Transformer substation equipment oil leakage detection method and device, model processing method and device and computer equipment

Abstract

The invention relates to a transformer substation equipment oil leakage detection method and device, a model processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: performing visual angle classification processing on an equipment image of substation equipment through a trained visual angle classification model to obtain an image visual angle type of the equipment image; performing equipment part detection processing on the equipment image through the sub oil leakage detection model corresponding to the image view angle type to obtain equipment part information in the equipment image; obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image; and performing oil leakage detection processing on the target detection image through the oil leakage detection model to obtain an oil leakage detection result of the substation equipment. By adopting the method, the model processing precision of the visual angle classification model and the oil leakage detection model can be improved, so that the oil leakage detection accuracy of the substation equipment is improved.

Description

Technical Field

[0001] The present application relates to the field of smart grid technology, and in particular to a substation equipment oil leakage detection method, model processing method, device, computer equipment, storage medium and computer program product. Background Art

[0002] Substations play a vital role in the power system. They are not only responsible for voltage conversion and power distribution, but also undertake many important functions such as power dispatching, system protection, power factor adjustment, load management, etc. They are the key link in the stable operation of the power system. However, there are many equipment oil leakage problems in all substations. How to effectively solve the problem of oil leakage in substation equipment has far-reaching significance for ensuring the stability and reliability of the power system.

[0003] At present, the technical solutions for substation equipment oil leakage scenarios can be basically summarized as sensor-based detection. Sensor-based detection, such as acoustic detection and chemical detection, uses the physical or chemical properties of equipment when it leaks oil to capture the occurrence of oil leakage defects. This places high demands on the sensor equipment of the substation and is also easily interfered by the external environment, resulting in low accuracy in oil leakage detection of substation equipment. Summary of the invention

[0004] Based on this, it is necessary to provide a substation equipment oil leakage detection method, model processing method, device, computer equipment, computer readable storage medium and computer program product that can improve the accuracy of oil leakage detection of substation equipment in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for detecting oil leakage in substation equipment. The method comprises:

[0006] The trained perspective classification model is used to perform perspective classification processing on the equipment image of the substation equipment to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model;

[0007] By using a sub-oil leakage detection model corresponding to the image viewing angle type, the device image is subjected to device component detection processing to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model;

[0008] Obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image;

[0009] The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

[0010] In one embodiment, the trained view classification model is used to perform view classification processing on the device image of the substation device to obtain the image view type of the device image, including:

[0011] Preprocessing the device image of the substation device to obtain a processed device image;

[0012] Inputting the processed device image into a first target backbone network in the view classification model to obtain a first device feature of the processed device image;

[0013] Inputting the first device feature into a first neck network in the view classification model to obtain a first fused device feature of the first device feature;

[0014] The first fused device feature is input into a classification head in a first target head network of the view classification model to obtain an image view type of the device image.

[0015] In one embodiment, the device image is subjected to device component detection processing by using a sub-oil leakage detection model corresponding to the image viewing angle type to obtain device component information in the device image, including:

[0016] Inputting the device image into the second target backbone network in the sub-oil leakage detection model to obtain a second device feature of the device image;

[0017] Inputting the second device feature into the second neck network in the sub-oil leakage detection model to obtain a second fused device feature of the device image;

[0018] The second fused equipment feature is input into the detection head in the second target head network of the sub-oil leakage detection model to obtain the equipment component information of the equipment image.

[0019] In one embodiment, obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image includes:

[0020] If the equipment component information indicates that there are substation equipment components in the equipment image, the equipment image is cropped according to the component position information in the equipment component information to obtain a component area image in the equipment image, and the component area image is set as the target detection image;

[0021] If the equipment component information indicates that there is no substation equipment component in the equipment image, the equipment image is set as the target detection image.

[0022] In one embodiment, the target detection image is subjected to oil leakage detection processing by the oil leakage detection model to obtain the oil leakage detection result of the substation equipment, including:

[0023] Inputting the target detection image into each of the sub-oil leakage detection models of the oil leakage detection model to obtain the sub-oil leakage detection results output by each of the sub-oil leakage detection models;

[0024] The oil leakage detection result is obtained according to each of the sub-oil leakage detection results.

[0025] In a second aspect, the present application provides a model processing method. The method comprises:

[0026] Through the state space model, the backbone network in the original model is improved to obtain an improved model;

[0027] The head network in the improved model is improved by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head;

[0028] According to the target improvement model, an initial view classification model and a sub-oil leakage detection model to be trained are obtained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained;

[0029] The initial perspective classification model is iteratively trained to obtain a trained perspective classification model, and the initial oil leakage detection model is iteratively trained to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in the first aspect.

[0030] In one embodiment, the backbone network in the original model is improved by using the state space model to obtain an improved model, including:

[0031] Fusing the linear attention model with the state space model to obtain a fused state space model;

[0032] The bottleneck layer in the C2f module of the backbone network is replaced with the fused state space model to obtain the improved model.

[0033] In a third aspect, the present application also provides a substation equipment oil leakage detection device. The device comprises:

[0034] A perspective classification module is used to perform perspective classification processing on the equipment image of the substation equipment through the trained perspective classification model to obtain the image perspective type of the equipment image; the perspective classification model is obtained by improving the backbone network and the head network in the original model;

[0035] a component detection module, configured to perform device component detection processing on the device image through a sub-oil leakage detection model corresponding to the image perspective type, and obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model; the oil leakage detection model is obtained by improving the backbone network and the head network, and the task head in the oil leakage detection model is different from the task head in the perspective classification model;

[0036] An image determination module, used to obtain a target detection image of the substation equipment according to the equipment component information and the equipment image;

[0037] The oil leakage detection module is used to perform oil leakage detection processing on the image to be detected through the oil leakage detection model to obtain the oil leakage detection result of the substation equipment.

[0038] In a fourth aspect, the present application also provides a model processing device. The device comprises:

[0039] The backbone improvement module is used to improve the backbone network in the original model through the state space model to obtain an improved model;

[0040] A head improvement module, used to improve the head network in the improved model by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head;

[0041] A model building module, used to improve the model according to the target, to obtain an initial view classification model and a sub-oil leakage detection model to be trained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained;

[0042] A model training module is used to iteratively train the initial perspective classification model to obtain a trained perspective classification model, and to iteratively train the initial oil leakage detection model to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in the first aspect.

[0043] In a fifth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0044] The trained perspective classification model is used to perform perspective classification processing on the equipment image of the substation equipment to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model;

[0045] By using a sub-oil leakage detection model corresponding to the image viewing angle type, the device image is subjected to device component detection processing to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model;

[0046] Obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image;

[0047] The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

[0048] In a sixth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0049] Through the state space model, the backbone network in the original model is improved to obtain an improved model;

[0050] The head network in the improved model is improved by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head;

[0051] According to the target improvement model, an initial view classification model and a sub-oil leakage detection model to be trained are obtained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained;

[0052] The initial perspective classification model is iteratively trained to obtain a trained perspective classification model, and the initial oil leakage detection model is iteratively trained to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in the first aspect.

[0053] In a seventh aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0054] The trained perspective classification model is used to perform perspective classification processing on the equipment image of the substation equipment to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model;

[0055] By using a sub-oil leakage detection model corresponding to the image viewing angle type, the device image is subjected to device component detection processing to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model;

[0056] Obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image;

[0057] The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

[0058] In an eighth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0059] Through the state space model, the backbone network in the original model is improved to obtain an improved model;

[0060] The head network in the improved model is improved by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head;

[0061] According to the target improvement model, an initial view classification model and a sub-oil leakage detection model to be trained are obtained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained;

[0062] The initial perspective classification model is iteratively trained to obtain a trained perspective classification model, and the initial oil leakage detection model is iteratively trained to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in the first aspect.

[0063] In a ninth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0064] The trained perspective classification model is used to perform perspective classification processing on the equipment image of the substation equipment to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model;

[0065] By using a sub-oil leakage detection model corresponding to the image viewing angle type, the device image is subjected to device component detection processing to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model;

[0066] Obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image;

[0067] The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

[0068] In a tenth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0069] Through the state space model, the backbone network in the original model is improved to obtain an improved model;

[0070] The head network in the improved model is improved by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head;

[0071] According to the target improvement model, an initial view classification model and a sub-oil leakage detection model to be trained are obtained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained;

[0072] The initial perspective classification model is iteratively trained to obtain a trained perspective classification model, and the initial oil leakage detection model is iteratively trained to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in the first aspect.

[0073] The above-mentioned substation equipment oil leakage detection method, model processing method, device, computer equipment, storage medium and computer program product, through the trained perspective classification model, the equipment image of the substation equipment is subjected to perspective classification processing to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improvement model; the first target improvement model is obtained by improving the backbone network and the head network in the original model; through the sub-oil leakage detection model corresponding to the image perspective type, the equipment image is subjected to equipment component detection processing to obtain the equipment component information in the equipment image; the sub-oil leakage detection model belongs to the model branch in the trained oil leakage detection model; according to the equipment component information and the equipment image, the target detection image of the substation equipment is obtained; through the oil leakage detection model, the target detection image is subjected to oil leakage detection processing to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the second target improvement model is subjected to the same model improvement processing as the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model. By adopting this method, the image perspective type of the equipment image of the substation equipment is identified through the perspective classification model, and then the sub-oil leakage detection model corresponding to the image perspective type can be used to detect whether there are equipment components related to the substation equipment in the equipment image, so as to accurately locate the area of ​​the relevant equipment components in the equipment image according to the equipment component information, and then obtain the target detection image, and finally detect whether there is an oil leakage in the target detection image through the oil leakage detection model, and output the oil leakage detection result; by improving the backbone network and the head network of the original model, the nonlinear feature extraction ability of the improved perspective classification model is effectively enhanced, so that the perspective classification model can accurately classify the perspective type of the equipment image of the substation equipment at different perspectives, and then the oil leakage of the equipment is detected through the improved oil leakage detection model. After the improvement, the model processing accuracy of the perspective classification model and the oil leakage detection model is significantly improved, thereby greatly improving the accuracy of oil leakage detection of substation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 An application environment diagram of a method for detecting oil leakage in substation equipment in one embodiment;

[0075] Figure 2 A schematic diagram of a flow chart of a method for detecting oil leakage in substation equipment in one embodiment;

[0076] Figure 3 A schematic flow chart of the steps of performing viewpoint classification processing on a device image of a substation device in one embodiment;

[0077] Figure 4 A schematic diagram of a flow chart of a model processing method in one embodiment;

[0078] Figure 5 Schematic diagram of the structure of a C2f module in a backbone network of an original model in one embodiment;

[0079] Figure 6 A schematic diagram of the model structure of the Mamba model in one embodiment;

[0080] Figure 7 Schematic diagram of a model structure of a linear attention model in one embodiment;

[0081] Figure 8 Schematic diagram of the model structure of the MLLA model in one embodiment;

[0082] Fig. 9 A schematic diagram of a flow chart of a method for detecting oil leakage in substation equipment in another embodiment;

[0083] Fig.10 is a flow chart of a training phase in one embodiment;

[0084] Fig.11 is a flowchart of the reasoning phase in one embodiment;

[0085] Fig.12 It is a structural block diagram of a substation equipment oil leakage detection device in one embodiment;

[0086] Fig.13 is a structural block diagram of a model processing device in an embodiment;

[0087] Fig.14 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0088] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application 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 application and are not used to limit the present application.

[0089] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0090] The substation equipment oil leakage detection method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the image acquisition device 101 communicates with the server 102 through the network. The data storage system can store the data that the service client 102 needs to process. The data storage system can be integrated on the server 102, or it can be placed on the cloud or other network servers. Among them, the image acquisition device 101 can be but not limited to various drones, substation robots, substation cameras and other devices with image acquisition capabilities. The server 102 can be but not limited to various personal computers, laptops, smart phones, tablets, Internet of Things devices and portable wearable devices. The server 102 can also be implemented with an independent server or a server cluster consisting of multiple servers.

[0091] In one embodiment, Figure 2 As shown, a method for detecting oil leakage in substation equipment is provided. Figure 1 The server in the example is used as an example to illustrate, including the following steps:

[0092] Step S201, using the trained perspective classification model, the equipment image of the substation equipment is subjected to perspective classification processing to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model.

[0093] The view classification model refers to a model used to classify the view type of an input image (such as a device image).

[0094] The image perspective type refers to the acquisition perspective type of the image acquisition device corresponding to the device image. For example, the image perspective type can be a drone perspective, a camera perspective, or a robot perspective.

[0095] Among them, the original model can be an open source YoloV8 (You Only Look Once Version 8) model. The YoloV8 model structure is mainly composed of a backbone network (Backbone), a neck network (Neck), and a head network (Head). The backbone network and the head network can be improved to obtain a target improved model; then, the task head of the target head network of the target improved model is processed to respectively construct a first target improved model and a second target improved model; an initial perspective classification model to be trained is constructed based on the first target improved model, and an initial oil leakage detection model to be trained is constructed based on the second target improved model; a trained perspective classification model is obtained by training the initial perspective classification model to be trained, and a trained oil leakage detection model is obtained by training the initial oil leakage detection model.

[0096] Specifically, the image acquisition device is deployed in the substation scene to collect device images of different perspectives of the substation equipment in the substation scene through the image acquisition device, for example, the device image from a high altitude aerial perspective is collected by a drone, the device image from a close front perspective is collected by a substation robot, and the device image from a medium distance side perspective is collected by a substation camera; the image acquisition device transmits the collected device image to the server. The server can first pre-process the device image of the power station equipment to enhance the feature information in the device image, and then the server obtains the processed device image; then the server inputs the processed device image into the trained perspective classification model, so as to perform perspective classification processing on the processed device image through the trained perspective classification model, and output the image perspective type.

[0097] Step S202, performing device component detection processing on the device image through the sub-oil leakage detection model corresponding to the image viewing angle type to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model.

[0098] The equipment component information is used to describe whether there is an equipment component of the substation equipment in the equipment image (as well as the location information of the equipment component).

[0099] Specifically, the trained oil leakage detection model includes a plurality of sub-oil leakage detection models corresponding to different image viewing angle types; the server inputs the equipment image into the sub-oil leakage detection model corresponding to the image viewing angle type obtained in the above step S201, and performs equipment component detection processing on the equipment image through the sub-oil leakage detection model to detect whether the equipment image contains equipment components of the substation equipment, and outputs the equipment component information in the equipment image. The equipment components may be components such as main transformers, capacitor reactors, and mutual inductors.

[0100] Step S203: obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image.

[0101] The target detection image refers to image data used for oil leak detection.

[0102] Specifically, the server determines whether the equipment image contains substation equipment components according to the equipment component information, thereby performing different image processing on the equipment image, and finally the server obtains the target detection image.

[0103] Step S204, performing oil leakage detection processing on the target detection image through the oil leakage detection model to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

[0104] The oil leakage detection result is used to describe the oil leakage defect in the input image (such as the target detection image). For example, the oil leakage detection result can indicate whether there is an oil leakage defect in the image, as well as the specific location information of the existing oil leakage defect.

[0105] Specifically, the server inputs the target detection image into the oil leakage detection model, and performs oil leakage detection processing on the target detection image through the oil leakage detection model to detect the oil leakage defect of the target detection image. Then, the server obtains the oil leakage detection result of the substation equipment based on the oil leakage detection result of the target detection image output by the oil leakage detection model.

[0106] In the above-mentioned substation equipment oil leakage detection method, the image perspective type of the equipment image of the substation equipment is identified by the perspective classification model, and then the sub-oil leakage detection model corresponding to the image perspective type can be used to detect whether there are equipment components related to the substation equipment in the equipment image, so as to accurately locate the area of ​​the relevant equipment components in the equipment image according to the equipment component information, and then obtain the target detection image, and finally detect whether there is an oil leakage in the target detection image through the oil leakage detection model, and output the oil leakage detection result; by improving the backbone network and the head network of the original model, the nonlinear feature extraction ability of the improved perspective classification model is effectively enhanced, so that the perspective classification model can accurately classify the perspective type of the equipment image of the substation equipment at different perspectives, and then the oil leakage of the equipment is detected by the improved oil leakage detection model. After the improvement, the model processing accuracy of the perspective classification model and the oil leakage detection model is significantly improved, thereby greatly improving the accuracy of oil leakage detection of substation equipment.

[0107] In one embodiment, Figure 3 As shown, in the above step S201, the trained view classification model is used to perform view classification processing on the device image of the substation device to obtain the image view type of the device image, which specifically includes the following contents:

[0108] Step S301, pre-processing the equipment image of the substation equipment to obtain a processed equipment image.

[0109] Specifically, the server performs preprocessing such as data enhancement processing on the equipment image of the substation equipment to obtain a processed equipment image.

[0110] Step S302: input the device image of the processed device image into the first target backbone network in the view classification model to obtain the first device feature of the processed device image.

[0111] It should be noted that the first target backbone network is mainly responsible for the feature extraction of the image perspective in the perspective classification model. The network structure of the first target backbone network is the same as that of the second target backbone network, but the network parameters are different. Taking the first target backbone network as an example, the main components of the backbone network of the original model are as follows: C2f module and bottleneck structure; among them, the C2f module is used to fuse some connections and features, and enhance the feature extraction ability of the model while introducing a small number of parameters, and reducing redundant calculations; the bottleneck structure significantly reduces the amount of calculation and the number of parameters by reducing the number of channels in the middle layer, while maintaining the expressiveness of the model. The first target backbone network replaces the bottleneck layer in the C2f module of the backbone network with the MLLA module (a module that integrates the state space model and the linear attention model).

[0112] Specifically, the server inputs the device image of the processed device image into the first target backbone network in the perspective classification model, and performs feature extraction processing of the image perspective on the processed device image through the first target backbone network. The server then obtains the first device feature of the processed device image output by the first target backbone network at different feature extraction stages.

[0113] The first device feature is used to describe the feature of the processed device image in terms of image viewing angle. The first device feature may be a feature map.

[0114] Step S303: input the first device feature into the first neck network in the view classification model to obtain a first fused device feature of the first device feature.

[0115] It should be noted that the first neck network is mainly responsible for the fusion of multi-scale first device features in the perspective classification model. The first neck network has the same network structure as the second neck network, but the network parameters are different. Taking the first neck network as an example, the main components of the first neck network are as follows: SPPF (Spatial Pyramid Pooling Fast) module and PAN (Path Aggregation Network) module; among them, the SPPF module is used for pooling operations of different scales, splicing first device features of different scales together, and improving the classification ability of image perspectives of different sizes; the first neck network can include two PAN modules for path aggregation of features at different levels, and enhance the expression ability of first device features through bottom-up and top-down paths.

[0116] Specifically, the server inputs the first device feature obtained in the above step S302 into the first neck network in the view classification model, performs feature fusion processing on the first device feature through the first neck network, and the server obtains the fused first fused device feature.

[0117] Step S304: input the first fused device feature into the classification head in the first target head network of the perspective classification model to obtain the image perspective type of the device image.

[0118] It should be noted that the first target head network is mainly responsible for the classification task of the view type in the view classification model. The task head in the first target head network is different from the task head in the second target head network, and the rest of the network structures of the two are the same.

[0119] Specifically, the server inputs the first fusion device feature into the first target head network of the perspective classification model. The classification head of the first target head network uses average pooling to classify each first fusion device feature to reduce the dimension of the first fusion device feature, and finally outputs the probability of each perspective type; the server obtains the image perspective type of the device image based on the probability of each perspective type. In addition, the re-parameterized convolution layer in the first target head network enables the perspective classification model to use a serial structure for inference processing during inference, which enhances the model's feature extraction capability while also providing faster speed during inference.

[0120] In this embodiment, the first target backbone network, the first neck network, and the first target head network in the trained perspective classification model cooperate with each other to jointly implement the perspective classification processing of the equipment image of the substation equipment, and finally obtain the image perspective type of the equipment image. The MLLA module in the first target backbone network enhances the feature extraction capability of the image perspective, thereby improving the classification accuracy of the image perspective type of the equipment image by the perspective classification model; the re-parameterized convolution layer in the first target head network enables the perspective classification model to adopt a serial structure for inference processing during inference, which enhances the feature extraction capability of the perspective classification model while further improving the inference speed of the perspective classification model.

[0121] In one embodiment, the above step S202 performs device component detection processing on the device image through the sub-oil leak detection model corresponding to the image perspective type to obtain device component information in the device image, which specifically includes the following contents: inputting the device image into the second target backbone network in the sub-oil leak detection model to obtain the second device feature of the device image; inputting the second device feature into the second neck network in the sub-oil leak detection model to obtain the second fused device feature of the device image; inputting the second fused device feature into the detection head in the second target head network of the sub-oil leak detection model to obtain the device component information of the device image.

[0122] It should be noted that the second target head network is mainly responsible for detection tasks in the sub-oil leakage detection model, such as equipment component detection tasks or oil leakage detection tasks.

[0123] Specifically, the server determines the sub-oil leak detection model corresponding to the image perspective type from the multiple model branches of the missed detection model, and then inputs the equipment image into the sub-oil leak detection model. Through the second target backbone network and the second neck network in the sub-oil leak detection model, the equipment image is sequentially subjected to feature extraction processing and multi-scale feature fusion processing to obtain the second fused equipment feature of the equipment image. For details, refer to the above steps S302 to S303, which will not be repeated here. The server inputs the second fused equipment feature into the detection head in the second target head network of the sub-oil leak detection model; wherein, the detection head is responsible for implementing the equipment component detection task and the oil leak detection task; the detection head includes multiple convolutional layers and deconvolutional layers, which are responsible for predicting the bounding box regression value and confidence of each detection box. Finally, the sub-oil leak detection model outputs the detection result; the server obtains the equipment component information of the equipment image according to the detection result.

[0124] In this embodiment, the second target backbone network, the second neck network, and the second target head network in the sub-oil leakage detection model cooperate with each other to jointly realize the detection and processing of the equipment components of the equipment image of the substation equipment, and finally obtain the equipment component information of the equipment image, and enhance the feature extraction capability of the equipment components through the MLLA module in the second target backbone network, thereby improving the detection accuracy of the sub-oil leakage detection model for the equipment components of the equipment image; the heavy parameterized convolution layer in the second target head network enables the sub-oil leakage detection model to adopt a serial structure for reasoning processing during reasoning, which enhances the feature extraction capability of the sub-oil leakage detection model while further improving the reasoning speed of the sub-oil leakage detection model.

[0125] In one embodiment, the above step S203 obtains a target detection image of the substation equipment based on the equipment component information and the equipment image, and specifically includes the following contents: if the equipment component information indicates that there are substation equipment components in the equipment image, then the equipment image is cropped according to the component position information in the equipment component information to obtain a component area image in the equipment image, and the component area image is set as the target detection image; if the equipment component information indicates that there are no substation equipment components in the equipment image, then the equipment image is set as the target detection image.

[0126] Specifically, if the equipment component information indicates that there are substation equipment components in the equipment image, the server will crop the equipment component area in the equipment image according to the component position information in the equipment component information, and then the server will crop the component area image in the equipment image and set the component area image as the target detection image. If the equipment component information indicates that there are no substation equipment components in the equipment image, the server does not need to crop the equipment image, and can set the equipment image as the target detection image, so that the oil leakage detection model can analyze whether there is an oil leakage defect in the substation equipment in combination with the overall image information of the equipment image.

[0127] In this embodiment, it is determined based on the equipment component information whether there are substation equipment components in the equipment image, and then it is decided whether to crop the equipment image. If there are substation equipment components in the equipment image, the oil leakage detection model can use the cropped component area image to perform oil leakage detection, reducing the impact of irrelevant background information in the equipment image on oil leakage detection, thereby improving the accuracy of oil leakage detection by the oil leakage detection model; if there are no substation equipment components in the equipment image, the oil leakage detection model can combine the overall image information of the equipment image to comprehensively analyze whether the substation equipment has oil leakage defects, which helps to improve the oil leakage detection accuracy of the equipment image.

[0128] In one embodiment, the above step S204, through the oil leakage detection model, performs oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment, which specifically includes the following contents: inputting the target detection image into each sub-oil leakage detection model of the oil leakage detection model to obtain the sub-oil leakage detection result output by each sub-oil leakage detection model; and obtaining the oil leakage detection result according to each sub-oil leakage detection result.

[0129] Specifically, the server inputs the target detection image into each sub-oil leakage detection model of the oil leakage detection model, and performs oil leakage detection processing on the target detection image through each sub-oil leakage detection model to obtain the sub-oil leakage detection results output by each sub-oil leakage detection model. The server obtains the oil leakage detection result by combining each sub-oil leakage detection result, for example, taking the average value of each sub-oil leakage detection result as the oil leakage detection result, or taking the sub-oil leakage detection result with the highest frequency among all sub-oil leakage detection results as the oil leakage detection result, and so on.

[0130] In this embodiment, the oil leakage detection result is comprehensively determined by the sub-oil leakage detection results output by each sub-oil leakage detection model of the oil leakage detection model, thereby avoiding the error in the final detection result caused by the abnormal output result of a single model, and effectively improving the accuracy of the oil leakage detection result obtained by the final processing.

[0131] In one embodiment, Figure 4 As shown, a model processing method is provided. In this embodiment, the method is applied to Figure 1 The server in the example is illustrated, and is implemented through the interaction between the image acquisition device and the server. In this embodiment, the method includes the following steps:

[0132] Step S401, improving the backbone network in the original model through the state space model to obtain an improved model.

[0133] Among them, the state-space model (i.e., Mamba model) is a deep learning network architecture, which is usually used for computer vision tasks; the state-space model aims to improve the performance and efficiency of the model by introducing residual connections and bottleneck structures.

[0134] Specifically, the server can optimize the bottleneck layer in the C2f module of the backbone network through the state space model and the linear attention model, and then the server obtains the optimized target backbone network (in order to distinguish the original model from the model with the improved backbone network, the model with the optimized backbone network can be called the improved model), so that the optimized improved model has stronger feature extraction capabilities.

[0135] Among them, the Linear Attention Transformer is an improved Transformer model that aims to solve the computational complexity problem of the traditional Transformer model when processing long sequences. The computational complexity of the attention mechanism of the traditional Transformer model is o(N 2), where N is the sequence length, and the linear attention model reduces the computational complexity to o(N) by introducing an attention mechanism with linear complexity. Therefore, the improved model that introduces the linear attention model not only improves the feature extraction capability, but also reduces the computational complexity of the feature processing process, thereby improving the inference speed and model processing efficiency of the improved model.

[0136] Step S402, by re-parameterizing the convolutional layer, improving the head network in the improved model to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head.

[0137] Specifically, the server introduces the re-parameterized convolution layer into the head network of the improved model, and the server obtains the optimized target head network (in order to distinguish the improved model from the model with the improved head network, the model with the optimized head network can be called the target improved model). The target improved model can adopt a multi-branch auxiliary training structure during training and convert it into a serial structure reasoning during reasoning. For example, the target improved model can use multi-branch parallel convolution layers for model training processing during training to accurately adjust the model parameters of each branch model, and connect the multi-branch convolution layers in series into one convolution layer during testing, thereby reducing the number of parameters and calculations, and providing a faster calculation speed for the target improved model in the reasoning process.

[0138] Step S403, improving the model according to the target, obtaining the initial view classification model and the sub-oil leakage detection model to be trained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; multiple sub-oil leakage detection models are used to form the initial oil leakage detection model to be trained.

[0139] Specifically, the task head in the target head network of the target improvement model includes a classification head and a detection head. The server can delete the detection head in the target head network, and then the server obtains the initial perspective classification model to be trained. Similarly, the server can also delete the classification head in the target head network, and then the server obtains the sub-oil leakage detection model to be trained. Then the server uses multiple sub-oil leakage detection models to construct an initial oil leakage detection model with a multi-branch structure.

[0140] Step S404, iteratively train the initial perspective classification model to obtain a trained perspective classification model, and iteratively train the initial oil leakage detection model to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the above-mentioned substation equipment oil leakage detection method.

[0141] Specifically, the server performs perspective annotation processing on the historical equipment images to obtain the image perspective type labels of the historical equipment images; the historical equipment images are input into the initial perspective classification model to obtain the historical image perspective types of the historical equipment images; according to the difference between the historical image perspective types and the image perspective labels of the historical equipment images, the initial perspective classification model is iteratively trained to obtain the trained perspective classification model. The historical equipment images refer to the historical oil leakage defect images of the substation equipment.

[0142] Furthermore, the server performs oil leak location annotation processing on the historical equipment images to obtain the oil leak location labels of the historical equipment images; the historical equipment images are input into the initial oil leak detection model to obtain the predicted oil leak location of the historical equipment images; according to the difference between the predicted oil leak location and the oil leak location label, the initial oil leak detection model is iteratively trained to obtain a trained oil leak detection model.

[0143] In the above model processing method, by improving the backbone network and head network of the original model, the nonlinear feature extraction ability of the improved perspective classification model is effectively enhanced, so that the perspective classification model can accurately classify the perspective type of the equipment images of the substation equipment at different perspectives, and then detect the oil leakage of the equipment through the improved oil leakage detection model. After the improvement, the model processing accuracy of the perspective classification model and the oil leakage detection model is significantly improved, thereby greatly improving the oil leakage detection accuracy of the substation equipment.

[0144] In one embodiment, the above step S401 improves the backbone network in the original model through the state space model to obtain an improved model, which specifically includes the following contents: fusing the linear attention model with the state space model to obtain a fused state space model; replacing the bottleneck layer in the C2f module of the backbone network with the fused state space model to obtain an improved model.

[0145] Among them, the C2f (Channel-to-Feature) module is a structure used to improve the performance of the backbone network and is often used in deep learning models in computer vision tasks. Its main purpose is to interact information between different channels to improve the expressiveness of features and the performance of the network.

[0146] Specifically, the structure diagram of the C2f module in the backbone network of the original model is as follows: Figure 5As shown in Figure 1, the C2f module includes CBS (Convolution-BatchNorm-Swish), split (an operation that divides the input feature map into multiple parts or channels), and a bottleneck layer. The parameters S and K of CBS are both hyperparameters. The server fuses the Mamba model and the linear attention model to obtain a fused state space model (also known as the MLLA model). The MLLA model mainly integrates the two key technologies of "forget gate" and "module design" in the Mamba model.

[0147] The forget gate in the Mamba model provides local deviation and position information, and its elements are strictly controlled between 0 and 1. When the model receives input, it will gradually and continuously decay the previous hidden state, so that the model is sensitive to the order of the input sequence. Mamba's module design improves performance by replacing the attention submodule with a linear attention model while maintaining similar floating-point calculations. The model structure of the Mamba model is as follows: Figure 6 As shown, the model structure of the linear attention model is as follows Figure 7 As shown, the model structure of the MLLA model is as follows Figure 8 As shown. Among them, MLP (Multi Layer Perceptron) is used to perform nonlinear transformation and feature extraction on input data; Norm (Normalization) is used to adjust the distribution of features, thereby improving the convergence speed and performance of the model; Linear (Linear Transformation) is used to perform linear transformation on input data; Linear Attention is an improvement on the attention mechanism, which aims to reduce the computational complexity by linearizing the calculation process; Conv (Convolution) is used to extract local features of input data; LA Block (Linear Attention Block) is the basic structural unit for realizing linear attention; SSM (State Space Model) is a model used to describe the behavior of dynamic systems.

[0148] In this embodiment, a fused state-space model is obtained by combining the Transformer architecture of the Mamba model and the linear attention mechanism, and the fused state-space model is used to replace the bottleneck layer in the C2f module of the backbone network, so that the improved model can take advantage of the efficient structure of the Mamba model and the fast calculation speed of the linear attention model, maintain efficient calculation and improve memory usage efficiency when processing long sequence data, thereby greatly improving the computational efficiency of the view classification model and the oil leak detection model.

[0149] In one embodiment, Fig. 9 As shown, another substation equipment oil leakage detection method is provided, and the method is applied to Figure 1 The server in the example is used as an example to illustrate, including the following steps:

[0150] Step S901: pre-process the equipment image of the substation equipment to obtain a processed equipment image.

[0151] Step S902: Performing perspective classification processing on the processed device image through the trained perspective classification model to obtain the image perspective type.

[0152] Among them, the view classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model.

[0153] Step S903: Perform device component detection processing on the device image through the sub-oil leakage detection model corresponding to the image viewing angle type to obtain device component information in the device image.

[0154] Among them, the sub-oil leak detection model belongs to the model branch in the trained oil leak detection model.

[0155] Step S904: if the equipment component information indicates that there are substation equipment components in the equipment image, the equipment image is cropped according to the component position information in the equipment component information to obtain a component area image in the equipment image, and the component area image is set as the target detection image.

[0156] Step S905: If the equipment component information indicates that there is no substation equipment component in the equipment image, the equipment image is set as a target detection image.

[0157] Step S906: input the target detection image into each sub-oil leakage detection model of the oil leakage detection model to obtain the sub-oil leakage detection results output by each sub-oil leakage detection model.

[0158] Step S907, obtaining the oil leakage detection result according to each sub-oil leakage detection result.

[0159] The above-mentioned substation equipment oil leakage detection method can achieve the following beneficial effects: the image perspective type of the equipment image of the substation equipment is identified through the perspective classification model, and then the sub-oil leakage detection model corresponding to the image perspective type can be used to detect whether there are equipment components related to the substation equipment in the equipment image, so as to accurately locate the area of ​​the relevant equipment components in the equipment image according to the equipment component information, and then obtain the target detection image, and finally detect whether there is an oil leakage in the target detection image through the oil leakage detection model, and output the oil leakage detection result; by improving the backbone network and the head network of the original model, the nonlinear feature extraction ability of the improved perspective classification model is effectively enhanced, so that the perspective classification model can accurately classify the perspective type of the equipment image of the substation equipment at different perspectives, and then the equipment oil leakage is detected through the improved oil leakage detection model. After the improvement, the model processing accuracy of the perspective classification model and the oil leakage detection model is significantly improved, thereby greatly improving the accuracy of oil leakage detection of substation equipment.

[0160] In order to more clearly illustrate the substation equipment oil leakage detection method provided by the embodiment of the present disclosure, the above-mentioned substation equipment oil leakage detection method is specifically described below with a specific embodiment. Fig.10 and Fig.11 Another substation equipment oil leakage detection method is provided, which can be applied to Figure 1 The server in the , specifically includes the following contents:

[0161] Fig.10 It is a flowchart of the training phase; where:

[0162] S1001: Collect historical oil leakage defect images of substations from multiple perspectives through image acquisition equipment such as drones, substation robots, and substation cameras.

[0163] S1002: The server performs data cleaning and data enhancement processing on the historical oil leak defect images, uses Labelimg to mark the oil leak locations in the historical oil leak defect data, and simultaneously marks the view category of the historical oil leak defect images, and then uses the processed image data to construct training and test data sets.

[0164] S1003: Based on yolov8, the Backbone and Head models are improved to build a view classification model and an oil leakage defect detection model, and the collected and labeled training data are sent to the view classification model and the oil leakage defect detection model for model training. The server inputs the training set into batches into the improved Backbone network to extract feature data.

[0165] S1004: The server inputs the feature data obtained in S1003 into the Neck network for multi-scale fusion to obtain fused feature data.

[0166] S1005: The server inputs the fused feature data obtained in S1004 into the re-parameterized improved Head network for task decoupling processing and outputs the oil leak detection result.

[0167] Fig.11 Figure 1 is a flowchart of the inference phase;

[0168] S1101: collecting device images of substation equipment in the substation in real time, and performing image preprocessing on the collected device images to obtain processed device images.

[0169] S1102: Input the processed device image into a trained view classification model to classify the image view.

[0170] S1103: According to the classification result in S1102, the processed image is selected to be input into the corresponding oil leakage detection model branch, and the sub-oil leakage detection model of the branch is used to detect whether the processed image contains common substation equipment components, such as main transformers, capacitor reactors, mutual inductors, etc.

[0171] S1104, including branch 1 and branch 2, wherein:

[0172] Branch 1: If there are parts, the part area in the processed image is cropped out and input into the oil leakage detection model to detect oil leakage defects.

[0173] Branch 2: If there are no parts, the processed image is directly input into the oil leakage detection model to detect the oil leakage defect.

[0174] In this embodiment, the substation data from different sources and different perspectives have good detection compatibility, providing multiple options for substation equipment images and saving operation and maintenance costs. In addition, the improved Backbone and improved Head models based on YoloV8 improve the nonlinear feature extraction capabilities of the perspective classification model and oil leakage detection, and to a certain extent also improve the inference speed during model deployment.

[0175] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0176] Based on the same inventive concept, the embodiment of the present application also provides a substation equipment oil leakage detection device for implementing the substation equipment oil leakage detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more substation equipment oil leakage detection device embodiments provided below can refer to the above limitations on the substation equipment oil leakage detection method, and will not be repeated here.

[0177] In one embodiment, Fig.12 As shown, a substation equipment oil leakage detection device 1200 is provided, comprising: a view classification module 1201, a component detection module 1202, an image determination module 1203 and an oil leakage detection module 1204, wherein:

[0178] The perspective classification module 1201 is used to perform perspective classification processing on the equipment image of the substation equipment through the trained perspective classification model to obtain the image perspective type of the equipment image; the perspective classification model is obtained by improving the backbone network and head network in the original model.

[0179] The component detection module 1202 is used to perform device component detection processing on the device image through the sub-oil leakage detection model corresponding to the image perspective type, and obtain the device component information in the device image; the sub-oil leakage detection model belongs to the model branch in the trained oil leakage detection model; the oil leakage detection model is obtained by improving the backbone network and the head network, and the task head in the oil leakage detection model is different from the task head in the perspective classification model.

[0180] The image determination module 1203 is used to obtain a target detection image of the substation equipment according to the equipment component information and the equipment image.

[0181] The oil leakage detection module 1204 is used to perform oil leakage detection processing on the image to be detected through the oil leakage detection model to obtain the oil leakage detection result of the substation equipment.

[0182] In one embodiment, the view classification module 1201 is further used to pre-process the device image of the substation device to obtain a processed device image; input the processed device image to the first target backbone network in the view classification model to obtain a first device feature of the processed device image;

[0183] The first device feature is input into the first neck network in the perspective classification model to obtain the first fused device feature of the first device feature; the first fused device feature is input into the classification head in the first target head network of the perspective classification model to obtain the image perspective type of the device image.

[0184] In one embodiment, the component detection module 1202 is also used to input the device image into the second target backbone network in the sub-oil leak detection model to obtain the second device feature of the device image; input the second device feature into the second neck network in the sub-oil leak detection model to obtain the second fused device feature of the device image; input the second fused device feature into the detection head in the second target head network of the sub-oil leak detection model to obtain the device component information of the device image.

[0185] In one embodiment, the image determination module 1203 is also used to crop the device image according to the component position information in the device component information to obtain a component area image in the device image, and set the component area image as the target detection image if the device component information indicates that there are substation equipment components in the device image, and set the device image as the target detection image if the device component information indicates that there are no substation equipment components in the device image.

[0186] In one embodiment, the oil leakage detection module 1204 is further used to input the target detection image into each sub-oil leakage detection model of the oil leakage detection model to obtain the sub-oil leakage detection results output by each sub-oil leakage detection model; and obtain the oil leakage detection result according to each sub-oil leakage detection result.

[0187] In one embodiment, Fig.13 As shown, a substation equipment oil leakage detection device 1300 is provided, including: a backbone improvement module 1301, a head improvement module 1302, a model building module 1303 and a model training module 1304, wherein:

[0188] The backbone improvement module 1301 is used to improve the backbone network in the original model through the state space model to obtain an improved model.

[0189] The head improvement module 1302 is used to improve the head network in the improved model by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head.

[0190] The model building module 1303 is used to improve the model according to the target to obtain the initial perspective classification model and the sub-oil leakage detection model to be trained; the task head in the initial perspective classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; multiple sub-oil leakage detection models are used to form the initial oil leakage detection model to be trained.

[0191] The model training module 1304 is used to iteratively train the initial perspective classification model to obtain a trained perspective classification model, and to iteratively train the initial oil leakage detection model to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the above-mentioned substation equipment oil leakage detection method.

[0192] In one embodiment, the backbone improvement module 1301 is also used to fuse the linear attention model with the state space model to obtain a fused state space model; and replace the bottleneck layer in the C2f module of the backbone network with the fused state space model to obtain an improved model.

[0193] Each module in the above-mentioned substation equipment oil leakage detection device and model processing can be fully or partially implemented by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0194] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.13As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as device images, image viewing angle types, device part information, and oil leakage detection results. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for detecting oil leakage in substation equipment is implemented.

[0195] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0196] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0197] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0198] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0199] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0200] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0201] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for detecting oil leakage in substation equipment, characterized in that: The method comprises: The trained perspective classification model is used to perform perspective classification processing on the equipment image of the substation equipment to obtain the image perspective type of the equipment image; the perspective classification model is constructed based on the first target improved model; the first target improved model is obtained by improving the backbone network and the head network in the original model; By using a sub-oil leakage detection model corresponding to the image viewing angle type, the device image is subjected to device component detection processing to obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model; Obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image; The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment; the oil leakage detection model is constructed based on the second target improvement model; the model improvement processing of the second target improvement model is the same as that of the first target improvement model, and the task head in the target head network of the second target improvement model is different from the task head in the target head network of the first target improvement model.

2. The method according to claim 1, characterized in that: The trained view classification model performs view classification processing on the device image of the substation device to obtain the image view type of the device image, including: Preprocessing the device image of the substation device to obtain a processed device image; Inputting the processed device image into a first target backbone network in the view classification model to obtain a first device feature of the processed device image; Inputting the first device feature into a first neck network in the view classification model to obtain a first fused device feature of the first device feature; The first fused device feature is input into a classification head in a first target head network of the view classification model to obtain an image view type of the device image.

3. The method according to claim 1, characterized in that The performing device component detection processing on the device image by using the sub-oil leakage detection model corresponding to the image viewing angle type to obtain device component information in the device image includes: Inputting the device image into the second target backbone network in the sub-oil leakage detection model to obtain a second device feature of the device image; Inputting the second device feature into the second neck network in the sub-oil leakage detection model to obtain a second fused device feature of the device image; The second fused equipment feature is input into the detection head in the second target head network of the sub-oil leakage detection model to obtain the equipment component information of the equipment image.

4. The method according to claim 3, characterized in that The step of obtaining a target detection image of the substation equipment according to the equipment component information and the equipment image includes: If the equipment component information indicates that there are substation equipment components in the equipment image, the equipment image is cropped according to the component position information in the equipment component information to obtain a component area image in the equipment image, and the component area image is set as the target detection image; If the equipment component information indicates that there is no substation equipment component in the equipment image, the equipment image is set as the target detection image.

5. The method according to claim 1, characterized in that The oil leakage detection model is used to perform oil leakage detection processing on the target detection image to obtain the oil leakage detection result of the substation equipment, including: Inputting the target detection image into each of the sub-oil leakage detection models of the oil leakage detection model to obtain the sub-oil leakage detection results output by each of the sub-oil leakage detection models; The oil leakage detection result is obtained according to each of the sub-oil leakage detection results.

6. A model processing method, characterized in that: The method comprises: Through the state space model, the backbone network in the original model is improved to obtain an improved model; The head network in the improved model is improved by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head; According to the target improvement model, an initial view classification model and a sub-oil leakage detection model to be trained are obtained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained; Iteratively train the initial perspective classification model to obtain a trained perspective classification model, and iteratively train the initial oil leakage detection model to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in any one of claims 1 to 3.

7. The method according to claim 6, characterized in that The backbone network in the original model is improved by the state space model to obtain an improved model, including: Fusing the linear attention model with the state space model to obtain a fused state space model; The bottleneck layer in the C2f module of the backbone network is replaced with the fused state space model to obtain the improved model.

8. A substation equipment oil leakage detection device, characterized in that: The device comprises: A perspective classification module is used to perform perspective classification processing on the equipment image of the substation equipment through the trained perspective classification model to obtain the image perspective type of the equipment image; the perspective classification model is obtained by improving the backbone network and the head network in the original model; a component detection module, configured to perform device component detection processing on the device image through a sub-oil leakage detection model corresponding to the image perspective type, and obtain device component information in the device image; the sub-oil leakage detection model belongs to a model branch in the trained oil leakage detection model; the oil leakage detection model is obtained by improving the backbone network and the head network, and the task head in the oil leakage detection model is different from the task head in the perspective classification model; An image determination module, used to obtain a target detection image of the substation equipment according to the equipment component information and the equipment image; The oil leakage detection module is used to perform oil leakage detection processing on the image to be detected through the oil leakage detection model to obtain the oil leakage detection result of the substation equipment.

9. A model processing device, characterized in that: The device comprises: The backbone improvement module is used to improve the backbone network in the original model through the state space model to obtain an improved model; A head improvement module, used to improve the head network in the improved model by re-parameterizing the convolution layer to obtain a target improved model; the task head in the target head network of the target improved model includes a classification head and a detection head; A model building module, used to improve the model according to the target, to obtain an initial view classification model and a sub-oil leakage detection model to be trained; the task head in the initial view classification model is the classification head, and the task head in the sub-oil leakage detection model is the detection head; a plurality of the sub-oil leakage detection models are used to form an initial oil leakage detection model to be trained; A model training module, used to iteratively train the initial perspective classification model to obtain a trained perspective classification model, and to iteratively train the initial oil leakage detection model to obtain a trained oil leakage detection model; the trained perspective classification model and the trained oil leakage detection model are used to implement the steps of the method described in any one of claims 1 to 5.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5, or 6 to 7 are implemented.

Images