Substation bird repelling control method and system based on target detection

Abstract

The invention is suitable for the field of target detection repelling, and provides a transformer substation bird repelling control method and system based on target detection, and the method comprises the steps: carrying out the preprocessing of a to-be-detected image through an image enhancement module, extracting image features through a backbone network stacking convolution layer, carrying out the feature fusion through a neck structure, and forming a multi-scale feature; outputting an edge feature and a category result of a target object in the to-be-detected image through a detection head module; an alarm signal is generated in response to a driving instruction generated based on the edge features and the category result, the alarm signal is transmitted to an intelligent control system through a communication module, and the intelligent control system issues an action instruction to a gas pipeline electromagnetic valve of an air compression injection system at an alarm point through a DO module after operation of a CPU module. And injecting compressed gas to repel the target object. According to the invention, birds in the transformer substation area can be repelled, and safe operation of the transformer substation can be effectively protected.

Description

Technical Field

[0001] The present invention belongs to the technical field of target detection and expulsion, and in particular relates to a substation bird expulsion control method and system based on target detection. Background Art

[0002] Substation switchyards are equipped with important primary electrical equipment such as circuit breakers, disconnectors, busbars, and current transformers. These equipment are installed outdoors. Since substations are generally located in suburban areas with good surrounding vegetation or close to farmland, sparrows, turtledoves and other birds are easily gathered there. Bird nests are built at the base of the circuit breaker insulation pillars, the disconnector contacts and bases, the current transformer joints, and the busbar connectors. Birds move frequently, and they hold long branches and metal wires in their mouths, which can easily cause equipment grounding short circuit accidents. In addition, bird droppings are highly corrosive and can corrode and oxidize the current-carrying components of primary electrical equipment, causing the equipment to increase in temperature, decrease in current-carrying capacity, and corrode and damage, thereby reducing power supply reliability. Therefore, it is necessary to drive away birds from substations.

[0003] Bird repelling requires target detection. In the field of target detection, deep learning technology, especially the YOLO series of algorithms, is favored due to its fast and accurate performance.

[0004] Traditional YOLO-based object detection algorithms still face some challenges when dealing with small-sized objects such as birds. The main problems include: Bird objects are small in size and do not occupy a large proportion of the image, resulting in less feature information and difficulty in effectively capturing them. In specific environments such as substations, factors such as lighting conditions and brightness changes may interfere with the detection performance of the algorithm. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a substation bird-repelling control method and system based on target detection, aiming to solve the technical problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] An embodiment of the present invention provides a substation bird-repelling control method based on target detection, comprising:

[0008] Constructing a target detection model based on the YOLOv5 network architecture, the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module;

[0009] Obtain the image to be detected in the substation area, use the image enhancement module to preprocess the image to be detected, use the backbone network to stack convolutional layers to extract image features, use the neck structure to fuse features to form multi-scale features, and output the edge features and category results of the target object in the image to be detected through the detection head module;

[0010] In response to the expulsion instruction generated based on the edge features and category results, an alarm signal is generated, and the alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module, and compressed gas is injected to drive away the target object, which is a bird.

[0011] Furthermore, the step of preprocessing the image to be detected by using the image enhancement module includes:

[0012] Build an image enhancement module, which includes a lightweight enhancement parameter prediction network and an image enhancement iteration unit;

[0013] Using the image to be detected as the input of the lightweight enhanced parameter prediction network to obtain parameter prediction of the curve parameter matrix;

[0014] The matrix is ​​divided in the channel dimension to obtain a plurality of light enhancement parameters, and the plurality of light enhancement parameters are respectively used as iteration parameters of each image enhancement iteration unit;

[0015] The input of each iteration unit is used as the intermediate representation image output by the previous iteration unit, and the image is cyclically enhanced, and the final output is used as the enhanced image.

[0016] Furthermore, the lightweight enhanced parameter prediction network includes 7 convolutional layers; among them, the first 6 convolutional layers are 16 or 32 convolutional kernels with a size of 3×3 and a stride of 1, and ReLU activation is adopted; the last convolutional layer is T×3 convolutional kernels with a size of 3×3 and a stride of 1, and Tanh activation is adopted.

[0017] Furthermore, in the step of performing image cyclic enhancement, a non-reference loss function is used to analyze the pixel spatial differences of the image itself and the correlation between the original image and the enhanced version of the image to calculate the loss; wherein the non-reference loss function includes spatial consistency loss, exposure control loss, color constancy loss and illumination smoothing loss.

[0018] Furthermore, in the step of extracting image features by stacking convolutional layers of a backbone network, the backbone network is a DarkNet53 network using CSP, and a coordinate attention mechanism is added to the backbone network to encode coordinate information in the horizontal and vertical directions and embed it into the channel of the backbone network, so that the backbone network pays more attention to small target features when performing target detection and reduces the interference of background noise.

[0019] Furthermore, the step of adding a coordinate attention mechanism to the backbone network includes:

[0020] Two spatial range pooling kernels (H, 1) and (1, W) are used to encode each channel of the feature input and aggregate them in two directions to form a pair of position-aware feature maps, where the output of the cth channel in height and width is expressed as:

[0021]

[0022] In the formula, x c (h,i) represents the value of the hth row and ith column of channel c; x c (j,w) represents the value of the wth column and jth row of channel c; Represents the output along the horizontal direction; Represents the output along the vertical direction;

[0023] Coordinate attention generation first embeds the coordinate information into the extracted feature map connection, and then encodes the spatial information in the horizontal and vertical directions through calculation, and obtains the intermediate feature map k, which is expressed as:

[0024] k=δ(F1([z h ,z w ]))

[0025] In the formula, [z h ,z w ] means to set z h and z w Splicing along the spatial channel; δ represents a nonlinear activation function; F1 represents a 1×1 convolution transformation function;

[0026] Split the intermediate feature map k along the spatial dimension to obtain two independent tensors k h and k w , after convolution transformation and activation function, the two tensors have the same number of channels as the input X, and finally we get g h and g w As attention weights, it is expressed as follows:

[0027] g h =σ(K h (k h ))

[0028] g w =σ(K w (k w ))

[0029] In the formula, K h and K w All are 1×1 convolution transformations; σ represents the sigmoid activation function.

[0030] Furthermore, the backbone network adopts the GIOU function as the bounding box loss function and uses non-maximum suppression to filter the target objects in the image; wherein the GIOU function is expressed as:

[0031]

[0032] Where C represents the area of ​​the minimum circumscribed rectangle of the predicted box and the true box; U represents the union area of ​​the predicted box and the true box; the value range of GIoU is [-1,1]. When the predicted box and the true box completely overlap, GIoU takes the maximum value of 1. When they are completely disjoint and far away, GIoU takes the value of -1. IoU is an indicator used to measure the degree of overlap between the predicted box and the true box.

[0033] Furthermore, the loss function used by the backbone network is the GIOU function that introduces power transformation, which is expressed as:

[0034]

[0035] In the formula, parameter α1>0, parameter α2>0, b represents the prediction box, b gt represents the real frame; According to b and b gt The regularization term for the calculation.

[0036] Furthermore, the air compression injection system comprises:

[0037] A screw air compressor with an exhaust pressure of 0.8-0.85Mpa;

[0038] A 3m 3 The normal pressure of the gas tank is 0.7Mpa. The gas tank is equipped with an electric contact pressure gauge. When the pressure of the gas tank drops to 0.6Mpa, the screw air compressor is automatically started. When the pressure is increased to 0.7Mpa, the screw air compressor is automatically stopped. The electric contact pressure gauge has a pressure analog module to transmit the pressure of the gas tank to the intelligent control system in real time.

[0039] A carbon steel main steam pipeline with a diameter of 60mm and 20 insulated branch pipelines with a diameter of 20mm. Each branch pipeline is equipped with an electromagnetic valve, and the valve switch is controlled by the control system DO module. A 90-degree elbow is installed at the outlet of the branch pipeline to make the compressed gas horizontal to the substation equipment point.

[0040] Another embodiment of the present invention provides a substation bird-repelling control system based on target detection, comprising:

[0041] A model construction module is used to construct a target detection model based on the YOLOv5 network architecture, wherein the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module;

[0042] The target detection module is used to obtain the image to be detected in the substation area, pre-process the image to be detected using the image enhancement module, extract image features using the backbone network stacked convolutional layers, fuse features using the neck structure to form multi-scale features, and output the edge features and category results of the target object in the image to be detected through the detection head module;

[0043] The target driving away module is used to generate an alarm signal in response to the driving away instruction generated based on the edge features and category results. The alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module to inject compressed gas to drive away the target object, which is a bird.

[0044] In summary, the beneficial effects of the substation bird-repelling control method and system based on target detection of the present invention are:

[0045] First, by constructing a target detection model based on the YOLOv5 network architecture, the present invention can achieve rapid and accurate detection of target objects such as birds in the substation area; the image enhancement module can improve the quality of the image to be detected, which helps the target detection model to quickly and accurately detect the target object in the image, and solves the problem that low-light images in the substation scene affect the speed and accuracy of bird target detection;

[0046] Second, by processing the image to be detected in the substation area, the present invention can extract the feature information in the image and fuse it with multi-scale features, thereby improving the accuracy and robustness of target detection. The edge features and category results of the target object finally output can provide a basis for subsequent expulsion operations;

[0047] Third, the present invention generates expulsion instructions based on target detection results, and can control the operation of the air compression injection system through an intelligent control system to achieve the expulsion of target objects such as birds. The control method based on target detection is real-time and accurate, and can effectively protect the safe operation of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0049] Figure 1 The present invention is a flowchart of a method for controlling bird-repelling in a substation based on target detection;

[0050] Figure 2A sub-flow chart of a substation bird-repelling control method based on target detection according to the present invention;

[0051] Figure 3 This is a structural block diagram of a substation bird-repelling control system based on target detection according to the present invention;

[0052] Figure 4 The present invention provides a structural block diagram of a computer device. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0054] In the field of target detection, deep learning technology, especially the YOLO series of algorithms, is favored for its fast and accurate performance. However, these algorithms still face some challenges when dealing with small-sized targets such as birds. The main problems include: bird targets are small in size and do not account for a large proportion of the image, resulting in less feature information and difficulty in being effectively captured. In specific environments such as substations, factors such as lighting conditions and brightness changes may interfere with the detection performance of the algorithm.

[0055] To solve the above problems, the present invention provides a substation bird-repelling control method and system based on target detection. The present invention constructs a target detection model based on the YOLOv5 network architecture, and the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module; an image to be detected in the substation area is acquired, the image to be detected is preprocessed by the image enhancement module, the image features are extracted by stacking convolution layers of the backbone network, the neck structure is used to perform feature fusion to form multi-scale features, and the edge features and category results of the target object in the image to be detected are output through the detection head module; in response to a driving instruction generated based on the edge features and category results, an alarm signal is generated, and the alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module, and the compressed gas is injected to drive away the target object.

[0056] The following is a detailed description of the specific implementation of a substation bird-repelling control method and system based on target detection of the present invention in conjunction with specific embodiments.

[0057] Please refer to Figure 1 In one embodiment of the present invention, a substation bird-repelling control method based on target detection is provided. The control method of the present invention comprises the following steps:

[0058] S101: constructing a target detection model based on the YOLOv5 network architecture, wherein the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module;

[0059] In the object detection model based on YOLOv5 in the embodiment of the present invention, image enhancement is an important part of data preprocessing, which enhances the generalization ability of the object detection model by applying various image transformations. Commonly used image enhancement techniques include random flipping, rotation, scaling, cropping, etc.

[0060] In an embodiment of the present invention, an image enhancement module can also be used to perform image enhancement processing on images to be detected that are obtained in poor lighting environments, which helps to achieve rapid and accurate detection of target objects such as birds in the substation area; wherein the image enhancement module can improve the quality of the images to be detected, and helps the target detection model to quickly and accurately detect the target objects in the image, thereby solving the problem that low-light images in substation scenarios affect the speed and accuracy of target detection of birds.

[0061] Furthermore, the neck structure of the embodiment of the present invention is located between the backbone network and the detection head module, and its main function is to integrate feature maps of different levels to improve the model's detection capability for targets of different scales. In the YOLOv5 target detection model of the present invention, the neck network adopts the FPN (Feature Pyramid Network) structure to generate multi-scale feature maps through top-down and bottom-up feature fusion.

[0062] For further information, please refer to Figure 1 The substation bird-repelling control method based on target detection of the present invention also includes:

[0063] S102: Obtain an image to be detected in the substation area, use an image enhancement module to preprocess the image to be detected, use a backbone network stacked convolution layer to extract image features, use the neck structure to perform feature fusion to form multi-scale features, and output edge features and category results of the target object in the image to be detected through a detection head module;

[0064] The present invention can process the image to be detected in the substation area to extract the feature information in the image and fuse it with multi-scale features, thereby improving the accuracy and robustness of target detection. The edge features and category results of the target object finally output can provide a basis for subsequent expulsion operations.

[0065] S103: In response to the expulsion instruction generated based on the edge features and category results, an alarm signal is generated, and the alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module, and compressed gas is injected to drive away the target object, which is a bird.

[0066] The structure of the air compression injection system of the embodiment of the present invention is as follows:

[0067] A screw air compressor with an exhaust pressure of 0.8-0.85Mpa;

[0068] A 3m 3 The normal pressure of the gas tank is 0.7Mpa. The gas tank is equipped with an electric contact pressure gauge. When the pressure of the gas tank drops to 0.6Mpa, the screw air compressor is automatically started. When the pressure is increased to 0.7Mpa, the screw air compressor is automatically stopped. The electric contact pressure gauge has a pressure analog module to transmit the pressure of the gas tank to the intelligent control system in real time.

[0069] A carbon steel main steam pipeline with a diameter of 60mm, 20 insulated branch pipes with a diameter of 20mm (bird repellent pipes), each branch pipe is installed with an electromagnetic valve, the valve switch is controlled by the control system DO module, and a 90-degree elbow is installed at the outlet of the branch pipe to make the compressed gas and the substation equipment point in a horizontal direction.

[0070] Preferably, in step S103 of the embodiment of the present invention, the following control method can be used to eject the compressed gas away from the target object:

[0071] Manual mode: You can manually issue bird-repelling orders to 20 bird-repelling pipelines in the intelligent control system, open a single management solenoid valve to discharge compressed gas, and drive away birds;

[0072] Timing mode: You can set the time when birds are most active (such as 06:00-07:00 in the morning and 18:00-19:00 in the evening), so that the intelligent control system automatically issues instructions, and 20 bird-repelling pipelines open the solenoid valves at the same time to discharge compressed gas to drive away the birds;

[0073] Intelligent mode: that is, the target detection model of the present invention is used to detect birds. Specifically, when the intelligent industrial camera captures the activity of birds at the equipment identification point, an alarm is automatically given, and the signal is transmitted to the intelligent control system through the communication module. After the control system is operated by the CPU module, the bird-repelling command is sent to the gas pipeline solenoid valve at the alarm point through the D0 module to discharge compressed gas and drive away the birds;

[0074] Comprehensive mode: Combine the timing mode and intelligent mode to form a comprehensive bird-repelling mode with timing full measurement points and intelligent early warning single measurement points.

[0075] For further information, please refer to Figure 2 In a preferred embodiment of the present invention, the step of preprocessing the image to be detected by using the image enhancement module includes:

[0076] S201: Building an image enhancement module, the image enhancement module includes a lightweight enhancement parameter prediction network and an image enhancement iteration unit;

[0077] S202: using the image to be detected as the input of the lightweight enhanced parameter prediction network to obtain parameter prediction of the curve parameter matrix;

[0078] Among them, in an embodiment of the present invention, the lightweight enhanced parameter prediction network includes 7 convolutional layers; among them, the first 6 convolutional layers are 16 or 32 convolutional kernels with a size of 3×3 and a stride of 1, and ReLU activation is adopted; the last convolutional layer is T×3 3×3 convolutional kernels with a stride of 1, and Tanh activation is adopted.

[0079] In addition, in step S202, in order to prevent the lightweight enhancement parameter prediction network from destroying the smooth relationship between adjacent pixels in the image, the lightweight enhancement parameter prediction network of the embodiment of the present disclosure does not contain downsampling and batch normalization layers. In this way, the global information and spatial correlation of the image can be retained to a greater extent.

[0080] S203: Divide the matrix in the channel dimension to obtain a plurality of light enhancement parameters, and use the plurality of light enhancement parameters as iteration parameters of each image enhancement iteration unit respectively;

[0081] S204: The input of each iteration unit is used as the intermediate representation image output by the previous iteration unit, and image cyclic enhancement is performed, and the final output is used as the enhanced image.

[0082] The present invention uses multiple iterations of enhancement and uses the output of the previous light enhancement curve as the input of the next iteration. Multiple iterations enable the model to adjust the image within a wider dynamic range.

[0083] Furthermore, in the step of performing image cyclic enhancement, a non-reference loss function is used to analyze the pixel spatial differences of the image itself and the correlation between the original image and the enhanced version of the image to calculate the loss; wherein the non-reference loss function includes spatial consistency loss, exposure control loss, color constancy loss and illumination smoothing loss.

[0084] The spatial consistency loss is used to keep the brightness difference between adjacent pixels in the enhanced image similar to that of the input image, ensure that adjacent regions in the enhanced image maintain consistent contrast and edge information, and avoid excessive stretching or reduction of contrast during the enhancement process. This loss function encourages the enhanced image to remain consistent in local areas by calculating the brightness difference between neighboring pixels.

[0085] The exposure control loss is used to control the exposure of the image to ensure that the image is not overexposed or underexposed. It limits the overly bright or dark areas by comparing the average brightness of the local area with the ideal exposure value. In the specific implementation, the exposure value is set to 0.6, which means that the brightness in the image should be close to the medium gray level;

[0086] The color constancy loss means that the average of each color channel in the image should tend to gray (that is, the average values ​​of the red, green, and blue channels should be similar). The color constancy loss is used to correct the color distortion in the enhanced image to ensure that the color of the image looks natural and the color relationship of the three channels remains consistent. It reduces color shift by comparing the average value difference of the red, green, and blue channels;

[0087] The illumination smoothness loss (Illumination Smoothness Loss) limits the changes in the illumination curve parameters to keep the brightness changes of the enhanced image smooth, aiming to avoid sudden changes in illumination in the enhanced image and ensure smooth transition of illumination, thereby avoiding unnatural lighting effects. The loss calculates the gradient of the illumination curve parameters to ensure that the illumination adjustment of adjacent pixels maintains a smooth transition.

[0088] The present invention uses the above-mentioned loss function to ensure that the model can still generate high-quality enhanced images without a reference image, making the image look natural and rich in details, which helps to further detect the target object in the image using the target detection model.

[0089] In the image enhancement process of the present invention, the image quality is gradually optimized in an iterative manner, and each iteration is fine-tuned based on the result of the previous iteration, thereby gradually improving the image quality; the iterative process can be iterated multiple times according to actual needs until a satisfactory effect is achieved. This flexibility allows the enhancement process to be adjusted according to specific needs; in the iterative enhancement process, the detailed information of the image is retained, avoiding information loss caused by excessive enhancement, which is particularly important for image analysis and feature extraction; the input of each iteration unit is used as the intermediate representation image output by the previous iteration unit, and the image is cyclically enhanced, and the final output is used as the enhanced image. This cyclic enhancement mechanism can further improve the image quality.

[0090] Furthermore, in the step of extracting image features by stacking convolutional layers of a backbone network, the backbone network is a DarkNet53 network using CSP, and a coordinate attention mechanism is added to the backbone network to encode coordinate information in the horizontal and vertical directions and embed it into the channel of the backbone network, so that the backbone network pays more attention to small target features when performing target detection and reduces the interference of background noise.

[0091] Specifically, in an embodiment of the present invention, the step of adding a coordinate attention mechanism to the backbone network includes: encoding each channel of the feature input using two spatial range pooling kernels (H, 1) and (1, W), and aggregating them in two directions to form a pair of position-aware feature maps, wherein the output of the cth channel in height and width is expressed as:

[0092]

[0093] In the formula, x c (h,i) represents the value of the hth row and ith column of channel c; x c (j,w) represents the value of the wth column and jth row of channel c; Represents the output along the horizontal direction; Represents the output along the vertical direction;

[0094] Coordinate attention generation first embeds the coordinate information into the extracted feature map connection, and then encodes the spatial information in the horizontal and vertical directions through calculation, and obtains the intermediate feature map k, which is expressed as:

[0095] k=δ(F1([z h ,z w ]))

[0096] In the formula, [z h ,z w ] means to set z h and z wSplicing along the spatial channel; δ represents a nonlinear activation function; F1 represents a 1×1 convolution transformation function;

[0097] Split the intermediate feature map k along the spatial dimension to obtain two independent tensors k h and k w , after convolution transformation and activation function, the two tensors have the same number of channels as the input X, and finally we get g h and g w As attention weights, it is expressed as follows:

[0098] g h =σ(K h (k h ))

[0099] g w =σ(K w (k w ))

[0100] In the formula, K h and K w All are 1×1 convolution transformations; σ represents the sigmoid activation function.

[0101] Furthermore, the backbone network adopts the GIOU function as the bounding box loss function and uses non-maximum suppression to filter the target objects in the image; wherein the GIOU function is expressed as:

[0102]

[0103] Where C represents the area of ​​the minimum circumscribed rectangle of the predicted box and the true box; U represents the union area of ​​the predicted box and the true box; the value range of GIoU is [-1,1]. When the predicted box and the true box completely overlap, GIoU takes the maximum value of 1. When they are completely disjoint and far away, GIoU takes the value of -1. IoU is an indicator used to measure the degree of overlap between the predicted box and the true box.

[0104] Furthermore, the loss function adopted by the backbone network is a GIOU function with power transformation, which can converge better. The GIOU function with power transformation introduced in the present invention is expressed as:

[0105]

[0106] In the formula, parameter α1>0, parameter α2>0, b represents the prediction box, b gt represents the real frame; According to b and b gt The regularization term for the calculation.

[0107] See also Figure 3In another embodiment of the present invention, a substation bird-repelling control system based on target detection is provided, comprising:

[0108] A model construction module 301 is used to construct a target detection model based on the YOLOv5 network architecture, wherein the target detection model includes an image enhancement module, a backbone network, a neck structure, and a detection head module;

[0109] The target detection module 302 is used to obtain the image to be detected in the substation area, pre-process the image to be detected using the image enhancement module, extract image features using the backbone network stacked convolutional layers, use the neck structure to perform feature fusion to form multi-scale features, and output the edge features and category results of the target object in the image to be detected through the detection head module;

[0110] The target driving away module 303 is used to generate an alarm signal in response to the driving away instruction generated based on the edge features and category results. The alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module to inject compressed gas to drive away the target object, which is a bird.

[0111] In summary, the present invention generates an expulsion instruction based on the target detection result, and can control the operation of the air compression injection system through an intelligent control system to achieve the expulsion of target objects such as birds. The control method based on target detection is real-time and accurate, and can effectively protect the safe operation of the substation.

[0112] like Figure 4 As shown, another embodiment of the present invention provides a computer device, the computer device 400 includes a memory and a processor;

[0113] The memory stores a computer program executable by the processor;

[0114] When the processor executes the computer program, a substation bird-repelling control method based on target detection as provided in the above embodiment is implemented.

[0115] Yet another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is read and executed, a substation bird-repelling control method based on target detection as provided in the above embodiment is implemented.

[0116] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the processes does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0117] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, the functional units in the embodiments of the present invention may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0118] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a memory and includes several requests for a computer device (which can be a personal computer, a server or a network device, etc., specifically a processor in a computer device) to perform some or all of the steps of the method described in each embodiment of the present invention.

[0119] In the embodiments provided by the present invention, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0120] A person skilled in the art can understand that part or all of the steps in the various methods of the embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0121] The above is a detailed introduction to a substation bird-repelling control method and system based on target detection disclosed in an embodiment of the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A substation bird-repelling control method based on target detection, characterized in that: include: Constructing a target detection model based on the YOLOv5 network architecture, the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module; Obtain the image to be detected in the substation area, use the image enhancement module to preprocess the image to be detected, use the backbone network to stack convolutional layers to extract image features, use the neck structure to fuse features to form multi-scale features, and output the edge features and category results of the target object in the image to be detected through the detection head module; In response to the expulsion instruction generated based on the edge features and category results, an alarm signal is generated, and the alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module, and compressed gas is injected to drive away the target object, which is a bird.

2. A substation bird-repelling control method based on target detection according to claim 1, characterized in that: The step of preprocessing the image to be detected by using the image enhancement module includes: Build an image enhancement module, which includes a lightweight enhancement parameter prediction network and an image enhancement iteration unit; Using the image to be detected as the input of the lightweight enhanced parameter prediction network to obtain parameter prediction of the curve parameter matrix; The matrix is ​​divided in the channel dimension to obtain a plurality of light enhancement parameters, and the plurality of light enhancement parameters are respectively used as iteration parameters of each image enhancement iteration unit; The input of each iteration unit is used as the intermediate representation image output by the previous iteration unit, and the image is cyclically enhanced, and the final output is used as the enhanced image.

3. A substation bird-repelling control method based on target detection according to claim 2, characterized in that: The lightweight enhanced parameter prediction network includes 7 convolutional layers; among them, the first 6 convolutional layers are 16 or 32 convolutional kernels with a size of 3×3 and a stride of 1, and ReLU activation is adopted; the last convolutional layer is T×3 convolutional kernels with a size of 3×3 and a stride of 1, and Tanh activation is adopted.

4. A substation bird-repelling control method based on target detection according to claim 3, characterized in that: In the step of performing image cyclic enhancement, a non-reference loss function is used to analyze the pixel spatial differences of the image itself and the correlation between the original image and the enhanced version of the image to calculate the loss; wherein the non-reference loss function includes spatial consistency loss, exposure control loss, color constancy loss and illumination smoothing loss.

5. A substation bird-repelling control method based on target detection according to claim 4, characterized in that: In the step of extracting image features by stacking convolutional layers of a backbone network, the backbone network is a DarkNet53 network using CSP, and a coordinate attention mechanism is added to the backbone network to encode coordinate information in the horizontal and vertical directions and embed it into the channel of the backbone network, so that the backbone network pays more attention to small target features when performing target detection and reduces the interference of background noise.

6. A substation bird-repelling control method based on target detection according to claim 5, characterized in that: The step of adding a coordinate attention mechanism to the backbone network includes: Two spatial range pooling kernels (H, 1) and (1, W) are used to encode each channel of the feature input and aggregate them in two directions to form a pair of position-aware feature maps, where the output of the cth channel in height and width is expressed as: In the formula, x c (h,i) represents the value of the hth row and ith column of channel c; x c (j,w) represents the value of the wth column and jth row of channel c; Represents the output along the horizontal direction; Represents the output along the vertical direction; Coordinate attention generation first embeds the coordinate information into the extracted feature map connection, and then encodes the spatial information in the horizontal and vertical directions through calculation, and obtains the intermediate feature map k, which is expressed as: k=δ(F1([z h ,z w ])) In the formula, [z h ,z w ] means to set z h and z w Splicing along the spatial channel; δ represents a nonlinear activation function; F1 represents a 1×1 convolution transformation function; Split the intermediate feature map k along the spatial dimension to obtain two independent tensors k h and k w , after convolution transformation and activation function, the two tensors have the same number of channels as the input X, and finally we get g h and g w As attention weights, it is expressed as follows: g h =σ(K h (k h )) g w =σ(K w (k w )) In the formula, K h and K w All are 1×1 convolution transformations; σ represents the sigmoid activation function.

7. A substation bird-repelling control method based on target detection according to claim 6, characterized in that: The backbone network uses the GIOU function as the bounding box loss function and uses non-maximum suppression to filter the target objects in the image; wherein the GIOU function is expressed as: Where C represents the area of ​​the minimum circumscribed rectangle of the predicted box and the true box; U represents the union area of ​​the predicted box and the true box; the value range of GIoU is [-1,1]. When the predicted box and the true box completely overlap, GIoU takes the maximum value of 1. When they are completely disjoint and far away, GIoU takes the value of -1. IoU is an indicator used to measure the degree of overlap between the predicted box and the true box.

8. A substation bird-repelling control method based on target detection according to claim 7, characterized in that: The loss function used by the backbone network is the GIOU function with the introduction of power transformation, which is expressed as: In the formula, parameter α1>0, parameter α2>0, b represents the predicted box, and bgt represents the real box; Represents the regularization term calculated based on b and bgt.

9. A substation bird-repelling control method based on target detection according to claim 8, characterized in that: The compressed air injection system comprises: A screw air compressor with an exhaust pressure of 0.8-0.85Mpa; A 3m 3 The normal pressure of the gas tank is 0.7Mpa. The gas tank is equipped with an electric contact pressure gauge. When the pressure of the gas tank drops to 0.6Mpa, the screw air compressor is automatically started. When the pressure is increased to 0.7Mpa, the screw air compressor is automatically stopped. The electric contact pressure gauge has a pressure analog module to transmit the pressure of the gas tank to the intelligent control system in real time. A carbon steel main steam pipeline with a diameter of 60mm and 20 insulated branch pipelines with a diameter of 20mm. Each branch pipeline is equipped with an electromagnetic valve, and the valve switch is controlled by the control system DO module. A 90-degree elbow is installed at the outlet of the branch pipeline to make the compressed gas horizontal to the substation equipment point.

10. A control system for implementing a substation bird-repelling control method based on target detection as claimed in any one of claims 1 to 9, characterized in that: The control system comprises: A model construction module is used to construct a target detection model based on the YOLOv5 network architecture, wherein the target detection model includes an image enhancement module, a backbone network, a neck structure and a detection head module; The target detection module is used to obtain the image to be detected in the substation area, pre-process the image to be detected using the image enhancement module, extract image features using the backbone network stacked convolutional layers, fuse features using the neck structure to form multi-scale features, and output the edge features and category results of the target object in the image to be detected through the detection head module; The target driving away module is used to generate an alarm signal in response to the driving away instruction generated based on the edge features and category results. The alarm signal is transmitted to the intelligent control system through the communication module. After the intelligent control system is operated by the CPU module, the action instruction is sent to the gas pipeline solenoid valve of the air compression injection system of the alarm point through the DO module, and the compressed gas is injected to drive away the target object, which is a bird.

Images