Device detection method, system, computer storage medium and electronic device

Abstract

The present application discloses a detection method, system, computer storage medium and electronic device for equipment. It relates to the field of drone inspection technology and image detection or other related fields. The method includes: obtaining the original image data of the target equipment, preprocessing the original image data, and obtaining the preprocessed image data; inputting the preprocessed image data into the target detection model, and outputting the image detection result, wherein the image detection result includes a category prediction result and a position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the location information of the fault area in the target equipment; the preprocessed image data and the image detection result are sent to the staff operation terminal. Through this application, the problems of low detection efficiency and low detection accuracy in the detection of power equipment and transmission lines in the related technology are solved.

Description

Technical Field

[0001] The present application relates to the fields of drone inspection technology and image detection or other related fields, and specifically, to a device detection method, system, computer storage medium and electronic device. Background Art

[0002] In the power industry, in order to ensure the stable operation of transmission line equipment, it is often necessary to conduct regular inspections of transmission lines and power equipment. The traditional inspection method requires manual inspection of the lines by climbing the tower, which is time-consuming, labor-intensive, inefficient, costly, and has certain safety risks. With the rapid development of artificial intelligence technology, especially breakthroughs in the field of deep learning, machines can handle more complex tasks. Therefore, in the power industry, this technology is increasingly widely used, especially in the field of power inspection. The deep learning model carried by drones can realize the automated detection of transmission lines and equipment.

[0003] Although the above-mentioned deep learning model has obvious advantages in inspection effect, the model has a large number of parameters and a complex structure. The computing and storage requirements required for model training are also showing an increasing trend, which makes it difficult to deploy and run these models in resource-constrained environments and cannot meet real-time requirements. Especially in mobile devices and edge computing scenarios, the high computational complexity of these models will cause communication delays and high load on network bandwidth, making it impossible to efficiently implement detection tasks.

[0004] With regard to the problems of low detection efficiency and low detection accuracy when detecting power equipment and transmission lines in related technologies, no effective solution has been proposed so far. Summary of the invention

[0005] The main purpose of the present application is to provide a device detection method, system, computer storage medium and electronic device to solve the problems of low detection efficiency and low detection accuracy when detecting power equipment and transmission lines in the related art.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a device detection method is provided. The method comprises: obtaining the original image data of the target device, preprocessing the original image data, and obtaining the preprocessed image data, wherein the original image data comprises the image data of the power equipment and the image data of the transmission line; inputting the preprocessed image data into the target detection model, and outputting the image detection result, wherein the image detection result comprises the category prediction result and the position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the location information of the fault area in the target device; and sending the preprocessed image data and the image detection result to the staff operation terminal.

[0007] Furthermore, the target detection model includes a feature extraction module and a detection head module. The preprocessed image data is input into the target detection model, and the output image detection result includes: inputting the preprocessed image data into the feature extraction module, extracting features in the preprocessed image data through the feature extraction module, and outputting a fused feature map; obtaining N attention dimension parameters, inputting the fused feature map and the N attention dimension parameters into the detection head module, and obtaining an image detection result, wherein the image detection result is obtained by the detection head module processing the fused feature map using the N attention dimension parameters in sequence, and N is a positive integer.

[0008] Furthermore, the feature extraction module includes a sub-feature extraction module and a feature fusion module, the sub-feature extraction module includes M convolutional network modules, M is a positive integer, and the features in the preprocessed image data are extracted by the feature extraction module. The output fusion feature map includes: inputting the preprocessed image data into the sub-feature extraction module to obtain M layers of feature maps, wherein the M convolutional network modules in the sub-feature extraction module perform feature extraction processing on the preprocessed image data in turn to obtain M layers of feature maps; inputting the M layers of feature maps into the feature fusion module, and using the feature fusion module to perform fusion processing on the M layers of feature maps to obtain a fused feature map.

[0009] Furthermore, each layer of feature maps is obtained in the following manner: for a convolutional network module, the preprocessed image data is normalized by the convolutional network module to obtain initial image features; normalization weights are obtained, and a preset activation function is obtained, the initial image features are processed according to the normalization weights to obtain normalized features, and the normalized features are input into the preset activation function to obtain normalized function features; a gated decision function is obtained, and the normalized function features are calculated using the gated decision function to obtain image refinement features, and the image refinement features are segmented to obtain sub-feature maps, wherein the image refinement features are used to characterize the importance of the normalization function features; the initial image features are mapped by a fully connected layer to obtain processed channel features, and dot product calculations are performed on the sub-feature maps and the processed channel features to obtain a layer of feature maps associated with the convolutional network module.

[0010] Furthermore, before inputting M layers of feature maps into the feature fusion module, the method also includes: obtaining the attention weight of each layer of feature maps to obtain M groups of neuron data, wherein each group of neuron data is used to enhance the feature representation of each layer of feature maps, and each group of neuron data includes first neuron data and second neuron data; calculating the mean of each layer of feature maps according to each group of neuron data to obtain M feature means, and calculating the variance of each layer of feature maps according to each group of neuron data to obtain M feature variances; for a layer of feature maps, determining the initial feature weight value of the feature map according to the feature variance and feature mean associated with the feature map, and determining the target feature weight value of the feature map according to the initial feature weight value; processing each layer of feature maps according to the target feature weight value of each layer to obtain M processed feature maps; inputting M layers of feature maps into the feature fusion module includes: inputting the M processed feature maps into the feature fusion module.

[0011] Furthermore, the feature map of each layer is processed according to the target feature weight value of each layer to obtain M processed feature maps, including: obtaining a preset activation function, performing spatial dimension grouping processing on the M target feature weight values ​​to obtain feature parameters, and calculating the inverse of the feature parameters according to the preset activation function to obtain a feature function value; performing dot product calculation on the feature function value and the feature map of each layer to obtain M processed feature maps.

[0012] Furthermore, before inputting the preprocessed image data into the target detection model, the method also includes: obtaining a weight function and obtaining a loss function of a preset image detection model, adjusting the loss function using the weight function to obtain an adjusted loss function, wherein the target detection model is obtained by processing the preset image detection model; optimizing the preset image detection model using the adjusted loss function to obtain a processed image detection model, and determining the processed image detection model as the target detection model.

[0013] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a detection system for equipment is provided. The system includes: a drone control system, including a hardware system, a detection module and a communication module, the hardware system is used to obtain the original image data of the target equipment, preprocess the original image data, obtain the preprocessed image data, input the preprocessed image data into the detection module, the detection module is used to input the preprocessed image data into the target detection model when receiving the preprocessed image data, output the image detection result, and send the preprocessed image data and the image detection result to the communication module, the communication module is used to send the preprocessed image data and the image detection result to the operation end; the operation end is connected to the communication module for communication, and is used to generate inspection instruction information when receiving the preprocessed image data and the image detection result, and send the inspection instruction information to the operator.

[0014] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a detection device for equipment is provided. The device comprises: a first acquisition unit, used to acquire the original image data of the target equipment, preprocess the original image data, and obtain the preprocessed image data, wherein the original image data includes the image data of the power equipment and the image data of the transmission line; an input unit, used to input the preprocessed image data into the target detection model, and output the image detection result, wherein the image detection result includes the category prediction result and the position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the location information of the fault area in the target equipment; a sending unit, used to send the preprocessed image data and the image detection result to the staff operation terminal.

[0015] According to another aspect of an embodiment of the present invention, a computer storage medium is provided. The computer storage medium is used to store a program. When the program is executed, the device where the computer storage medium is located is controlled to execute a device detection method.

[0016] According to another aspect of an embodiment of the present invention, an electronic device is provided, comprising one or more processors and a memory; the memory stores computer-readable instructions, and the processor is used to execute the computer-readable instructions, wherein a device detection method is executed when the computer-readable instructions are executed.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, a detection method of a device is executed.

[0018] Through the present application, the following steps are adopted: obtaining original image data of the target device, preprocessing the original image data, and obtaining preprocessed image data, wherein the original image data includes image data of the power equipment and image data of the transmission line; inputting the preprocessed image data into the target detection model, and outputting image detection results, wherein the image detection results include category prediction results and position prediction results, the category prediction results are used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction results are used to characterize the location information of the fault area in the target device; the preprocessed image data and the image detection results are sent to the staff operation end, which solves the problems of low detection efficiency and low detection accuracy when detecting power equipment and transmission lines in the related technology, by obtaining the original image data of the device, inputting the original image data into the target detection model after preprocessing, obtaining image detection results, and then sending the image detection results to the operation end, thereby achieving the effect of improving the safety of detecting power equipment and transmission lines and improving detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0020] Figure 1 is a schematic diagram of a detection system of a device provided according to an embodiment of the present application;

[0021] Figure 2 is a flow chart of a detection method for a device provided in an embodiment of the present application;

[0022] Figure 3 is a schematic diagram of the network structure of the target detection model provided according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a detection head module in a target detection model provided according to an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a convolutional network module of a target detection model provided according to an embodiment of the present application;

[0025] Figure 6 is a schematic diagram of a detection device of a device provided according to an embodiment of the present application;

[0026] Figure 7 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0031] It should be noted that the collected information used in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse to use it.

[0032] The present invention is described below in conjunction with preferred implementation steps. The present application embodiment provides a detection system for a device. The detection system for a device provided by the present application embodiment is introduced below. Figure 1is a schematic diagram of a detection system of a device provided in an embodiment of the present application, such as Figure 1 As shown, the system includes:

[0033] The drone control system 101 includes a hardware system, a detection module and a communication module. The hardware system is used to obtain raw image data of a target device, preprocess the raw image data to obtain preprocessed image data, and input the preprocessed image data into the detection module. The detection module is used to input the preprocessed image data into a target detection model when receiving the preprocessed image data, output an image detection result, and send the preprocessed image data and the image detection result to the communication module. The communication module is used to send the preprocessed image data and the image detection result to the operation terminal 102;

[0034] Specifically, in order to conduct safe, fast and effective inspections of transmission lines and power equipment, a drone control system 101 can be built to conduct real-time inspections of target equipment. It should be noted that the target equipment refers to the inspection target, which can be the transmission line and power equipment, and the drone control system 101 is equipped with a hardware system, a detection module and a communication module, wherein the hardware system includes a GPS (Global Positioning System) positioning system, a camera acquisition device and a drone flight system. The GPS positioning system can obtain the exact position of the inspection target (that is, the transmission line and power equipment), the camera acquisition device can collect image data of the transmission line and power equipment, and the drone flight system can control the flight posture, speed and flight direction of the drone, that is, the drone control system 101 can use the camera acquisition device to collect the original image data of the transmission line and power equipment during the flight of the drone, and then pre-process the original image data and send it to the detection module provided with a target detection model.

[0035] After the detection module receives the preprocessed image data of the transmission line and the power equipment, the target detection model therein can be used to perform real-time detection on the above image data to obtain the image detection result, and the preprocessed image data and the image detection result are sent to the communication module through the remote data transmission technology, wherein the target detection model can be a detection model improved based on the YOLOv7 lightweight detection model, the convolutional network and the parameter-free attention mechanism module, which can enhance the weight of the key features in the image data, thereby enhancing the expression ability of the target detection model. After receiving the preprocessed image data and the image detection result, the communication module can forward the data to the operator 102 of the staff, so that the staff can timely check the relevant detection results based on the operator 102 and make maintenance instructions in time.

[0036] The operating terminal 102 is in communication connection with the communication module, and is used for generating inspection instruction information upon receiving the pre-processed image data and the image detection result, and sending the inspection instruction information to the operator.

[0037] Specifically, after the operating end 102 receives the pre-processed image data and image detection results received by the communication module, the image detection results can be analyzed to obtain inspection instruction information, wherein the inspection instruction information is used to represent the instruction information for maintenance processing of transmission lines and power equipment. Then, after receiving the information, the maintenance operator can respond and process it in a timely manner, thereby reducing resource waste.

[0038] The detection system of the device provided in the embodiment of the present application includes a hardware system, a detection module and a communication module through the drone control system 101. The hardware system is used to obtain the original image data of the target device, pre-process the original image data to obtain the pre-processed image data, and input the pre-processed image data into the detection module. The detection module is used to input the pre-processed image data into the target detection model when receiving the pre-processed image data, output the image detection result, and send the pre-processed image data and the image detection result to the communication module. The communication module is used to send the pre-processed image data and the image detection result to Operation terminal 102; Operation terminal 102 is communicatively connected with the communication module, and is used for generating inspection instruction information upon receiving preprocessed image data and image detection results, and sending the inspection instruction information to the operator, thereby solving the problems of low detection efficiency and low detection accuracy in the related technology when detecting power equipment and transmission lines, by acquiring the original image data of the equipment, inputting the original image data into the target detection model after preprocessing, obtaining the image detection result, and then sending the image detection result to the operation terminal, thereby achieving the effect of improving the safety of detecting power equipment and transmission lines and improving the detection efficiency and accuracy.

[0039] The present invention is described below in conjunction with preferred implementation steps. Figure 2 is a flow chart of a detection method for a device provided in an embodiment of the present application, such as Figure 2 As shown, the method comprises the following steps:

[0040] Step S201, acquiring original image data of a target device, and preprocessing the original image data to obtain preprocessed image data, wherein the original image data includes image data of the power device and image data of the power transmission line.

[0041] Specifically, the target device can be a transmission line and power equipment in a transmission system. The original image data can refer to images of the transmission line and power equipment collected by a camera acquisition device in a drone control system. After obtaining the original image data of the target device, in order to make the image data meet the input standard of the target detection model, the original image data can be preprocessed by format unification, size adjustment, data annotation, etc., and basic enhancement work such as random rotation, grayscale conversion, and noise filtering can be implemented to obtain preprocessed image data. For example, after processing the original image data, preprocessed image data with a size of 640×640×3 can be obtained. This method can not only provide high-quality and wide-coverage data support for the training of subsequent detection models, but also improve the robustness and generalization ability of the model to environmental changes.

[0042] Step S202, input the preprocessed image data into the target detection model, and output the image detection result, wherein the image detection result includes a category prediction result and a position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the location information of the fault area in the target device.

[0043] Specifically, the target detection model can be a model based on the YOLOv7 lightweight detection model. After obtaining the preprocessed image data, the target detection model can be used to detect and process the image, so as to obtain a target detection result including a category prediction result and a position prediction result. The target detection model is based on the YOLOv7 model architecture and is integrated with the spatial and channel reconstruction convolution SCConv, which can compress the spatial and channel redundancy between features and reduce the computational complexity of the target detection model. Secondly, a parameter-free attention mechanism is introduced, which can infer three-dimensional attention for the feature map within the layer without increasing the network parameters, and can enhance the representation weight of the key features in the preprocessed image data. Then, an SC-Dyhead detection module is set up, which can achieve lightweight while enhancing the expression ability of the detection head. Finally, the model is integrated with the Slide loss function, which can assign higher weights to difficult samples and enhance the expression ability, application effect and robustness of the model.

[0044] Step S203, sending the pre-processed image data and image detection results to the staff operation terminal.

[0045] Specifically, after using the target detection model to detect the preprocessed image data and obtain the image detection results, the preprocessed image data and the image detection results can be sent to the staff operation end. After receiving the relevant information, the staff can analyze the image detection results and determine the inspection instruction information used to characterize the maintenance of the transmission lines and power equipment. After receiving the information, the maintenance operator can respond and process it in a timely manner, reduce resource waste, and combine the human-computer interaction mode to enhance the reliability of the inspection results. For example, the drone control system can transmit the pre-processed image data and image detection results to the operation end in real time through wireless communication (such as Wi-Fi or mobile network). The operation end provides a display and interactive interface, and can also display the flight path, altitude, speed and other parameters of the drone, while presenting the detection results to the staff and allowing the staff to conduct real-time human-computer interaction with the drone; in addition, the operation end can remotely operate and control the drone through network communication. The staff can remotely control the flight path and mission execution of the drone through an intuitive human-computer interaction interface, and instantly view and review the images and analysis results sent back by the drone. If necessary, manual review or in-depth analysis can be performed. For example, the staff can provide feedback and instructions by marking or annotating the detection results to improve the accuracy and reliability of the inspection results.

[0046] The detection method for equipment provided in the embodiment of the present application obtains original image data of the target equipment, pre-processes the original image data, and obtains pre-processed image data, wherein the original image data includes image data of the power equipment and image data of the transmission line; the pre-processed image data is input into the target detection model, and the image detection result is output, wherein the image detection result includes a category prediction result and a position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the pre-processed image data, and the position prediction result is used to characterize the location information of the fault area in the target equipment; the pre-processed image data and the image detection result are sent to the staff operation end, which solves the problems of low detection efficiency and low detection accuracy when detecting power equipment and transmission lines in the related technology, and obtains the original image data of the equipment, inputs the original image data into the target detection model after pre-processing, obtains the image detection result, and then sends the image detection result to the operation end, thereby achieving the effect of improving the safety of detecting power equipment and transmission lines and improving the detection efficiency and accuracy.

[0047] In order to improve the detection accuracy of the target detection model, optionally, in the detection method of the device provided in the embodiment of the present application, the target detection model includes a feature extraction module and a detection head module, and the preprocessed image data is input into the target detection model, and the output image detection result includes: inputting the preprocessed image data into the feature extraction module, extracting features in the preprocessed image data through the feature extraction module, and outputting a fused feature map; obtaining N attention dimension parameters, inputting the fused feature map and the N attention dimension parameters into the detection head module, and obtaining an image detection result, wherein the image detection result is obtained by the detection head module processing the fused feature map using the N attention dimension parameters in sequence, and N is a positive integer.

[0048] Specifically, the target detection model can be a model based on the YOLOv7 lightweight detection model. Figure 3 is a schematic diagram of the network structure of the target detection model provided in the embodiment of the present application, such as Figure 3 As shown, in order to enhance the expressiveness and robustness of the target detection model, a variety of module structures can be added to the model. For example, a feature extraction module composed of a sub-feature extraction module and a feature fusion module can be added. The sub-feature extraction module can be an efficient Backbone network architecture of ELAN, which includes a convolutional network module. The convolutional network module can be an SCConv convolutional network module, that is, a more lightweight ELAN-S structure is obtained. The feature fusion module can be a Neck structure. Before inputting the multi-layer feature map into the feature fusion module for feature fusion processing, the feature fusion module can include a SimAM attention mechanism. In addition, a detection head module composed of Rep and conv can be added, and a Dyhead module can be introduced into the detection head module. When the target detection model is constructed based on the above modules and the target detection model is used for image detection processing, the pre-processed image data is first input into the feature extraction module, and then the feature information in the image data is extracted by the adjustment extraction module, so as to obtain a fused feature map after the fusion of multiple layers of feature maps of different scales.

[0049] Furthermore, after obtaining the fused feature map, the feature map is input into the detection head module composed of Rep and conv to predict the target category and position, and then output the target detection result, wherein the target category refers to the category information of the area with faults in the preprocessed image data, and the position refers to the position information of the area with faults in the preprocessed image data. It should be noted that Figure 4 is a schematic diagram of a detection head module in a target detection model provided in an embodiment of the present application, such as Figure 4As shown in the figure, the Dyhead module is introduced into the detection head module. After redesigning with the above SCConv convolutional network module, the SC-Dyhead detection module can be obtained. When the detection head module makes predictions, multiple attention dimension parameters can be added. For example, The three-dimensional attention dimension parameters, and then the fusion feature map, are sequentially subjected to the attention mechanisms of scale perception, space perception, and task perception. This can further improve the lightweight of the detection head module and enhance the expression ability of the detection head without increasing the amount of calculation. That is, the image detection result can be calculated by the following formula : ,in, This embodiment uses a target detection model provided with a feature extraction module and a detection head module to perform image detection, thereby improving the accuracy of image detection and also improving the efficiency of detection.

[0050] In order to achieve high-precision recognition and analysis, optionally, in the detection method of the device provided in the embodiment of the present application, the feature extraction module includes a sub-feature extraction module and a feature fusion module, the sub-feature extraction module includes M convolutional network modules, M is a positive integer, and the features in the preprocessed image data are extracted by the feature extraction module. The output fusion feature map includes: inputting the preprocessed image data into the sub-feature extraction module to obtain M-layer feature maps, wherein the M convolutional network modules in the sub-feature extraction module perform feature extraction processing on the preprocessed image data in turn to obtain M-layer feature maps; inputting the M-layer feature maps into the feature fusion module, and using the feature fusion module to fuse the M-layer feature maps to obtain a fused feature map.

[0051] It should be noted that the sub-feature extraction module can be a module composed of the ELAN efficient Backbone network architecture. In order to reduce and remove the spatial and channel redundant parameters of the feature map in the convolutional neural network, enhance the spatial feature extraction capability, and reduce the computing resources required for the model, the Backbone network architecture can introduce the convolutional network module into the ELAN module in the backbone network, that is, to obtain a more lightweight ELAN-S structure, wherein each convolutional network module can be a lightweight spatial and channel reconstruction convolutional network. For example, the convolutional network module can be an SCConv convolutional network module, which can be composed of a spatial reconstruction unit and a channel reconstruction unit in series, which can reduce the redundancy between space and channels in the standard convolution, thereby reducing the network computing cost and model storage, and has better feature extraction capability; the feature fusion module can be a module composed of a Neck structure.

[0052] Specifically, when the feature extraction module is used for feature extraction processing, the preprocessed image data can first be input into the sub-feature extraction module, and multiple convolutional network modules perform feature extraction processing on the preprocessed image data in turn to obtain a multi-layer feature map. For example, when the convolutional network module is a three-layer convolutional network, the obtained feature map is a feature map of three layers of different scales. After obtaining the multi-layer feature map, the feature fusion module can integrate these feature maps into a single, comprehensive feature representation, that is, to obtain a fused feature map. For example, all feature maps are first spliced ​​together along a specific dimension, and then different weights are assigned to each feature map, which are added together, and finally the attention mechanism is used to represent important features. This embodiment uses the addition of a feature extraction module in the target detection model to perform feature extraction operations, which can more effectively extract information from image data and achieve high-precision recognition and analysis, while being able to significantly maintain a high detection performance, thereby achieving the demand for lightweight detection models.

[0053] In order to enhance the spatial expression ability of the feature map and capture local features and structural information, optionally, in the detection method of the device provided in the embodiment of the present application, each layer of the feature map is obtained in the following manner: for a convolutional network module, the preprocessed image data is normalized by the convolutional network module to obtain the initial image feature; the normalization weight is obtained, and the preset activation function is obtained, the initial image feature is processed according to the normalization weight to obtain the normalized feature, and the normalized feature is input into the preset activation function to obtain the normalized function feature; the gated decision function is obtained, and the normalized function feature is calculated by using the gated decision function to obtain the image refinement feature, and the image refinement feature is segmented to obtain a sub-feature map, wherein the image refinement feature is used to characterize the importance of the normalization function feature; the initial image feature is mapped by the fully connected layer to obtain the processed channel feature, and the sub-feature map and the processed channel feature are dotted to obtain a layer of feature map associated with the convolutional network module.

[0054] Specifically, when the convolutional network module is used to extract features from the preprocessed image data, in the spatial dimension, the preprocessed image data (i.e., feature X) is first normalized (i.e., GN processing) by the convolutional network module to obtain the initial image features (i.e., feature ), that is, the initial image features are calculated by the following formula: . Further, obtain the normalized weights and preset activation function, that is, obtain the random initialization Normalized channel weights of parameters And sigmoid preset activation function, and then the initial image features are normalized according to the above weights Processing to obtain normalized features , where this method can map the product value of the output feature and the normalized weight to the range of (0, 1), and use the threshold for judgment. If the normalized feature is greater than the threshold, it indicates that the normalized feature is a valid feature, otherwise it is considered an invalid feature. Then the normalized feature is input into the above activation function to obtain the normalized function feature, that is, .

[0055] Furthermore, after obtaining the gate decision function Finally, the normalized function features are calculated using the gated decision function to obtain image refinement features. It should be noted that this method performs cross reconstruction in the spatial dimension, which can enhance effective features and suppress invalid features, and finally obtain the processed image refinement features, that is, the image refinement features can be calculated by the following formula: , where W represents the strong and weak image refinement features after the determination process, Represents the gate decision function, and sigmoid represents the preset activation function.

[0056] Further, Figure 5 is a schematic diagram of a convolutional network module of a target detection model provided in an embodiment of the present application, such as Figure 5 As shown in the figure, in order to obtain effective information while ensuring the lightness of information redundancy, it is also necessary to perform redundant processing on the image refinement features obtained in the spatial dimension using the segmentation-conversion-fusion strategy to obtain sub-feature maps. That is, segmentation processing refers to the segmentation processing of the input image refinement features in the spatial reconstruction unit in order to separately process and analyze the feature relationships within different channel groups to facilitate subsequent dimensionality reduction and feature reorganization. In the channel processing stage, the input image refinement features are segmented and divided into multiple sub-feature maps in the spatial reconstruction unit, each of which contains specific information. The conversion process refers to the use of one-dimensional convolution (i.e., 1x1 convolution) to reduce the number of channels for each subset of sub-feature maps, and then the global features are mapped back to the dimension of the original number of channels through a one-dimensional convolution or a fully connected layer, reducing the amount of calculation while removing redundant information between channels, that is, the initial image features are mapped through a fully connected layer to obtain the processed channel features, and finally a layer of feature maps associated with a convolutional network module is obtained by fusion processing, that is, the processed channel features and the sub-feature maps are fused through a dot multiplication operation in the channel reconstruction unit, and the information of each sub-feature map is integrated into a final feature map to form a more representative and discriminative feature representation, that is, a layer of feature maps associated with the convolutional network module (i.e., channel refinement features) is obtained. This embodiment processes the input pre-processed image data in the spatial dimension through a convolutional network module, which can obtain richer spatial information and refined features, and can also enhance the spatial expression ability of the feature map and capture local features and structural information.

[0057] In order to determine the importance of the feature map, optionally, in the detection method of the device provided in the embodiment of the present application, before inputting M layers of feature maps into the feature fusion module, the method also includes: obtaining the attention weight of each layer of feature maps to obtain M groups of neuron data, wherein each group of neuron data is used to enhance the feature representation of each layer of feature maps, and each group of neuron data includes first neuron data and second neuron data; calculating the mean of each layer of feature maps according to each group of neuron data to obtain M feature means, and calculating the variance of each layer of feature maps according to each group of neuron data to obtain M feature variances; for a layer of feature maps, determining the initial feature weight value of the feature map according to the feature variance and feature mean associated with the feature map, and determining the target feature weight value of the feature map according to the initial feature weight value; processing each layer of feature maps according to the target feature weight value of each layer to obtain M processed feature maps; inputting M layers of feature maps into the feature fusion module includes: inputting the M processed feature maps into the feature fusion module.

[0058] Specifically, before the multi-layer feature map is input into the feature fusion module for feature fusion processing, in order to improve the lightweight of the network, the SimAM attention mechanism without adding additional parameters can be introduced. It is added before the backbone network is output to the neck network (ie, the Neck structure), so that the spatial attention can be adaptively focused according to the input feature map, and the importance of different features can be found by constructing a binary energy function, thereby giving the features within the layer Figure 3 First, we need to obtain the attention weight of each layer of feature map, and then enhance the feature representation, that is, by calculating the energy of each neuron in the feature map to determine different attention weights, and obtain multiple sets of neuron data. It should be noted that the attention weight can identify which features are important by comparing the features of a position with the features of the surrounding positions.

[0059] Furthermore, the mean and variance of each layer of feature maps can be calculated using each group of neuron data to obtain multiple feature means and feature variances, and then the initial feature weight value of each layer of feature maps is determined according to the feature variance and feature mean, and the target feature weight value of each layer of feature maps is determined according to the initial feature weight value. Finally, each layer of feature maps is processed based on the target feature weight value to obtain multiple processed feature maps, which are then input into the feature fusion module, wherein the initial feature weight value can be calculated by the following formula:

[0060] ;

[0061] in, , t and They represent the target neuron (i.e., the first neuron data) and other neurons (i.e., the second neuron data) of each layer of the input feature map, i represents the index of the spatial dimension, represents the number of neurons on the channel, Indicates the preset parameters. represents the weighted value, represents the bias transformation, y represents the binary label, represents the i-th feature in the input data, Represent the feature mean and feature variance respectively. In order to reduce repeated calculations at each position, all neurons on the same channel reuse the feature mean and feature variance obtained above. The target feature weight value can be obtained through the following mathematical relationship:

[0062] ;

[0063] When the target feature weight The lower the value, the lower the similarity with the surrounding neurons, which means that the importance of the feature map corresponding to the target feature weight value is also higher. This embodiment calculates the importance of each layer of feature maps by using a parameter-free attention mechanism, which can more comprehensively consider the information in the feature map, thereby improving the model's discrimination ability and detection speed, and laying the foundation for subsequent detection of image data features.

[0064] Optionally, in the detection method of the device provided in the embodiment of the present application, each layer of the feature map is processed according to the target feature weight value of each layer to obtain M processed feature maps, including: obtaining a preset activation function, performing spatial dimension grouping processing on the M target feature weight values ​​to obtain feature parameters, and calculating the inverse of the feature parameters according to the preset activation function to obtain a feature function value; performing dot product calculation on the feature function value and the feature map of each layer to obtain M processed feature maps.

[0065] Specifically, in order to enhance the feature representation of each layer of feature maps, the features can also be enhanced. First, a preset activation function can be obtained, and then all target feature weight values ​​are grouped across channels and spatial dimensions to obtain feature parameters E. The feature parameters are then used to obtain the processed feature maps. The processed feature maps can be calculated by the following formula: , where X represents the feature map of each layer, represents a preset activation function, and E represents a feature parameter. In this embodiment, the feature map is enhanced again by using the feature parameter, so as to lay a foundation for the subsequent detection of the features of the image data.

[0066] In order to improve the detection capability of the target detection model, optionally, in the detection method of the device provided in the embodiment of the present application, before the preprocessed image data is input into the target detection model, the method also includes: obtaining a weight function, and obtaining a loss function of a preset image detection model, adjusting the loss function using the weight function to obtain an adjusted loss function, wherein the target detection model is obtained by processing the preset image detection model; optimizing the preset image detection model using the adjusted loss function to obtain a processed image detection model, and determining the processed image detection model as the target detection model.

[0067] Specifically, since the UAV control system is prone to an imbalance in the number of positive and negative samples when inspecting and photographing the image data of the target device, the target detection model tends to focus on routine problems, which will make it difficult for the model to distinguish between lines and ground cracks. Therefore, a loss function can be set in the detection head module of the target detection model. Based on the setting of the loss function of YOLOv7, the regression loss function CIOU can be improved to a Slide loss function, that is, the sample weights are assigned through the weight function Slide, and the loss function is adjusted using the weight function to obtain the adjusted loss function, which can then achieve the goal of quickly focusing the center of gravity on difficult-to-distinguish samples. Among them, the adjusted loss function of the detection head module in YOLOv7 Including object confidence loss (This loss function uses binary cross entropy loss function), object classification loss (This loss function uses the binary cross entropy loss function) and coordinate regression loss The three parts, that is, the adjusted loss function in the detection head module is determined by the following formula: , where the coordinate regression loss It is obtained by improving the Llbox coordinate loss through the slide weight function.

[0068] It should be noted that in order to improve the detection ability of the target detection model, it is necessary to effectively increase the attention paid to difficult samples. Therefore, the average value of all coordinate regression losses can be used as the threshold , will be less than the threshold is classified as a negative sample, which is greater than the threshold The samples are classified as positive samples, and those near the threshold are given higher weights, where the Slide weight function is as follows:

[0069] ;

[0070] The adjusted loss function can be further used to optimize the preset image detection model to obtain the target detection model. In this embodiment, the target detection model is adjusted by using the loss function, thereby improving the efficiency and accuracy of detection.

[0071] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0072] The present application also provides a device detection apparatus. It should be noted that the device detection apparatus of the present application can be used to execute the device detection method provided in the present application. The device detection apparatus provided in the present application is introduced below.

[0073] Figure 6 is a schematic diagram of a detection device according to an embodiment of the present application, such as Figure 6 As shown, the device includes: a first acquisition unit 60, an input unit 61, and a sending unit 62.

[0074] A first acquisition unit 60 is used to acquire original image data of a target device, and preprocess the original image data to obtain preprocessed image data, wherein the original image data includes image data of the power device and image data of the power transmission line;

[0075] An input unit 61 is used to input the preprocessed image data into the target detection model and output the image detection result, wherein the image detection result includes a category prediction result and a position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the position information of the fault area in the target device;

[0076] The sending unit 62 is used to send the pre-processed image data and the image detection result to the staff operation terminal.

[0077] The detection device of the equipment provided in the embodiment of the present application obtains the original image data of the target equipment through the first acquisition unit 60, pre-processes the original image data, and obtains the pre-processed image data, wherein the original image data includes the image data of the power equipment and the image data of the transmission line; the input unit 61 inputs the pre-processed image data into the target detection model, and outputs the image detection result, wherein the image detection result includes the category prediction result and the position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the pre-processed image data, and the position prediction result is used to characterize the location information of the fault area in the target equipment; the sending unit 62 sends the pre-processed image data and the image detection result to the staff operation end, which solves the problem of low detection efficiency and low detection accuracy when detecting power equipment and transmission lines in the related technology, by acquiring the original image data of the equipment, inputting the original image data into the target detection model after pre-processing, obtaining the image detection result, and then sending the image detection result to the operation end, thereby achieving the effect of improving the safety of detecting power equipment and transmission lines and improving the detection efficiency and accuracy.

[0078] Optionally, in the detection device of the equipment provided in the embodiment of the present application, the input unit 61 includes: a first input module, used to input the preprocessed image data into a feature extraction module, extract features in the preprocessed image data through the feature extraction module, and output a fused feature map; a first acquisition module, used to obtain N attention dimension parameters, input the fused feature map and N attention dimension parameters into the detection head module, and obtain an image detection result, wherein the image detection result is obtained by the detection head module using N attention dimension parameters in sequence to process the fused feature map, and N is a positive integer.

[0079] Optionally, in the detection device of the equipment provided in the embodiment of the present application, the input unit 61 includes: a second input module, used to input the preprocessed image data into a sub-feature extraction module to obtain an M-layer feature map, wherein the M convolutional network modules in the sub-feature extraction module perform feature extraction processing on the preprocessed image data in turn to obtain an M-layer feature map; a third input module, used to input the M-layer feature map into a feature fusion module, and use the feature fusion module to perform fusion processing on the M-layer feature map to obtain a fused feature map.

[0080] Optionally, in the detection device of the device provided in the embodiment of the present application, the input unit 61 includes: each layer of feature maps is obtained in the following manner: a first processing module, which is used to normalize the preprocessed image data for a convolutional network module through the convolutional network module to obtain initial image features; a second acquisition module, which is used to obtain normalization weights and a preset activation function, process the initial image features according to the normalization weights to obtain normalized features, and input the normalized features into the preset activation function to obtain normalized function features; a third acquisition module, which is used to obtain a gated decision function, calculate the normalized function features using the gated decision function to obtain image refinement features, segment the image refinement features to obtain sub-feature maps, wherein the image refinement features are used to characterize the importance of the normalized function features; a second processing module, which is used to map the initial image features through a fully connected layer to obtain processed channel features, perform dot multiplication on the sub-feature map and the processed channel features, and obtain a layer of feature maps associated with the convolutional network module.

[0081] Optionally, in the detection device of the device provided in the embodiment of the present application, the device also includes: a second acquisition unit, used to obtain the attention weight of each layer of feature maps before inputting M layers of feature maps into the feature fusion module, and obtain M groups of neuron data, wherein each group of neuron data is used to enhance the feature representation of each layer of feature maps, and each group of neuron data includes first neuron data and second neuron data; a calculation unit, used to calculate the mean of each layer of feature maps according to each group of neuron data, to obtain M feature means, and calculate the variance of each layer of feature maps according to each group of neuron data, to obtain M feature variances; a determination unit, used to determine the initial feature weight value of the feature map for a layer of feature maps according to the feature variance and feature mean associated with the feature map, and determine the target feature weight value of the feature map according to the initial feature weight value; a first processing unit, used to process each layer of feature maps according to the target feature weight value of each layer, to obtain M processed feature maps; the input unit 61 includes: inputting the M processed feature maps into the feature fusion module.

[0082] Optionally, in the detection device of the equipment provided in the embodiment of the present application, the input unit 61 includes: a fourth acquisition module, used to obtain a preset activation function, perform spatial dimension grouping processing on M target feature weight values, obtain feature parameters, and calculate the inverse of the feature parameters according to the preset activation function to obtain a feature function value; a calculation module, used to perform dot product calculation on the feature function value and each layer of feature map to obtain M processed feature maps.

[0083] Optionally, in the detection device of the equipment provided in the embodiment of the present application, the device also includes: a third acquisition unit, used to obtain a weight function and a loss function of a preset image detection model before inputting the preprocessed image data into the target detection model, and adjust the loss function using the weight function to obtain an adjusted loss function, wherein the target detection model is obtained by processing the preset image detection model; a second processing unit, used to optimize the preset image detection model using the adjusted loss function to obtain a processed image detection model, and determine the processed image detection model as the target detection model.

[0084] The detection device of the above-mentioned equipment includes a processor and a memory. The above-mentioned first acquisition unit 60, input unit 61, sending unit 62, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0085] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the problems of low detection efficiency and low detection accuracy in related technologies for detecting power equipment and transmission lines can be solved by adjusting kernel parameters.

[0086] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0087] An embodiment of the present invention provides a computer storage medium, which is used to store a program. When the program is running, the device where the computer storage medium is located is controlled to execute a device detection method.

[0088] Figure 7 is a schematic diagram of an electronic device provided according to an embodiment of the present application, such as Figure 7 As shown, an embodiment of the present invention provides an electronic device, the electronic device 70 includes a processor, a memory, and a program stored in the memory and executable on the processor, the processor is used to execute computer-readable instructions, wherein the computer-readable instructions execute a device detection method when executed. The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0089] The present application also provides a computer program product, including a computer program, which implements the steps of a detection method for a device in each embodiment of the present application when the computer program is executed by a processor.

[0090] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0091] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0092] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0094] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0095] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0096] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0097] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0098] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A device detection method, characterized in that: include: Acquire original image data of the target device, and preprocess the original image data to obtain preprocessed image data, wherein the original image data includes image data of the power equipment and image data of the power transmission line; Input the preprocessed image data into a target detection model, and output an image detection result, wherein the image detection result includes a category prediction result and a position prediction result, the category prediction result is used to characterize the fault category of the detection result classification of the preprocessed image data, and the position prediction result is used to characterize the position information of the fault area in the target device; Sending the preprocessed image data and the image detection result to a staff operation terminal; Wherein, the target detection model includes a feature extraction module and a detection head module, the preprocessed image data is input into the target detection model, and the output image detection result includes: inputting the preprocessed image data into the feature extraction module, extracting features in the preprocessed image data through the feature extraction module, and outputting a fused feature map; obtaining N attention dimension parameters, inputting the fused feature map and the N attention dimension parameters into the detection head module, and obtaining the image detection result, wherein the image detection result is obtained by the detection head module processing the fused feature map using the N attention dimension parameters in sequence, N is a positive integer, the target detection model is a model based on the YOLOv7 lightweight detection model, the YOLOv7 lightweight detection model includes CBS+SimAM, Rep+Dyhead+SCConv and SPPCSPC, the detection head module introduces the Dyhead module, and the feature extraction module is an SCConv convolutional network module; The feature extraction module includes a sub-feature extraction module and a feature fusion module, the sub-feature extraction module includes M convolutional network modules, M is a positive integer, the features in the preprocessed image data are extracted by the feature extraction module, and the output fusion feature map includes: inputting the preprocessed image data into the sub-feature extraction module to obtain M layers of feature maps, wherein the M convolutional network modules in the sub-feature extraction module sequentially perform feature extraction processing on the preprocessed image data to obtain the M layers of feature maps; inputting the M layers of feature maps into the feature fusion module, and using the feature fusion module to perform fusion processing on the M layers of feature maps to obtain the fused feature map; wherein each layer of feature maps is obtained by the following method: for a convolutional network module, the preprocessed image data is extracted by the convolutional network module The processed image data is normalized to obtain initial image features; a normalization weight is obtained, and a preset activation function is obtained, the initial image features are processed according to the normalization weight to obtain normalized features, the normalized features are input into the preset activation function to obtain normalized function features; a gated decision function is obtained, the normalized function features are calculated using the gated decision function to obtain image refinement features, the image refinement features are segmented to obtain a sub-feature map, wherein the image refinement features are used to characterize the importance of the normalized function features; the initial image features are mapped through a fully connected layer to obtain processed channel features, the sub-feature map and the processed channel features are multiplied to obtain a layer of feature maps associated with the convolutional network module.

2. The method according to claim 1, characterized in that Before inputting the M-layer feature map into the feature fusion module, the method further includes: Obtaining the attention weight of each layer of feature map to obtain M groups of neuron data, wherein each group of neuron data is used to enhance the feature representation of each layer of feature map, and each group of neuron data includes first neuron data and second neuron data; The mean of each layer of feature maps is calculated according to each group of neuron data to obtain M feature means, and the variance of each layer of feature maps is calculated according to each group of neuron data to obtain M feature variances; For a layer of feature maps, determining an initial feature weight value of the feature map according to a feature variance and a feature mean associated with the feature map, and determining a target feature weight value of the feature map according to the initial feature weight value; Process the feature map of each layer according to the target feature weight value of each layer to obtain M processed feature maps; The inputting of the M layers of feature maps into the feature fusion module includes: inputting the M processed feature maps into the feature fusion module.

3. The method according to claim 2, characterized in that Each layer of feature maps is processed according to the target feature weight value of each layer, and M processed feature maps are obtained, including: Obtain a preset activation function, perform spatial dimension grouping processing on the M target feature weight values ​​to obtain feature parameters, and calculate the inverse of the feature parameters according to the preset activation function to obtain a feature function value; A dot product calculation is performed on the feature function value and each layer of the feature map to obtain the M processed feature maps.

4. The method according to claim 1, characterized in that: Before inputting the preprocessed image data into the target detection model, the method further includes: Obtaining a weight function and a loss function of a preset image detection model, and adjusting the loss function using the weight function to obtain an adjusted loss function, wherein the target detection model is obtained by processing the preset image detection model; The preset image detection model is optimized using the adjusted loss function to obtain a processed image detection model, and the processed image detection model is determined as the target detection model.

5. A detection system for an equipment, characterized in that: include: A drone control system includes a hardware system, a detection module, and a communication module, wherein the hardware system is used to obtain raw image data of a target device, preprocess the raw image data to obtain preprocessed image data, and input the preprocessed image data into the detection module; the detection module is used to input the preprocessed image data into a target detection model when receiving the preprocessed image data, output an image detection result, and send the preprocessed image data and the image detection result to the communication module; the communication module is used to send the preprocessed image data and the image detection result to an operation terminal; Wherein, the target detection model includes a feature extraction module and a detection head module, the preprocessed image data is input into the target detection model, and the output image detection result includes: inputting the preprocessed image data into the feature extraction module, extracting features in the preprocessed image data through the feature extraction module, and outputting a fused feature map; obtaining N attention dimension parameters, inputting the fused feature map and the N attention dimension parameters into the detection head module, and obtaining the image detection result, wherein the image detection result is obtained by the detection head module processing the fused feature map using the N attention dimension parameters in sequence, N is a positive integer, the target detection model is a model based on the YOLOv7 lightweight detection model, the YOLOv7 lightweight detection model includes CBS+SimAM, Rep+Dyhead+SCConv and SPPCSPC, the detection head module introduces the Dyhead module, and the feature extraction module is an SCConv convolutional network module; The feature extraction module includes a sub-feature extraction module and a feature fusion module, the sub-feature extraction module includes M convolutional network modules, M is a positive integer, the feature extraction module is used to extract features from the preprocessed image data, and outputting a fusion feature map includes: inputting the preprocessed image data into the sub-feature extraction module to obtain an M-layer feature map, wherein the M convolutional network modules in the sub-feature extraction module sequentially perform feature extraction processing on the preprocessed image data to obtain the M-layer feature map; inputting the M-layer feature map into the feature fusion module, and using the feature fusion module to perform fusion processing on the M-layer feature map to obtain the fused feature map; Among them, each layer of feature map is obtained in the following manner: for a convolutional network module, the preprocessed image data is normalized by the convolutional network module to obtain initial image features; normalization weights are obtained, and a preset activation function is obtained, the initial image features are processed according to the normalization weights to obtain normalized features, and the normalized features are input into the preset activation function to obtain normalized function features; a gated decision function is obtained, and the normalized function features are calculated by using the gated decision function to obtain image refinement features, and the image refinement features are segmented to obtain sub-feature maps, wherein the image refinement features are used to characterize the importance of the normalization function features; the initial image features are mapped by a fully connected layer to obtain processed channel features, and the sub-feature maps and the processed channel features are multiplied to obtain a layer of feature maps associated with the convolutional network module; The operating end is communicatively connected with the communication module, and is used for generating inspection instruction information upon receiving the pre-processed image data and the image detection result, and sending the inspection instruction information to the operator.

6. A computer storage medium, characterized in that: The computer storage medium is used to store a program, wherein when the program is executed, the device where the computer storage medium is located is controlled to execute the detection method of the device according to any one of claims 1 to 4.

7. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the detection method of the device described in any one of claims 1 to 4.

Images