] in, are the linear transformation weights of each head. Divide Q, K, and V into 8 heads, each with a dimension of 128. For each head i:

[0058] Then Scaled Dot-Product Attention: For each head, calculate the attention score and apply it to the value.

[0059]

[0060] in, is the query matrix; is the key matrix; Value matrix; This is a scaling factor to prevent the dot product from being too large and causing the gradient to disappear. Because there are 173056 positions, each position generates a 2x2 matrix, so the overall dot product result shape is (173056, 2, 2).

[0061] Then perform splicing and linear transformation: splice the outputs of all heads together:

[0062] in, It means concatenating the output vector of each head on the last dimension feature; assuming there are 8 attention heads, is the output of each head; Represents the result after splicing. The dimension of each head output is set here. , after splicing, we get a full 1024 dimension, and the output shape after splicing is .

[0063] Then through linear transformation:

[0064] in, Represents a dimension The weight matrix of Through weight The shape of the transformed output is determined by the number of columns in the weight matrix to match the input dimension of the next layer of the model. The calculation method of the parameters of the second multi-head attention layer is similar to the first one, and will not be repeated in this embodiment.

[0065] Then restore the shape of the output processed by the 3D multi-head attention mechanism: Step 1: permute to convert (173056, Batch_size, 1024) to (Batch_size, 1024, 173056).

[0066] Step 2: Reshape the result of the first step permutation, that is, convert (Batch_size, 1024, 173056) to (Batch_size, 1, 1024, 416, 416).

[0067] Step 3: permute, convert the (Batch_size, 1, 1024, 416, 416) reshaped in the second step into (Batch_size, 1024, 1, 416, 416).

[0068] Then (Batch_size, 1024, 1, 416, 416) is convolved once to generate (Batch_size, 2048, 1, 416, 416), and (Batch_size, 2048, 1, 416, 416) is convolved once again to generate (Batch_size, 4096, 1, 416, 416).

[0069] Then perform the residual connection:

[0070] in, is the result after adding the residual term, is the feature map input to the residual block, is the residual term.

[0071] by is the input of the main path; is the first residual term; is the second residual term.

[0072] Then average pooling is performed. The purpose of average pooling is to reduce the spatial dimension of the feature map by averaging in the local area, which helps to reduce overfitting, ensure the main information of the feature map, and reduce the consumption of computing resources. Its input is (B, 4096, 1, 416, 416) and the output is (B, 4096, 1, 1, 1).

[0073] Adaptive average pooling layer formula: ; in, Indicates that the feature map obtained after the pooling operation is in the batch , channel c, depth ,high ,width The value of the position, Represents the original input feature map in batch b, channel c, depth d, height h, width The value of the position, represents the average pooling factor, and Respectively represent the scaling ratio of the pooling window in height and width, represents the rounding down operation, i and j represent the row and column indices in the pooling window, respectively. , Represent the height and width of the output feature map respectively.

[0074] Exemplarily, determine the height direction of the pooling window: input height , output height , the height of the pooling window ; In width direction: enter the width , output width , the height of the pooling window ; Then divide the entire 416x416 feature map into a large window for pooling:

[0075] Depth (D') = 1, so the only depth index is 0; height (H') = 1, so the only height index is 0; width (W') = 1, so the only width index is 0.

[0076] For each batch and each channel , calculate all spatial positions The final output Y retains only one value for each batch and channel, which represents the average activation value of the channel in the entire space.

[0077] In addition, the flattening operation converts a multi-dimensional tensor into a two-dimensional tensor, which is usually used to connect the output of the convolutional layer to the fully connected layer. The pooled [B, 4096, 1, 1, 1] is flattened to [B, 4096], that is, the redundant dimensions are removed.

[0078]

[0079] in, Represents the flattened output matrix, which is a two-dimensional matrix in which each row represents the flattened feature vector of a sample and is input to the fully connected layer of the subsequent network; Represents the feature map output from the average pooling layer, which is a multi-dimensional feature tensor, here is [B, 4096, 1, 1, 1].

[0080] The input of the fully connected layer is: [B, 4096], which is linearly transformed. The output is: [B, 1000]. The fully connected layer can be regarded as a linear transformation, expressed as:

[0081] in, is the input feature map of the fully connected layer, with a dimension of [B, 4096], where B represents the batch, and each sample has 4096 features; W is the weight matrix of the fully connected layer, with a dimension of [4096, 1000], which converts 4096 input features into 1000 output nodes, each of which corresponds to a part of the final output layer of the network; b is the bias vector with a dimension of 1000, which is used to add a bias value to each output node to help the model better fit the data; Y is the output feature map of the fully connected layer, with a dimension of [B, 1000]. For each sample in the batch, the fully connected layer outputs a 1000-dimensional vector, which is usually used for subsequent classification and regression tasks.

[0082] The specific calculation steps of linear transformation are: For each input sample ,if , corresponding Batch_size=2.

[0083] The first step is matrix multiplication:

[0084] Among them, W is the weight matrix of the fully connected layer, is the output feature map corresponding to the i-th sample.

[0085] Specifically, for the i-th sample, the j-th feature of the output is, , calculated as:

[0086] in, is the value of the input tensor X at the i-th sample and k-th feature position, is the value of the weight matrix W in the kth row and jth column, Represents the output feature map corresponding to the jth feature of the i-th sample.

[0087] The second step is to add bias:

[0088] Right now:

[0089] in, is the value of the bias vector at the jth position.

[0090] Then expand and repeat the tensor, specifically: (1) Expand: Input: [B, 1000], to expand the dimension.

[0091] Output: [B, 1000, 1, 1, 1].

[0092]

[0093] in, Corresponding batch size B, Corresponding output channel number , (Because the dimension size after expansion is 1), Represents the feature map after the input feature map is expanded.

[0094] (2) Repeat: Input: [B, 1000, 1, 1, 1], repeat.

[0095] Output: [B, 1000, 416, 416, 1].

[0096]

[0097] in, This means that the depth dimension is repeated 416 times. It means the height dimension is repeated 416 times. Indicates that the width dimension remains unchanged because the target dimension size is 1. [B, 1000, 416, 416, 1] will be used as the output feature map of the multidimensional detail enhancement module, that is, the second feature map. Represents the feature map where the input feature map is repeated.

[0098] Step S104: enhance the width dimension of the second feature map, and splice the first feature map and the enhanced second feature map in the channel dimension to obtain a spliced ​​feature map, and perform channel dimension reduction and feature fusion on the spliced ​​feature map through 1x1x1 convolution to obtain a fused feature map, and weight the fused feature map to obtain a super-resolution image.

[0099] It should be pointed out that in this step, the first feature map and the second feature map are input into the dual-stream feature fusion module. The first function of this module is to integrate the output features from the multi-scale spatial extraction network module and the multi-dimensional detail enhancement module, effectively combining two different perspectives and feature expressions, and increasing the diversity of features; the second function is to apply the attention mechanism to dynamically adjust the feature response of each channel through global pooling and fully connected layers, which not only compresses the streamlined features, but also emphasizes the more informative parts. The specific working method is as follows: First, increase the width dimension of the multi-dimensional detail enhancement module output:

[0100] in, represents the function used to resize the input feature map; Represents the input feature map, that is, the feature map input by the multi-dimensional detail enhancement module; Defines the target size after interpolation; Determines the alignment strategy of corner pixels in the interpolation method. When set to True, the corner pixels of the input and output tensors will be aligned, thereby keeping the position of the image edges unchanged, which helps to reduce geometric deformation. is the adjusted feature map.

[0101] Then execute the splicing feature: Concatenate the MESNet and resized SFGNN outputs in the channel dimension:

[0102] in, It is a splicing operation along the dimension; M represents the feature map output by the multi-scale space extraction network module; C is a new feature map after splicing the feature maps output by the multi-scale space extraction network module and the multi-dimensional detail enhancement module; It means that we assume that we bring in a value to get the concatenated graph.

[0103] Then perform the fusion feature: Use 1X1X1 convolution for channel dimension reduction and fusion:

[0104] in, Represents a 1X1X1 three-dimensional convolution operation; represents the fused feature map, with a shape of {batch_size, 512, 416, 416, 64}; B represents the batch of the fused feature map; Indicates the number of channels of the fused feature map; Indicates the depth of the fused feature map; Indicates the height of the fused feature map; Indicates the width of the fused feature map. For each batch batch_size, output channel ,depth ,high ,width , the feature formula after fusion is:

[0105] in, Represents the fused image batch; Indicates the number of channels of the fused feature map; is the depth of the fused feature map; is the height of the fused feature map; is the width of the fused feature map; is the original number of channels, that is, the number of channels of the output feature map after the previous concatenation operation; is the number of additional channels; The feature map output by the concatenation operation is the input feature map of the fusion feature operation; is a weight tensor used to transform the input features from channel c to the new ; It is a point convolution, i.e. 1x1x1 convolution, which is used to change the channel dimension without affecting the spatial dimension; is the bias term, which is the new channel Provides additional numerical adjustments.

[0106] Then, the CBAM attention mechanism is applied, and the Convolutional Block Attention Module (CBAM) is introduced to combine channel attention and spatial attention to enhance the fusion features. Specifically: The channel attention weight is calculated through the channel attention module:

[0107] in, Represents the output attention feature map, which has the same shape as the input feature map, but each channel has been weighted by the attention mechanism; Represents global average pooling, which helps extract global spatial information, reduce spatial dimension information, and retain channel dimension information for calculating channel attention; (assuming r=16 is the dimensionality reduction ratio) and All represent convolution weights; The activation function is used to introduce nonlinearity. The results are nonlinearly mapped.

[0108] Perform global average pooling:

[0109] in, Represents the output after global average pooling, which is a new tensor that contains the global information obtained through the pooling operation. In the channel dimension, the output is a feature vector that represents the average feature of each channel; Represents a 3D average pooling operation, which is applied to the fused feature map to calculate the average value of a channel from each spatial position to extract global information.

[0110]

[0111] in, It is the output feature map after the maximum pooling operation, which indicates the maximum value extracted from each channel. The output of each channel is compressed into a scalar (maximum value). It is a 3D maximum pooling operation, which is applied to the input feature map. This operation applies maximum pooling to each channel in the spatial dimensions (depth, height, width), and selects the maximum value in each spatial region. The pooling operation can extract the most significant features while reducing the spatial dimensions and retaining the global significant information. Represents the output feature map after pooling.

[0112] Perform global average pooling and add the results:

[0113] Among them, M is the final global pooling result obtained by adding the output feature maps after the global average pooling and maximum pooling operations; Represents the output after global average pooling, which is a new tensor that contains the global information obtained through the pooling operation. In the channel dimension, the output is a feature vector that represents the average features through each channel.

[0114] Perform two layers of convolution kernel ReLU activation:

[0115] in, Represents the output tensor after processing by the convolution kernel (ReLU), Contains the final weighted kernel adjustment result of the fused features; Represents the convolution operation; It is the calculation of the convolution operation plus the bias term; is the convolutional layer weight applied to the pooled tensor, which is used to process the spatial information of the input; M is the feature map after the previous pooling operation, which contains the global information after pooling; is the bias term, which is added to the feature map together with the convolution operation to further adjust the output; is the output tensor shape.

[0116] Perform the first weighting operation:

[0117] in, Represents the weighted feature map, with a shape of [Batch_size, Channels, Depth, Height, Width], which means that the spatial dimensions (depth, height, width) and number of channels of the feature map have been adjusted through the channel attention mechanism.

[0118] Calculate spatial attention weights:

[0119] in, Indicates that and Splicing is performed in the channel dimension, Indicates that the concatenation operation occurs on the channel dimension.

[0120] Perform spatial pooling:

[0121] in, Represents the spatial average pooling result of the output, which represents the average value of all channels in the feature map; is the normalization factor; It is to accumulate each channel c in the weighted feature map, which is a given spatial position. Sum all channel values ​​of ; For all batches b and all spatial locations Perform the above operations.

[0122]

[0123] in, Represents the output spatial maximum pooling result, which represents the maximum value of all channels in the feature map; Indicates that the maximum value is selected for each channel c in the weighted feature map, which is for a given spatial position The maximum value of all channel values ​​is extracted.

[0124] The result of splicing pooling is:

[0125] in, Represents the result of concatenating spatial average pooling and spatial maximum pooling.

[0126] Perform convolution and sigmoid activation:

[0127] in, is the convolution kernel weight, which is used to extract useful spatial features from the concatenated feature map to generate attention weights; Represents the result of concatenating spatial average pooling and spatial maximum pooling; is the bias vector, which is used for adjustment after the convolution operation to ensure the output is offset within the appropriate range. Represents the spatial attention map of the output; Perform the second weighting operation:

[0128] in, Represents the weighted eigenvalue map of the output.

[0129] Execute the activation function (finally apply the ReLU activation function to introduce nonlinearity):

[0130] in: It is the final feature map of the output, that is, the final output feature map of the entire network (super-resolution image).

[0131] In summary, the traditional super-resolution reconstruction network can achieve good reconstruction effects on a single blurred image, but since most of the objects it faces are restoration of old images or single images generated by low-resolution visible light devices, it does not take into account the motion blur caused by the movement of the shooting equipment in actual applications, making it difficult to promote the traditional super-resolution reconstruction network in actual applications in the power field. Therefore, in order to solve the above-mentioned problems, the embodiment of the present invention proposes a combination of spatiotemporal feature extraction and dual attention mechanism (time + space), which enables the model to aggregate information from the dual dimensions of time and space when processing time series data, thereby enhancing the overall feature expression ability, and being able to reconstruct and restore the motion blurred images of drones during the shooting process, providing a reliable method for insulator image reconstruction.

[0132] like Figure 2 As shown, an embodiment of the present invention further provides a transmission line inspection motion blurred image super-resolution reconstruction system, the system comprising: A fuzzy image acquisition module 10 is used to acquire a fuzzy image of the insulator, perform time series processing on the fuzzy image to obtain a time series image, and perform time attention, time aggregation and time reconstruction operations on the time series image in sequence to obtain a reconstructed image; A first convolution processing module 20 is used to sequentially perform multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing on the reconstructed image to obtain a first feature map; A second convolution processing module 30 is used to sequentially perform 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimensional repetition processing on the reconstructed image to obtain a second feature map; The feature fusion module 40 is used to enhance the width dimension of the second feature map, and to splice the first feature map and the enhanced second feature map in the channel dimension to obtain a spliced ​​feature map, and to perform channel dimension reduction and feature fusion on the spliced ​​feature map through a 1x1x1 convolution to obtain a fused feature map, and to weight the fused feature map to obtain a super-resolution image.

[0133] On the other hand, the present invention further proposes a storage medium on which one or more programs are stored, and when the programs are executed by a processor, the above-mentioned method for super-resolution reconstruction of motion blurred images of power transmission line inspection is implemented.

[0134] On the other hand, the present invention further proposes an electronic device, including a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the above-mentioned transmission line inspection motion blurred image super-resolution reconstruction method.

[0135] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain storage, communication, propagation or transmission of a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0136] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0137] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0138] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for super-resolution reconstruction of motion blurred images of power transmission line inspection, characterized in that: The method comprises: Acquire a blurred image of the insulator, perform time series processing on the blurred image to obtain a time series image, and perform time attention, time aggregation and time reconstruction operations on the time series image in sequence to obtain a reconstructed image; The reconstructed image is sequentially subjected to multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing to obtain a first feature map; The reconstructed image is sequentially subjected to 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing to obtain a second feature map; The width dimension of the second feature map is enhanced, and the first feature map and the enhanced second feature map are spliced ​​in the channel dimension to obtain a spliced ​​feature map, and the spliced ​​feature map is subjected to channel dimension reduction and feature fusion through a 1x1x1 convolution to obtain a fused feature map, and the fused feature map is weighted to obtain a super-resolution image.

2. The method for super-resolution reconstruction of motion blurred images of power transmission line inspection according to claim 1, characterized in that: The steps of acquiring a blurred image of the insulator, performing time series processing on the blurred image to obtain a time series image, and sequentially performing time attention, time aggregation and time reconstruction operations on the time series image to obtain a reconstructed image include: Input blurred image, and then go through permute function dimension transformation, Conv3D convolution, and permute function dimension transformation to get time series image. Input the time series image into the Conv3D convolution layer, the pooling layer and the activation function layer to obtain a time feature map; The temporal feature graph is sequentially subjected to dimension rearrangement and spatial dimension expansion by a permute function to obtain a spatial dimension graph, and the query, key, and value of the spatial dimension graph are calculated to weight the query, key, and value of the spatial dimension graph to output a weighted feature image; The weighted feature image is rearranged in dimension by a permute function and expanded in spatial dimension to obtain a feature vector, and the feature vector of each spatial position is extracted to form a two-dimensional matrix; Performing temporal aggregation according to the two-dimensional matrix to obtain an aggregated image, and rearranging the tensor dimension of the aggregated image; The rearranged aggregated image is subjected to a deconvolution operation to obtain a reconstructed image.

3. The method for super-resolution reconstruction of motion blurred images of power transmission line inspection according to claim 2 is characterized in that: The step of rearranging the weighted feature image through the permute function dimension and expanding the spatial dimension to obtain a feature vector, and extracting the feature vector of each spatial position to form a two-dimensional matrix includes: The two-dimensional matrix is ​​formed according to the following formula: ； in, represents the flattened matrix, represents the feature value of the Hth row and the Wth column at time step T, B represents the number of samples input to the feature map at one time, C represents the number of channels of the feature map, H represents the height of the input feature map, W represents the width of the input feature map, and T represents the number of time steps; In the two-dimensional matrix, each row represents a feature vector of a spatial position at different time steps, and each column corresponds to the feature vectors of all spatial positions at the same time step; The step of performing time aggregation according to the two-dimensional matrix to obtain an aggregated image comprises: Time aggregation is performed according to the following formula: ； in, represents an aggregate image, is a weight matrix, are the weight values ​​learned by the fully connected layer. represents the transpose of the weight matrix; The step of performing a deconvolution operation on the rearranged aggregated image to obtain a reconstructed image comprises: The deconvolution operation is performed according to the following formula: ； in, represents the size of the output feature map, represents the size of the input feature map, Represents the step size, the movement step of the convolution kernel on the input feature map each time, Indicates padding, the number of zeros added to each dimension of the input data. Represents the size of the convolution kernel, the size of the convolution kernel in each dimension, Represents output padding, which is used to add additional zero padding to the output feature map.

4. The method for super-resolution reconstruction of motion blurred images of power transmission line inspection according to claim 3 is characterized in that: The step of sequentially performing multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing on the reconstructed image to obtain a first feature map includes: The channels of the reconstructed image are enlarged by two consecutive 3D convolution operations, and the enlarged reconstructed image is subjected to dilated convolution processing according to the following formula: ； in, Indicates the output size, including the output depth, width and height. represents the input size, Indicates filling, represents the depth of the convolution kernel, represents the size of the convolution kernel, represents the step length; The result of the dilated convolution is processed by a residual block, and then the output of the residual block is added to the result of the dilated convolution through a residual connection to output a feature map. The residual block consists of two convolutional layers, and each convolutional layer keeps the number of channels and spatial dimensions unchanged. Perform average pooling on the feature map: ； in, It is a channel In the sample The weighted eigenvalues ​​on , and are the height and width of the input feature map, respectively. is the bth sample, cth channel, 1st time step, hth height position, The value of the width position; Map the average pooled feature map to a new tensor: ； in, It is Samples, The attention weight of each channel, Indicates that through a The convolutional layer is used for mapping. Represents the activation function, which is used to make the weight range output between 0 and 1; The attention weights are extended to each spatial position via a broadcasting mechanism: ； in, is the attention weight after expansion through the broadcast mechanism; Attention weights after applying broadcasting: ； in, is the output tensor, where the features of each spatial position are adjusted by the attention weight of the corresponding channel, is the bth sample, cth channel, dth time step, hth height position, The value of the width position; The output tensor is processed by dimensionality reduction and dimensionality conversion to obtain the first feature map.

5. The method for super-resolution reconstruction of motion blurred images of power transmission line inspection according to claim 4 is characterized in that: The step of sequentially performing 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing on the reconstructed image to obtain the second feature map includes: The number of channels of the reconstructed image is increased by using 3D convolution, and then input into the spatial attention mechanism, the number of channels of the input feature map is reduced by three 3D convolutions, and nonlinearity is introduced by applying an activation function, the generated spatial attention map is element-wise multiplied with the input feature map, and a weighted feature map is output; The weighted feature map is passed through three linear layers to generate query, key, and value matrices respectively; Divide the query, key, and value matrices into multiple heads, each head corresponds to an embedding dimension, and the embedding dimension of each head is: , represents the total embedding dimension in the 3D multi-head attention mechanism, Represents the total number of heads in the 3D multi-head attention mechanism; For each head, the attention score is calculated and applied to the value: ； in, is the query matrix; is the key matrix; Value matrix, L is the sequence length; N is the batch size; The outputs of all heads are stitched together, and the shape of the stitched image is restored; The residual connection is performed according to the following formula: ； in, is the result after adding the residual term, is the feature input to the residual module, is the residual term; Dynamic pooling is performed according to the following formula: ； in, Indicates that the feature map obtained after the pooling operation is in the batch , channel c, depth ,high ,width The value of the position, Represents the original input feature map in batch b, channel c, depth d, height h, width The value of the position, represents the average pooling factor, and Respectively represent the scaling ratio of the pooling window in height and width, represents the rounding down operation, i and j represent the row and column indices in the pooling window, respectively. , Represent the height and width of the output feature map respectively.

6. The method for super-resolution reconstruction of motion blurred images of power transmission line inspection according to claim 5, characterized in that: The step of increasing the width dimension of the second feature map, splicing the first feature map and the increased second feature map in the channel dimension to obtain a spliced ​​feature map, performing channel dimension reduction and feature fusion on the spliced ​​feature map through 1x1x1 convolution to obtain a fused feature map, and weighting the fused feature map to obtain a super-resolution image includes: Get the concatenated feature map according to the following formula: ； in, represents the concatenation operation along the dimension, M represents the first feature map of the output, represents the second feature map of the output, C represents the concatenated feature map, It means that we assume that we bring in values ​​to get the concatenated graph; Feature fusion is performed according to the following formula: ； in, represents the fused feature map, represents the fused image batch, Indicates the number of channels of the fused feature map, represents the depth of the fused feature map, Represents the height of the fused feature map, Represents the width of the fused feature map, Indicates the original number of channels, Indicates the number of additional channels, represents the feature map output by the concatenation operation, Represents a weight tensor used to transform the input features from channel c to the new , represents a 1x1x1 convolution, represents the bias term, Indicates the number of channels of the fused feature map; The weighting is done according to the following formula: ； in, represents the weighted eigenvalue map of the output, represents the first weighted operation result obtained in the global average pooling operation, represents element-wise multiplication, represents the spatial attention map of the output, represents the feature map after the fusion operation, represents the attention feature map, represents the output super-resolution image, Represents the activation function.

7. A transmission line inspection motion blurred image super-resolution reconstruction system, characterized in that: The system comprises: A fuzzy image acquisition module is used to acquire a fuzzy image of the insulator, perform time series processing on the fuzzy image to obtain a time series image, and perform time attention, time aggregation and time reconstruction operations on the time series image in sequence to obtain a reconstructed image; A first convolution processing module, used for sequentially performing multi-layer convolution processing, hole convolution processing, residual processing, attention mechanism processing, dimensionality reduction processing and dimensionality conversion processing on the reconstructed image to obtain a first feature map; A second convolution processing module is used to sequentially perform 3D convolution processing, spatial attention mechanism processing, tensor reshaping, permutation operation, 3D multi-head self-attention mechanism processing, shape recovery, high-dimensional convolution processing, residual connection, dynamic pooling, flattening operation, full connection operation, tensor expansion and dimension repetition processing on the reconstructed image to obtain a second feature map; A feature fusion module is used to enhance the width dimension of the second feature map, and to splice the first feature map and the enhanced second feature map in the channel dimension to obtain a spliced ​​feature map, and to perform channel dimension reduction and feature fusion on the spliced ​​feature map through a 1x1x1 convolution to obtain a fused feature map, and to weight the fused feature map to obtain a super-resolution image.

8. A storage medium, characterized in that: The storage medium stores one or more programs, which, when executed by the processor, implement the method for super-resolution reconstruction of motion blurred images of power transmission line inspection as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute the computer program stored in the memory, it implements the method for super-resolution reconstruction of motion blurred images of power transmission line inspection as described in any one of claims 1-6.

Images