Lightweight target detection network-based hidden danger target detection method and system along power transmission line

Abstract

The invention discloses a method and system for detecting hidden danger targets along a power transmission line based on a lightweight target detection network, and the method comprises the steps: obtaining a high-resolution remote sensing image along the power transmission line in a target region, carrying out the cutting, and constructing a hidden danger target data set along the power transmission line; constructing a lightweight target detection network, inputting the hidden danger target data set along the power transmission line into the lightweight target detection network, and training the lightweight target detection network; and utilizing the trained lightweight target detection network to predict hidden danger targets along the power transmission lines in other areas except the target area. According to the method, the hidden danger target along the power transmission line can be efficiently, conveniently and comprehensively detected, and the detection efficiency of the hidden danger target along the power transmission line is greatly improved.

Description

Technical Field

[0001] The present invention relates to a safety monitoring technology along a power transmission line, and in particular to a method, system, device and storage medium for detecting hidden danger targets along a power transmission line based on a lightweight target detection network. Background Art

[0002] With the rapid development of the power industry, the safety and stability of transmission lines, as an important part of the power grid, are directly related to the operating efficiency of the entire power grid and the safety of users' electricity use. However, since high-voltage transmission lines extend hundreds or even thousands of kilometers in all directions, the geographical environment is complex and diverse, and the climatic conditions are changeable, there are many targets with potential safety hazards along the transmission lines, such as color steel tiles, plastic greenhouses, ground films, etc. If encountering bad weather and other conditions, these safety hazard targets may be lifted up, thus affecting the safety of the transmission lines.

[0003] Traditionally, the safety of power transmission lines and the detection of potential hazards along the lines mainly rely on manual periodic inspections. Although this method can detect potential hazards to a certain extent, due to the limitation of the inspection cycle, it is often impossible to grasp the changes in the potential safety hazards along the lines in a timely manner. Moreover, in harsh climates and special geographical environments, the difficulty and danger of manual inspections are greatly increased. In addition, in the face of a wide range of transmission line networks, manual inspections are inefficient and costly, and it is difficult to meet the high requirements of modern power grids for safety and stability.

[0004] In order to overcome the limitations of traditional manual inspections, surveillance cameras deployed along the power grid are subsequently used to obtain surrounding images and conduct safety hazard target monitoring based on deep learning. Although it is more convenient than manual inspection methods, there are still some problems. These mainly include the small proportion of image targets obtained by surveillance cameras, poor image contrast, large amount of neural network calculations, and low calculation efficiency.

[0005] Therefore, how to efficiently, conveniently and comprehensively realize the detection of hidden danger targets along the transmission lines is a technical problem that needs to be solved urgently. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a method, system, device and storage medium for detecting hidden danger targets along transmission lines based on a lightweight target detection network, which can greatly improve the efficiency of detecting hidden danger targets along transmission lines.

[0007] Technical solution: A method for detecting hidden danger targets along a power transmission line based on a lightweight target detection network of the present invention comprises:

[0008] Obtain high-resolution remote sensing images along the transmission lines in the target area and crop them to construct a target dataset of hidden dangers along the transmission lines;

[0009] Construct a lightweight object detection network, input the hidden danger target dataset along the transmission line into the lightweight object detection network, and train the lightweight object detection network; wherein, the lightweight object detection network includes a convolutional module, a lightweight feature extraction module, an efficient convolutional module, a feature fusion module, and an attention prediction module connected in sequence; the convolutional module and the lightweight feature extraction module extract features from the images in the input hidden danger target dataset along the transmission line, and input the extracted feature maps into the efficient convolutional module; the efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain the global image information; after being processed by the convolutional module, the lightweight feature extraction module, and the efficient convolutional module, feature maps of different sizes are obtained and input into the feature fusion module; the feature fusion module fuses the features at different levels in the backbone network, and inputs the fused feature maps into the attention prediction module; the attention prediction module detects hidden danger targets along the transmission line with different proportions from the feature maps of different resolutions;

[0010] Use the trained lightweight object detection network to predict the hidden danger targets along the transmission line in other areas except the target area.

[0011] Further, the convolutional module and the lightweight feature extraction module extract features from the input image and input the extracted feature maps into the efficient convolutional module, including:

[0012] The lightweight feature extraction module uses a 1×1 convolutional kernel conv1 to integrate the channel information S of the input features, maps the input features to low-dimensional features with the number of channels as the input features, and uses a multi-branch 3×3 convolutional kernel conv3 to extract features from the low-dimensional features to obtain the feature map F. The calculation method is as follows:

[0013]

[0014] wherein, conv i is an i×i standard convolution;

[0015] In addition, batch normalization processing is also performed on the feature map, and the formula is as follows:

[0016]

[0017] wherein, x is the input feature map; y represents the output feature map; conv1 is a 1×1 standard convolution; conv2 is a 2×2 standard convolution; Γ i (x) is the projection of the input feature in the low-dimensional embedding and transformation; i is the number of feature maps; c is the maximum number of feature maps; BN is the batch normalization processing layer.

[0018] Furthermore, it is characterized in that the efficient convolution module processes and aggregates the feature information captured by the convolution module and the lightweight feature extraction module to obtain the global image information, including:

[0019] Based on the standard Transformer module, the efficient convolution module convolves the feature maps using a 1×1 convolution kernel. One branch extracts detailed information through an optimized multi-head self-attention mechanism layer and then connects with the other branch to output the processed and aggregated feature information, obtaining the global image information.

[0020] Furthermore, the optimized multi-head self-attention mechanism layer uses convolutional projection instead of linear projection. It projects the input two-dimensional tokens using a convolutional projection layer and flattens the projected tokens into one dimension for subsequent processing. The formula is as follows:

[0021]

[0022] Where, is the token input of the q / k / v matrix of the i-th layer; Conv is the standard convolution; x i is the unperturbed token before convolutional projection; s is the convolution kernel size.

[0023] Furthermore, the feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature maps into the attention prediction module, including:

[0024] The fusion feature module adds skip connections between the same-level input and output of the feature maps to fuse features at different levels and control the computational cost. The feature fusion formula is as follows:

[0025]

[0026] Where, is the intermediate feature at the i-th level on the top-down path; is the output feature at the i-th level on the bottom-up path; is the output feature at the (i - 1)-th level on the bottom-up path; is the input feature at the i-th level; is the input feature at the (i + 1)-th level; Conv is the standard 1×1 convolution; Resize is the downsampling or upsampling operation; w0, w1, w0′, w1′, w2′ represent weight parameters; ε represents a very small number to prevent the denominator from being 0;

[0027] Extra weights are assigned to each input to reflect the different contributions of input features at different resolutions. The weighting formula is as follows:

[0028]

[0029] Among them, X Output is the output feature; is the input feature; w i is the learnable weight, ε > 0.

[0030] Furthermore, the attention prediction module detects potential hazard targets along transmission lines in different proportions from feature maps of different resolutions, including:

[0031] The attention prediction module performs inference on the input feature map in both the channel and spatial dimensions, performs batch normalization and pixel normalization on the input feature map respectively, multiplies the attention map with the feature map, and adaptively obtains the attention area of the image. The formula of the attention prediction module is as follows:

[0032]

[0033] Among them, B in is the input feature map; B out is the output result feature map; γ is the trainable affine transformation parameter; β is the bias; μ β and σ β are the mean and variance of the mini-batch β respectively;

[0034] M s = sigmoid(W η (BN(F1)))

[0035] W γ = λ i / ∑ j＝0 λ j

[0036] Among them, sigmoid is the activation function; W γ is the coefficient; BN represents the batch normalization processing layer; F1 represents the input attention map; M s is the output feature; η is the channel ratio factor, λ i is the weight.

[0037] Furthermore, during the training of the lightweight object detection network, the mean squared error is used as the loss function to optimize the lightweight object detection network, including:

[0038] The loss function is the mean squared error MSE, and the formula for calculating MSE is as follows:

[0039]

[0040] Among them, MSE is the mean squared error; n is the number of images in the potential hazard target dataset along the transmission line; f(x i ) is the prediction value of the lightweight object detection network; y iIt is the true value of the hidden danger target along the transmission line.

[0041] Based on the same inventive concept, a hidden danger target detection system for transmission lines along the line based on a lightweight object detection network of the present invention includes:

[0042] A data acquisition module for acquiring high-resolution remote sensing images along the transmission line in the target area and cropping them to construct a hidden danger target dataset along the transmission line.

[0043] A model construction and training module for constructing a lightweight object detection network, inputting the hidden danger target dataset along the transmission line into the lightweight object detection network, and training the lightweight object detection network; wherein, the lightweight object detection network includes a convolutional module, a lightweight feature extraction module, an efficient convolutional module, a feature fusion module, and an attention prediction module connected in sequence; the convolutional module and the lightweight feature extraction module extract features from the images in the input hidden danger target dataset along the transmission line, and input the extracted feature maps into the efficient convolutional module; the efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain global image information; after being processed by the convolutional module, the lightweight feature extraction module, and the efficient convolutional module, feature maps of different sizes are obtained and input into the feature fusion module; the feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature maps into the attention prediction module; the attention prediction module detects hidden danger targets along the transmission line in different proportions from feature maps of different resolutions.

[0044] A prediction module for predicting hidden danger targets along the transmission line in other areas except the target area by using the trained lightweight object detection network.

[0045] Based on the same inventive concept, a hidden danger target detection device for transmission lines along the line based on a lightweight object detection network of the present invention includes a processor and a memory, wherein computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the above-mentioned hidden danger target detection method for transmission lines along the line based on a lightweight object detection network.

[0046] Based on the same inventive concept, a computer-readable storage medium of the present invention stores a computer program, and when the program is executed by a processor, it implements the steps of the above-mentioned hidden danger target detection method for transmission lines along the line based on a lightweight object detection network.

[0047] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are as follows:

[0048] The present invention proposes a lightweight feature extraction module, which reduces the computational complexity of feature extraction, solves the problem of slow network computing rate to a certain extent, and can quickly extract features of remote sensing images;

[0049] The present invention proposes an attention prediction module, which can adaptively obtain the attention area of an image and can achieve high-precision extraction of targets of different sizes;

[0050] The present invention proposes a hidden danger target detection scheme along a transmission line based on a lightweight object detection network, which can efficiently and quickly detect hidden danger targets along the transmission line and ensure the safety and stability of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is a schematic flowchart of a method for detecting hidden danger targets along a transmission line based on a lightweight object detection network disclosed in an embodiment of the present invention;

[0052] Figure 2 is a schematic structural diagram of a lightweight object detection network disclosed in an embodiment of the present invention;

[0053] Figure 3 is a schematic structural diagram of a system for detecting hidden danger targets along a transmission line based on a lightweight object detection network disclosed in an embodiment of the present invention;

[0054] Figure 4 is a schematic structural diagram of a device for detecting hidden danger targets along a transmission line based on a lightweight object detection network disclosed in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the specific beneficial effects described above, and it will be more clearly understood from the following detailed description that the above and other purposes that the present invention can achieve.

[0056] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in connection with the embodiments disclosed in the present invention can be implemented in hardware, software, or a combination of the two. Specifically, whether to execute in a hardware or software manner depends on the specific application and design and tree conditions of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0057] As used in this invention, the term "embodiment" means that the specific features, structures or characteristics described in connection with the embodiments can be included in at least one embodiment of the invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0058] Embodiment 1

[0059] Please refer to Figure 1 , Figure 1 which is a schematic flow chart of a hidden danger target detection method for transmission lines along the line based on a lightweight object detection network disclosed in an embodiment of the present invention. Among them, Figure 1 the described hidden danger target detection method for transmission lines along the line is applied to the power system, such as for safety monitoring along transmission lines, etc., which is not limited in the embodiments of the present invention. As Figure 1 shown, the hidden danger target detection method for transmission lines along the line based on the lightweight object detection network may include the following operations:

[0060] S1. Obtain high-resolution remote sensing images along the transmission lines in the target area and crop them to construct a hidden danger target dataset for transmission lines along the line.

[0061] In this embodiment, the obtained high-resolution remote sensing images along the transmission lines in the target area are cropped into images of 640×640 size for constructing a hidden danger target dataset for transmission lines along the line.

[0062] S2. Construct a lightweight object detection network, input the hidden danger target dataset for transmission lines along the line into the lightweight object detection network, and train the lightweight object detection network.

[0063] Among them, as Figure 2 shown, the lightweight object detection network includes a convolutional module, a lightweight feature extraction module, an efficient convolutional module, a feature fusion module, and an attention prediction module connected in sequence; the input remote sensing images along the transmission lines are subjected to feature extraction on the images in the hidden danger target dataset for transmission lines along the line by the convolutional module and the lightweight feature extraction module; the extracted feature maps are input into the efficient convolutional module, and the efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain global image information; after the processing of the above two steps, feature maps of different sizes are obtained and input into the feature fusion module to fuse the features at different levels in the backbone network; finally, the fused feature maps are input into the attention prediction module to detect hidden danger targets for transmission lines along the line with different proportions from the feature maps of different resolutions.

[0064] In this embodiment, the convolution module and the lightweight feature extraction module extract features from the input image, and input the extracted feature maps into the efficient convolution module. Specifically as follows:

[0065] The lightweight feature extraction module first integrates the channel information S of the input features using a 1×1 convolution kernel conv1, maps the input features to low-dimensional features with the number of channels as the input features, and then uses a multi-branch 3×3 convolution kernel conv3 to extract features from the low-dimensional features to obtain the feature map F. The formula for calculating F is as follows:

[0066]

[0067] Among them, conv i is the i×i standard convolution.

[0068] In addition, batch normalization is also performed on the feature map to prevent gradient loss. The formula is as follows:

[0069]

[0070] Among them, x is the input feature map; y represents the output feature map; conv1 represents the 1×1 standard convolution; conv2 represents the 2×2 standard convolution; Γ i (x) represents the projection of the input feature in the low-dimensional embedding and transformation; i represents the number of feature maps; c represents the maximum number of feature maps; BN is the batch normalization layer.

[0071] In this embodiment, the efficient convolution module processes and aggregates the feature information captured by the convolution module and the lightweight feature extraction module to obtain the global image information. Specifically as follows:

[0072] The efficient convolution module is based on the standard Transformer module, convolves the feature maps using a 1×1 convolution kernel respectively. One branch extracts more detailed information through the optimized multi-head self-attention mechanism layer and then connects with the other branch, and outputs the processed and aggregated feature information to obtain the global image information.

[0073] Among them, the optimized multi-head self-attention mechanism layer uses convolution projection instead of linear projection, reduces the number of parameters required by the module, and reduces the computational complexity. The convolution projection layer is used to project the input two-dimensional tokens, and the projected tokens are flattened into one dimension for subsequent processing. The formula is as follows:

[0074]

[0075] Among them, is the token input of the q / k / v matrix of the i-th layer; Conv is the standard convolution; x i is the unperturbed token before convolution projection; s is the convolution kernel size.

[0076] In this embodiment, the feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature map into the attention prediction module. Specifically as follows:

[0077] The fused feature module adds skip connections between the same-level input and output of the feature map to fuse features at different levels and control the computational cost. The feature fusion formula is as follows:

[0078]

[0079] Where, is the intermediate feature at the i-th level on the top-down path; is the output feature at the i-th level on the bottom-up path; is the output feature at the (i - 1)-th level on the bottom-up path; is the input feature at the i-th level; is the input feature at the (i + 1)-th level; Conv is a standard 1×1 convolution; Resize is a downsampling or upsampling operation; w0, w1, w0′, w1′, w2′ represent weight parameters; ε represents a very small number to prevent the denominator from being 0;

[0080] An additional weight is assigned to each input to reflect the different contributions of input features at different resolutions. The weighting formula is as follows:

[0081]

[0082] Where, X Output represents the output feature; is the input feature; w i is the learnable weight, ε > 0.

[0083] In this embodiment, the attention prediction module detects potential hazard targets along transmission lines at different scales from feature maps of different resolutions. Specifically as follows:

[0084] The attention prediction module performs inference on the input feature map in both the channel and spatial dimensions, performs batch normalization and pixel normalization on the input feature map respectively, multiplies the attention map with the feature map, and adaptively obtains the attention region of the image. The formula of the attention prediction module is as follows:

[0085]

[0086] Where, B in represents the input feature map, B out represents the output result feature map, γ is the trainable affine transformation parameter, β is the bias, μ β and σ β are the mean and variance of the mini-batch β respectively.

[0087] M s = sigmoid(W η (BN(F1)))

[0088] W γ = λ i / ∑ j＝0 λ j

[0089] Among them, sigmoid is the activation function; W γ is the coefficient; BN represents the batch normalization processing layer; F1 represents the input attention map; M s is the output feature; η is the channel scale factor, and λ i is the weight.

[0090] During the training of the lightweight object detection network, the mean square error is used as the loss function to optimize the lightweight object detection network, as follows:

[0091] The loss function is the mean square error MSE, and the formula for calculating MSE is as follows:

[0092]

[0093] Among them, MSE is the mean square error; n is the number of images in the hidden danger target dataset along the transmission line; f(x i ) is the prediction value of the lightweight object detection network; y i is the true value of the hidden danger target along the transmission line.

[0094] S3. Use the trained lightweight object detection network to predict the hidden danger targets along the transmission line in other areas except the target area.

[0095] Embodiment 2

[0096] Please refer to Figure 3 , Figure 3 which is a schematic structural diagram of a hidden danger target detection system along a transmission line based on a lightweight object detection network disclosed in an embodiment of the present invention. This system can realize the safety monitoring along the transmission line and specifically includes:

[0097] A data acquisition module, which is used to acquire high-resolution remote sensing images along the transmission line in the target area and perform cropping to construct a hidden danger target dataset along the transmission line;

[0098] The model construction and training module is used to construct a lightweight object detection network, input the hidden danger target dataset along the transmission line into the lightweight object detection network, and train the lightweight object detection network. Among them, the lightweight object detection network includes a convolution module, a lightweight feature extraction module, an efficient convolution module, a feature fusion module, and an attention prediction module connected in sequence. The lightweight feature extraction module extracts features and downsamples the input image. The efficient convolution module processes and aggregates the feature information captured by the convolution module to obtain global image information. The feature fusion module fuses features at different levels in the backbone network. The attention prediction module detects hidden danger targets along the transmission line in different proportions in feature maps with different resolutions.

[0099] The prediction module is used to predict hidden danger targets along the transmission line in other areas except the target area by using the trained lightweight object detection network.

[0100] In an optional embodiment, the hidden danger target detection along the transmission line based on the lightweight object detection network includes: a) obtaining and cropping the high-resolution remote sensing image along the transmission line in the target area to construct a hidden danger target dataset along the transmission line; b) constructing a lightweight object detection network, inputting the hidden danger target dataset along the transmission line into the lightweight object detection network, and training the lightweight object detection network; c) predicting hidden danger targets along the transmission line in other areas by using the trained lightweight object detection network.

[0101] Embodiment 3

[0102] Please refer to Figure 4 , Figure 4 which is a schematic structural diagram of a hidden danger target detection device along the transmission line based on a lightweight object detection network disclosed in an embodiment of the present invention. Among them, Figure 4 the described device can be applied to the power system, such as for safety monitoring along the transmission line, etc., and the embodiments of the present invention do not make limitations.

[0103] As Figure 4 shown, the device may include a processor and a memory. Computer instructions are stored in the memory. The processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method as described in the above embodiment and can achieve the same technical effects as the above method.

[0104] The memory may include computer system-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the memory may be used for reading from and writing to a non-removable, non-volatile magnetic medium (commonly referred to as a "hard disk drive"). A program / utility having a set (at least one) of program modules may be stored in, for example, the memory. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, and an implementation of a network environment may be included in each or some combination of these examples. The program modules generally execute the functions and / or methods in the embodiments described in the present invention.

[0105] The processor executes various functional applications and data processing by running the programs stored in the memory, such as implementing the method provided in Embodiment 1 of the present invention.

[0106] Embodiment 4

[0107] Embodiment 4 of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the steps of the method described in the above embodiments and can achieve the same technical effects as the above method.

[0108] The computer storage medium of the embodiments of the present invention may adopt any combination of one or more computer-readable media. The computer-readable media may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection having 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 this document, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0109] A computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0110] The program code contained on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0111] The computer program code for performing the operations of the present invention may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., connected through the Internet using an Internet service provider).

[0112] Of course, a storage medium containing computer-executable instructions provided by an embodiment of the present invention, the computer-executable instructions are not limited to the method operations as described above, and may also execute related operations in the methods provided by any embodiment of the present invention.

[0113] The specific embodiments described above further elaborate on the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A hidden danger target detection method along a transmission line based on a lightweight object detection network, characterized in that, Including: Obtain high-resolution remote sensing images along the transmission lines in the target area and perform cropping to construct a hidden danger target dataset along the transmission lines. Construct a lightweight object detection network, input the hidden danger target dataset along the transmission lines into the lightweight object detection network, and train the lightweight object detection network. Among them, the lightweight object detection network includes a convolutional module, a lightweight feature extraction module, an efficient convolutional module, a feature fusion module, and an attention prediction module connected in sequence. The convolutional module and the lightweight feature extraction module extract features from the images in the input hidden danger target dataset along the transmission lines and input the extracted feature maps into the efficient convolutional module. The efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain the global image information. After being processed by the convolutional module, the lightweight feature extraction module, and the efficient convolutional module, feature maps of different sizes are obtained and input into the feature fusion module. The feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature maps into the attention prediction module. The attention prediction module detects hidden danger targets along the transmission lines with different proportions from feature maps of different resolutions. Use the trained lightweight object detection network to predict hidden danger targets along the transmission lines in other areas except the target area.

2. The hidden danger target detection method along the transmission line based on the lightweight object detection network according to claim 1, characterized in that, The convolutional module and the lightweight feature extraction module extract features from the input image and input the extracted feature maps into the efficient convolutional module, including: The lightweight feature extraction module uses a 1×1 convolutional kernel conv1 to integrate the channel information S of the input feature, maps the input feature to a low-dimensional feature with the number of channels as the input feature, and uses a multi-branch 3×3 convolutional kernel conv3 to extract features from the low-dimensional feature to obtain a feature map F. The calculation method is as follows: Among them, conv i is an i×i standard convolution; In addition, batch normalization is also performed on the feature map. The formula is as follows: Among them, x is the input feature map; y represents the output feature map; conv1 is a 1×1 standard convolution; conv2 is a 2×2 standard convolution; Γ i (x) is the projection of the input feature in the low-dimensional embedding and transformation; i is the number of feature maps; c is the maximum number of feature maps; BN is the batch normalization processing layer.

3. The hidden danger target detection method along a transmission line based on a lightweight object detection network according to claim 1, wherein The efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain the global image information, including: The efficient convolutional module is based on a standard Transformer module. The feature map is convolved using a 1×1 convolutional kernel respectively. One branch extracts detailed information through an optimized multi-head self-attention mechanism layer and then connects with the other branch to output the processed and aggregated feature information to obtain the global image information.

4. The hidden danger target detection method along the transmission line based on the lightweight object detection network according to claim 3, wherein The optimized multi-head self-attention mechanism layer uses a convolutional projection instead of a linear projection, projects the input two-dimensional tokens using a convolutional projection layer, and flattens the projected tokens into one dimension for subsequent processing. The formula is as follows: Among them, is the token input of the q / k / v matrix of the i-th layer; Conv is a standard convolution; x i is the unperturbed token before convolution projection; s is the convolution kernel size.

5. The hidden danger target detection method along a transmission line based on a lightweight object detection network according to claim 1, wherein The feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature maps into the attention prediction module, including: The fusion feature module adds skip connections between the same-level input and output of the feature map to fuse features at different levels and control the calculation cost. The feature fusion formula is as follows: Among them, is the intermediate feature at the i-th level on the top-down path; is the output feature at the i-th level on the bottom-up path; is the output feature at the (i - 1)-th level on the bottom-up path; is the input feature at the i-th level; is the input feature at the (i + 1)-th level; Conv is a standard 1×1 convolution; Resize is a downsampling or upsampling operation; w0, w1, w′0, w′1, w′2 represent weight parameters; ε represents a very small number to prevent the denominator from being 0; Assign additional weights to each input to reflect the different contributions of input features with different resolutions. The weighting formula is as follows: Among them, X Output is the output feature; is the input feature; w i is the learnable weight, ε > 0.

6. The hidden danger target detection method along the transmission line based on the lightweight object detection network according to claim 1, characterized in that The attention prediction module detects potential hazard targets along transmission lines with different scales from feature maps of different resolutions, including: The attention prediction module performs reasoning on the input feature map in both the channel and spatial dimensions, performs batch normalization and pixel normalization on the input feature map respectively, multiplies the attention map with the feature map, and adaptively obtains the attention area of the image. The formula of the attention prediction module is as follows: Among them, B in is the input feature map; B out is the output result feature map; γ is the trainable affine transformation parameter; β is the bias; μ β and σ β are the mean and variance of the mini-batch β, respectively; M s = sigmoid(W η (BN(F1))) W γ = λ i / ∑ j＝0 λ j Among them, sigmoid is the activation function; W γ is the coefficient; BN represents the batch normalization processing layer; F1 represents the input attention map; M s is the output feature; η is the channel ratio factor; λ i is the weight.

7. The hidden danger target detection method along a transmission line based on a lightweight object detection network according to claim 1, wherein During the training process of the lightweight object detection network, the mean square error is used as the loss function to optimize the lightweight object detection network, including: The loss function is the mean square error MSE, and the formula for calculating MSE is as follows: Among them, MSE is the mean square error; n is the number of images in the hidden danger target dataset along the transmission line; f(x i ) is the predicted value of the lightweight object detection network; y i is the true value of the hidden danger target along the transmission line.

8. A hidden danger target detection system along a transmission line based on a lightweight object detection network, characterized in that, Including: The data acquisition module is used to obtain high-resolution remote sensing images along the transmission lines in the target area and perform cropping to construct a dataset of potential hazard targets along the transmission lines. The model construction and training module is used to construct a lightweight object detection network, input the dataset of potential hazard targets along the transmission lines into the lightweight object detection network, and train the lightweight object detection network. Among them, the lightweight object detection network includes a convolutional module, a lightweight feature extraction module, an efficient convolutional module, a feature fusion module, and an attention prediction module connected in sequence. The convolutional module and the lightweight feature extraction module extract features from the images in the input dataset of potential hazard targets along the transmission lines, and input the extracted feature maps into the efficient convolutional module. The efficient convolutional module processes and aggregates the feature information captured by the convolutional module and the lightweight feature extraction module to obtain the global image information. After being processed by the convolutional module, the lightweight feature extraction module, and the efficient convolutional module, feature maps of different sizes are obtained and input into the feature fusion module. The feature fusion module fuses the features at different levels in the backbone network and inputs the fused feature maps into the attention prediction module. The attention prediction module detects potential hazard targets along transmission lines with different scales from feature maps of different resolutions. The prediction module is used to predict potential hazard targets along transmission lines in other areas except the target area by using the trained lightweight object detection network.

9. A hidden danger target detection device along a transmission line based on a lightweight object detection network, characterized in that, It includes a processor and a memory. Computer instructions are stored in the memory, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method for detecting potential hazard targets along transmission lines based on a lightweight object detection network according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the program is executed by the processor, it implements the steps of the method for detecting potential hazard targets along transmission lines based on a lightweight object detection network according to any one of claims 1 to 7.

Images