Isolation switch on-off state detection method and device based on deep learning

Abstract

The invention provides an isolating switch on-off state detection method and device based on deep learning. The method comprises the following steps: acquiring an isolating switch image; detecting the isolation switch image by using a pre-trained isolation switch detection network to obtain a detection result of the opening and closing state of the isolation switch; wherein the isolation switch detection network performs feature extraction by using the first feature extraction and processing module and the second feature extraction and processing module; the first feature extraction and processing module is used for performing preliminary feature extraction on an input isolating switch image by adopting a multi-branch convolution structure so as to obtain preliminary features on the basis of maintaining the spatial resolution of the image; and the second feature extraction and processing module is used for carrying out deep feature mining on the preliminary features by adopting a deep separable convolutional layer and a dynamic convolutional layer so as to obtain key features of the disconnecting switch image. According to the detection network, through multi-branch convolution and deep feature mining, the detection precision can be improved, and the real-time detection requirement is met.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system equipment status detection, and specifically relates to a method and device for detecting the opening and closing status of an isolating switch based on deep learning. Background Art

[0002] In power systems, the stable operation of various devices is crucial to ensuring the reliability of power supply. As one of these key devices, disconnectors fulfill the crucial function of achieving electrical isolation in power lines, providing safety for equipment maintenance and troubleshooting. Their operating status is directly related to the overall stability and safety of the power system.

[0003] The continuous expansion of power systems and the increasing demand for intelligent systems are placing higher demands on the accuracy, efficiency, and real-time performance of disconnector status detection. Traditional methods relying on manual inspections or simple sensor monitoring are increasingly unable to meet these requirements. Consequently, methods such as visual inspection based on sensors and image processing have emerged. In recent years, detection methods based on deep learning have become a hot topic in research and application.

[0004] Current sensor-based detection methods suffer from high installation and maintenance costs and are easily affected by environmental factors, leading to inaccurate detection results. While image processing-based visual detection methods reduce costs, they place stringent demands on image quality. Traditional image processing methods struggle to capture subtle changes in the state of disconnectors, resulting in limited detection accuracy. Deep learning-based detection methods, in particular, generally rely on single-scale convolution operations, failing to simultaneously account for both local details and global structural features of disconnectors. They also lack key feature enhancement mechanisms, resulting in poor detection performance in complex scenarios. Furthermore, their complex network structure and high computational overhead make them unable to meet the practical needs of real-time detection. Summary of the Invention

[0005] In view of this, the present invention provides a method and device for detecting the opening and closing status of an isolating switch based on deep learning, aiming to solve at least one of the problems existing in the traditional isolating switch opening and closing status detection method, such as low detection accuracy, poor environmental adaptability, high computational complexity and poor real-time performance.

[0006] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for detecting the opening and closing status of an isolating switch based on deep learning, comprising the following steps:

[0008] Get the disconnector image;

[0009] Use the pre-trained disconnector detection network to detect the disconnector image and obtain the detection result of the disconnector open and closed state;

[0010] Among them, the isolating switch detection network uses the first feature extraction and processing module and the second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

[0011] Furthermore, the disconnector detection network includes:

[0012] Backbone feature extraction network, multi-scale feature processing module, attention enhancement module and state detection module;

[0013] The backbone feature extraction network includes a first feature extraction and processing module and a second feature extraction and processing module;

[0014] The first feature extraction and processing module is used to perform preliminary feature extraction on the input disconnector image. It extracts features from different angles through different convolutional layers in the multi-branch convolutional structure and then concatenates the results of each branch to obtain a preliminary feature map while maintaining the spatial resolution of the image.

[0015] The second feature extraction and processing module is used to perform deep feature mining on the preliminary features using a depthwise separable convolution layer and a dynamic convolution layer, thereby obtaining a key feature map of the disconnector image;

[0016] The multi-scale feature processing module is used to extract and fuse multi-scale features of key features to obtain a multi-scale feature map;

[0017] The attention enhancement module is used to calculate and adjust the importance weights in the multi-scale feature map based on the EA attention mechanism to obtain a feature map that highlights key features;

[0018] The state detection module is used to determine the final disconnector opening and closing state detection results based on the feature map that highlights the key features.

[0019] Furthermore, the first feature extraction and processing module is an MSCA-Conv module, and the image processing process of the MSCA-Conv module includes:

[0020] A set of Conv modules are used to perform convolution operations, dilated convolution operations, and depthwise separable convolution operations on the input isolation switch feature map, thereby obtaining a second isolation switch feature map, a third isolation switch feature map, and a fourth isolation switch feature map;

[0021] Normalizing the second isolating switch characteristic graph, the third isolating switch characteristic graph, and the fourth isolating switch characteristic graph through the BN layer;

[0022] pass The standardized isolating switch feature graphs are spliced ​​at a layer to obtain a fifth isolating switch feature graph;

[0023] Performing a convolution operation on the fifth isolating switch feature map through a Conv module to obtain a sixth isolating switch feature map;

[0024] Processing the sixth isolating switch feature map through a global average pooling module to obtain a seventh isolating switch feature map;

[0025] Generate attention weights for the seventh isolation switch feature map through a fully connected layer and normalize it using a Sigmoid function to obtain the eighth isolation switch feature map;

[0026] A point product operation is performed on the eighth isolating switch feature map and the input isolating switch feature map to obtain a ninth isolating switch feature map, and the ninth isolating switch feature map is used as the final output of the MSCA-Conv module.

[0027] Furthermore, the transformation formulas of the fifth isolating switch characteristic diagram and the ninth isolating switch characteristic diagram are as follows:

[0028]

[0029]

[0030] Where, represents the characteristic diagram of the fifth disconnector, Represents a splicing operation, Indicates standardized processing; 、 and They represent convolution operations with a kernel size of 1 and a stride of 1, dilated convolution operations with a kernel size of 3 and a stride of 1, and depthwise separable convolution operations with a kernel size of 5 and a stride of 1. Characteristic diagram of the disconnector representing the input; Represents the characteristic diagram of the ninth disconnector.

[0031] Furthermore, the second feature extraction and processing module is a DASC-Conv module. The image processing process of the DASC-Conv module includes:

[0032] Performing a convolution operation on the input isolation switch feature map through the first Conv module to obtain a tenth isolation switch feature map;

[0033] Performing a convolution operation on the tenth isolating switch feature map through the second Conv module to obtain an eleventh isolating switch feature map;

[0034] Performing maximum pooling processing on the eleventh disconnector feature map through a maximum pooling module to obtain a twelfth disconnector feature map;

[0035] Performing bilinear interpolation processing on the twelfth disconnector characteristic graph through a bilinear interpolation module to obtain a thirteenth disconnector characteristic graph;

[0036] The thirteenth isolating switch feature map is activated by the Swish function module, and a dot product operation is performed on the tenth isolating switch feature map to obtain the fourteenth isolating switch feature map, and the fourteenth isolating switch feature map is used as the final output of the DASC-Conv module.

[0037] Furthermore, the transformation formulas of the thirteenth isolating switch characteristic diagram and the fourteenth isolating switch characteristic diagram are as follows:

[0038]

[0039]

[0040] Where, The characteristic diagram of the thirteenth disconnector is shown. represents a bilinear interpolation operation, represents the maximum pooling operation, Represents a convolution operation with a kernel size of 3 and a stride of 1. Represents a convolution operation with a kernel size of 1 and a stride of 1. Characteristic diagram of the disconnector representing the input; express activation function, Represents the characteristic diagram of the tenth disconnector, Represents the characteristic diagram of the fourteenth disconnector.

[0041] Furthermore, the multi-scale feature processing module is the MSFE-Conv module. The image processing process of the MSFE-Conv module includes:

[0042] Performing convolution operations on the input isolation switch feature map through two Conv modules of different sizes to obtain a fifteenth isolation switch feature map and a sixteenth isolation switch feature map;

[0043] Performing a channel splicing operation on the fifteenth isolating switch characteristic diagram and the sixteenth isolating switch characteristic diagram through a channel splicing module to obtain a seventeenth isolating switch characteristic diagram;

[0044] Performing a global average pooling operation on the seventeenth isolating switch feature map through a global average pooling module to obtain an eighteenth isolating switch feature map;

[0045] Activate the eighteenth disconnector feature map through the GELU activation function to obtain a nineteenth disconnector feature map;

[0046] Performing a convolution operation on the nineteenth isolating switch feature map through a Conv module to obtain a twentieth isolating switch feature map;

[0047] A dot product operation is performed on the twentieth isolating switch feature map and the input isolating switch feature map to obtain a twenty-first isolating switch feature map, and the twenty-first isolating switch feature map is used as the final output of the MSFE-Conv module.

[0048] Furthermore, the transformation formulas of the seventeenth isolating switch characteristic diagram, the twentieth isolating switch characteristic diagram, and the twenty-first isolating switch characteristic diagram are as follows:

[0049]

[0050]

[0051]

[0052] Where, The characteristic diagram of the seventeenth disconnector is shown. Indicates channel splicing operation, Represents a convolution operation with a kernel size of 1 and a stride of 1. Represents a convolution operation with a kernel size of 3 and a stride of 1. Characteristic diagram of the disconnector representing the input; The characteristic diagram of the twentieth disconnector is shown. express activation function, represents the global average pooling operation, Represents the characteristic diagram of the twenty-first disconnector.

[0053] Furthermore, the state detection module is a Detect module, and the characteristic graph highlighting the key features is recorded as the twenty-second disconnector characteristic graph. Based on the characteristic graph highlighting the key features, the final disconnector opening and closing state detection result is determined, including:

[0054] Using a global average pooling module, averaging the feature values ​​of each channel in the twenty-second isolation switch feature map along the spatial dimension to generate a twenty-third isolation switch feature map;

[0055] Pass the twenty-third isolation switch feature map through a fully connected layer, map the channel dimension to the number of target categories, and output a feature vector;

[0056] Normalize the feature vector into a probability distribution through the Softmax function ,in ,in, represents the probability that the disconnector is in the open state, Represents the probability that the disconnector is in the closed state.

[0057] In a second aspect, the present invention provides a device for detecting the opening and closing status of an isolating switch based on deep learning, comprising:

[0058] An image acquisition module, used for acquiring an image of the disconnector;

[0059] A detection module is used to detect the disconnector image using a pre-trained disconnector detection network to obtain a detection result of the disconnector opening and closing state;

[0060] Among them, the isolating switch detection network uses the first feature extraction and processing module and the second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

[0061] In summary, the present invention provides a method and device for detecting the opening and closing status of an isolating switch based on deep learning, comprising obtaining an isolating switch image; detecting the isolating switch image using a pre-trained isolating switch detection network to obtain a detection result of the opening and closing status of the isolating switch; wherein the isolating switch detection network uses a first feature extraction and processing module and a second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to perform preliminary feature extraction on the input isolating switch image using a multi-branch convolution structure to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to perform deep feature mining on the preliminary features using a depth-separable convolution layer and a dynamic convolution layer, thereby obtaining the key features of the isolating switch image. The present invention is based on image processing, which avoids the high installation and maintenance costs of sensor detection and the problems of being interfered with by the environment. Compared with traditional visual detection, its detection network can better capture subtle state changes and improve detection accuracy through multi-branch convolution and deep feature mining, and takes into account local and global features, enhances key features, simplifies the network structure, reduces computational overhead, and meets real-time detection requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 A flowchart of a method for detecting the opening and closing status of an isolating switch based on deep learning provided by an embodiment of the present invention;

[0064] Figure 2 A technical roadmap for a method for detecting the opening and closing status of an isolating switch based on deep learning provided by an embodiment of the present invention;

[0065] Figure 3 MDMENet network structure diagram provided by an embodiment of the present invention;

[0066] Figure 4 A diagram of the MSCA-Conv module framework provided in an embodiment of the present invention;

[0067] Figure 5 A structural diagram of the DASC-Conv module provided in an embodiment of the present invention;

[0068] Figure 6 This is a structural diagram of the MSFE-Conv module provided in an embodiment of the present invention;

[0069] Figure 7 A structural diagram of the Detect module provided in an embodiment of the present invention;

[0070] Figure 8 A block diagram of a device for detecting the opening and closing status of an isolating switch based on deep learning provided by an embodiment of the present invention;

[0071] Figure 9 A block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purposes, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0073] The following first introduces some technical terms involved in the present invention:

[0074] (1) Isolating switch: A switching device mainly used for "isolating power supply, switching operation, and connecting and disconnecting low-current circuits". It plays an important role in the power system. The accurate detection of its opening and closing status is crucial to the safe and stable operation of the power system.

[0075] (2) Deep Learning (DL): It is a branch of machine learning. It builds a multi-level model based on artificial neural networks and automatically extracts features and patterns from data by learning from large amounts of data.

[0076] (3) Neural Network (NN): A computational model composed of a large number of neurons that simulates the workings of the human nervous system and learns the characteristics and patterns of data through the connections between neurons and the adjustment of weights. In deep learning, neural networks usually have multiple layers, including input layers, hidden layers, and output layers.

[0077] (4) Convolutional Neural Network (CNN): A neural network specifically designed to process grid-structured data (such as images). It uses components such as convolutional layers and pooling layers to automatically extract image features.

[0078] See also Figure 1 , an embodiment of the present invention provides a method for detecting the opening and closing state of an isolating switch based on deep learning, comprising the following steps:

[0079] S11: Acquire an image of the disconnector.

[0080] It is understandable that images of the disconnector can be acquired through certain image acquisition devices (such as high-definition cameras, drones, etc.) as input data for subsequent detection.

[0081] S12: Detect the disconnector image using a pre-trained disconnector detection network to obtain a detection result of the disconnector opening and closing state;

[0082] Among them, the isolating switch detection network uses the first feature extraction and processing module and the second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

[0083] It should be noted that the first feature extraction and processing module uses a multi-branch convolution structure to perform preliminary feature extraction on the input isolation switch image from different angles, and obtains preliminary features while maintaining the spatial resolution of the image. The multi-branch convolution structure means that multiple different convolution kernels or convolution methods are used to process the image in parallel, which can capture richer feature information in the image. The second feature extraction and processing module uses a depthwise separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features obtained by the first module, and then obtains the key features of the isolation switch image. The depthwise separable convolution layer can reduce the amount of calculation, and the dynamic convolution layer can adaptively adjust the convolution kernel according to the input, enhancing the ability to extract key features.

[0084] The detection method provided in this embodiment is based on the fundamental principles of convolutional neural networks (CNNs) in deep learning, extracting image features through convolution operations. The multi-branch convolutional structure of the first feature extraction and processing module uses different convolution kernels to convolve the image at multiple scales and directions, generating preliminary feature representations that contain both local image details and certain global structural information. The depthwise separable convolution layer and the dynamic convolution layer of the second feature extraction and processing module further process these preliminary features. The depthwise separable convolution layer reduces computational complexity, while the dynamic convolution layer dynamically adjusts convolution kernel weights based on image features, thereby more accurately extracting key features that are crucial for determining the open / close status of the disconnector. The features processed by the two feature extraction and processing modules are input into the subsequent detection network structure (although not specifically mentioned in this embodiment, it generally includes a classifier and other structures). Through analysis and processing of the key features, the detection result of the disconnector open / close status is output.

[0085] This method differs from the single-scale convolution operations commonly relied upon in traditional deep learning-based detection methods. Its multi-branch convolutional structure simultaneously extracts image features from multiple angles and scales, taking into account both local details and global structural features of the isolation switch, thereby improving the comprehensiveness and accuracy of feature extraction. Furthermore, in the second feature extraction and processing module, the depthwise separable convolutional layer reduces computational effort, making the model more efficient when processing images. The dynamic convolutional layer adaptively adjusts the convolution kernel weights based on input features, enhancing the model's adaptability to diverse image features and its ability to extract key features.

[0086] In one embodiment, the disconnect switch detection network includes:

[0087] Backbone feature extraction network, multi-scale feature processing module, attention enhancement module and state detection module;

[0088] The backbone feature extraction network includes a first feature extraction and processing module and a second feature extraction and processing module;

[0089] The first feature extraction and processing module is used to perform preliminary feature extraction on the input disconnector image. It extracts features from different angles through different convolutional layers in the multi-branch convolutional structure and then concatenates the results of each branch to obtain a preliminary feature map while maintaining the spatial resolution of the image.

[0090] The second feature extraction and processing module is used to perform deep feature mining on the preliminary features using a depthwise separable convolution layer and a dynamic convolution layer, thereby obtaining a key feature map of the disconnector image;

[0091] The multi-scale feature processing module is used to extract and fuse multi-scale features of key features to obtain a multi-scale feature map;

[0092] The attention enhancement module is used to calculate and adjust the importance weights in the multi-scale feature map based on the EA attention mechanism to obtain a feature map that highlights key features;

[0093] The state detection module is used to determine the final disconnector opening and closing state detection results based on the feature map that highlights the key features.

[0094] In this embodiment, the isolating switch detection network consists of a backbone feature extraction network, a multi-scale feature processing module, an attention enhancement module, and a state detection module. The backbone feature extraction network, as the foundation, further includes a first feature extraction and processing module and a second feature extraction and processing module. The first feature extraction and processing module uses a multi-branch convolutional structure to perform preliminary feature extraction on the input isolating switch image from multiple angles through different convolutional layers, and splices the results of each branch to obtain a preliminary feature map while maintaining the spatial resolution of the image, providing basic feature information for subsequent processing. The second feature extraction and processing module uses a depthwise separable convolutional layer and a dynamic convolutional layer to perform deep feature mining on the preliminary feature map, further refining the key feature map of the isolating switch image, making the features more critical and representative. The multi-scale feature processing module performs multi-scale feature extraction and fusion operations on the key feature map to obtain a multi-scale feature map, enhancing the diversity of features and the ability to capture information at different scales. The attention enhancement module, based on the EA attention mechanism, calculates and adjusts the importance weights of various components in the multi-scale feature map, highlighting key features and suppressing irrelevant information. This produces a feature map that emphasizes key features, allowing the model to focus more on important features. Finally, the state detection module, based on this feature map, analyzes and processes the final disconnector state detection result, ensuring that the entire detection network accurately determines the disconnector state.

[0095] In a specific implementation, the present invention proposes a multi-scale dynamic enhanced fusion network MDMENet (Multi-scale Dynamic Multi-branch Enhanced Network) based on the aforementioned embodiment, and designs a disconnector opening and closing state detection method based on the network, including first constructing an disconnector image dataset. Then preprocess the image dataset and divide the image dataset into a training set, a validation set, and a test set according to a certain ratio. Then design an MDMENet network and initialize the training parameters. Then load the training set and validation set into the MDMENet network for training and update the network parameters, and then input the test set into the trained network to verify the effect of the network, and then apply the trained MDMENet network to the disconnector opening and closing state detection task. The technical route of this method is as follows: Figure 2 As shown, it is introduced below in combination with some embodiments.

[0096] S1: Construct an isolating switch image dataset.

[0097] High-definition cameras and drones were used in conjunction with each other in the substation to capture static and dynamic images of disconnectors, capturing them at regular intervals. To enhance the dataset's applicability and robustness for algorithm training, we carefully collected images of disconnectors under various weather conditions (sunny, cloudy, rainy, and foggy), lighting intensities (strong light, soft light, and backlight), and states (open and closed). These images were screened and classified to construct a comprehensive and detailed disconnector image dataset.

[0098] In a specific embodiment, the time for periodically capturing images from the video is determined according to the current monitoring task of the isolating switch status. When the isolating switch needs to be opened or closed, the interval is usually about 1 second; otherwise, the interval is usually about 30 seconds.

[0099] S2: Perform data preprocessing and preprocessing division of the data set.

[0100] The disconnector image data was preprocessed and the dataset was divided. A Gaussian filter algorithm was applied to the collected disconnector images to smooth them, reducing high-frequency noise while preserving edge information. This resulted in higher-resolution images of single-pole vertical telescopic disconnectors. These processed images constitute a high-quality single-pole vertical telescopic disconnector image dataset, providing a solid foundation for subsequent image analysis and condition detection. Annotation tools were then used to annotate ice-covered areas. Finally, the grid overhead line ice image dataset was divided into training, validation, and test sets in a 7:2:1 ratio.

[0101] It's important to note that the Gaussian filter algorithm is a linear smoothing filter technique widely used in image processing. It uses the mathematical Gaussian function as its core and performs convolution operations on images to achieve noise reduction and smoothing. The core idea of ​​Gaussian filtering is to leverage the properties of the Gaussian function to assign a weight to each pixel in the image. This weight decreases as the pixel's distance from the center increases, thereby removing noise while preserving image edges and details as much as possible.

[0102] S3: Design the MDMENet network.

[0103] In one embodiment of the present invention, a multi-scale dynamic enhancement fusion network MDMENet is designed to detect the opening and closing status of the disconnector. Its structure is as follows: Figure 3 As shown. In MDMENet, the input disconnector image is first input into the multi-scale context aggregation module MSCA-Conv (Multi-Scale Context Aggregation Convolution) module for feature fusion, and then the key features in the disconnector image are gradually extracted through the dynamic adaptive spatial convolution module DASC-Conv (Dynamic AdaptiveSpatial Convolution) module, and the detection capability of the model is enhanced through nonlinear activation. At the same time, it is input into the multi-scale feature fusion module MSFE-Conv (Multi-Scale Feature Extraction & Fusion Convolution) module through multi-scale feature extraction and feature fusion. Then the EA attention mechanism is introduced, and finally the disconnector opening and closing state detection result is output through the Detect module. The network structure is shown as follows Figure 3 shown.

[0104] Step 3.1: Perform feature fusion on the input disconnector image.

[0105] In a further embodiment, the first feature extraction and processing module is designed as an MSCA-Conv module, and its structure is as follows: Figure 4 As shown. Any isolating switch image in the isolating switch image dataset is recorded as the isolating switch feature map , input into the MSCA-Conv module for feature fusion.

[0106] In the MSCA-Conv module, the input isolation switch characteristic map is first Perform a convolution (Convolution, Conv) operation with a convolution kernel size of 1 and a stride of 1, and input it to the BN (Batch Normalization, Batch Normalization) layer for normalization to obtain the output isolation switch feature map , the main purpose of this step is to extract the disconnector feature map This convolution operation enhances the representation ability of features while maintaining spatial resolution; Perform a dilated convolution operation with a kernel size of 3 and a stride of 1 (dilated ratio = 2), and input it into the BN layer for normalization to obtain the isolation switch feature map. The main purpose of this step is to Perform more extensive context feature extraction; in addition, Perform a depth-wise separable convolution operation with a kernel size of 5 and a step size of 1, and input it into the BN layer for normalization to obtain the isolation switch feature map ,The main purpose of this step is to extract a wider range of local ,features while reducing the amount of parameters.

[0107] Afterwards 、 、 Perform channel splicing to obtain the characteristic diagram of the isolation switch The main purpose of this step is to fuse multi-scale features and enhance the diversity of features; then Perform a convolution operation with a kernel size of 1 and a stride of 1, adjusting the number of channels to match the input The same, get the characteristic diagram of the disconnector The main purpose of this step is to integrate multi-scale features while maintaining spatial resolution; then Perform global average pooling to obtain the feature map ,The main purpose of this step is to compress the spatial dimension, preserve ,global information, and reduce the computational complexity.

[0108] Then Generate attention weights through the fully connected layer and use the Sigmoid function for normalization to obtain the isolation switch feature map The main purpose of this step is to enhance important features and suppress irrelevant features; finally and Perform dot multiplication to obtain the characteristic diagram of the disconnector ,The main purpose of this step is to further enhance the model’s sensitivity to key features through the ,attention mechanism, thereby improving the feature representation ability.

[0109] The MSCA-Conv module uses multi-scale feature extraction and feature fusion to enhance the diversity and expressiveness of features while maintaining the spatial resolution of the input isolation switch feature map, allowing the model to focus more on important features, thereby improving detection accuracy and robustness.

[0110] In a further embodiment, and The transformation formula is as follows:

[0111] (1)

[0112] (2)

[0113] Where, It is a convolution with kernel size 1 and stride 1. It is a convolution with a kernel size of 3 and a stride of 1. It is a convolution with a kernel size of 5 and a stride of 1. is the average pooling operation, is the activation function, FC is the fully connected layer, It is element-wise multiplication.

[0114] It should be noted that the Sigmoid function is a commonly used activation function, usually used for nonlinear transformations in neural networks. The output value of the Sigmoid function is between 0 and 1. When the input x approaches positive infinity, the output is close to 1; when the input approaches negative infinity, the output is close to 0. Its mathematical expression is: .

[0115] Average pooling is a pooling operation commonly used in convolutional neural networks (CNNs). It is mainly used to downsample the input feature map and reduce the size of the feature map, thereby reducing the amount of computation, alleviating overfitting, and extracting more representative features.

[0116] A fully connected layer (FC layer) is a common layer type in deep learning models and is widely used in neural networks, particularly in convolutional neural networks (CNNs) and fully connected neural networks (MLPs). Its primary function is to map input features to the output space and is typically used for classification, regression, or other tasks.

[0117] Step 3.2: Extract key features from the isolation switch feature map output by MSCA-Conv.

[0118] In a further embodiment of the present invention, the second feature extraction and processing module is designed as a DASC-Conv module, and its structure is as follows: Figure 5 shown.

[0119] In the DASC-Conv module, we first Perform a depth-wise separable convolution operation with a kernel size of 1 and a stride of 1 to obtain the isolation switch feature map. The main purpose of this step is to extract the local detail features in the disconnector feature map and reduce the number of parameters; then Perform a dynamic convolution operation with a convolution kernel size of 3 and a step size of 1 to obtain the isolation switch feature map The main purpose of this step is to capture the local structural features of the isolation switch and adaptively adjust the convolution kernel weights through dynamic convolution to enhance the feature expression ability; then Use MaxPool to perform the maximum pooling operation to obtain the isolation switch feature map , the main purpose of this step is to compress the spatial dimension and retain the global information. The isolation switch feature map is obtained by upsampling through bilinear interpolation and activating through the Swish function. The main purpose of this step is to compress the value range of the disconnector feature map to [0, 1], further enhancing the sensitivity of the model to the disconnector state; finally, and Perform dot multiplication to obtain the output isolation switch characteristic diagram ,The main purpose of this step is to fuse local features with global ,features through the dot product operation to enhance the model’s ,ability to judge the status of the disconnector.

[0120] The DASC-Conv module gradually extracts key features from disconnector images through operations such as depthwise separable convolution, dynamic convolution, and max pooling. It also enhances the model's detection capabilities through nonlinear activation. This module boasts a clear structure and efficient computation, effectively improving the accuracy and robustness of disconnector status detection.

[0121] The execution flow of the DASC-Conv module is as follows: Assume The dimension is (where H is the height, W is the width, and C is the number of channels), first Perform a convolution operation with a kernel size of 1, a step size of 1, and a number of C, and the dimension is Disconnector characteristic diagram ; then Perform a convolution operation with a kernel size of 3, a step size of 1, and a number of C, and the dimension is Disconnector characteristic diagram ; then Perform the maximum pooling operation and get the dimension Disconnector characteristic diagram In addition, Perform bilinear interpolation (BilinearInterpolation) to upsample and obtain the dimension Disconnector characteristic diagram ; Finally, Activated by the Swish function and with Perform dot multiplication to obtain the characteristic diagram of the disconnector , The dimension is .

[0122] In a further embodiment of the present invention, and The transformation formula is as follows:

[0123] (3)

[0124] (4)

[0125] In the formula, Swish is the activation function, MaxPool is the maximum pooling operation, It is a convolution operation with a kernel size of 1 and a stride of 1. It is a convolution operation with a kernel size of 3 and a stride of 1. BI stands for Bilinear Interpolation operation.

[0126] It should be noted that max pooling is a common downsampling operation, often used in convolutional neural networks (CNNs). Its function is to extract the maximum value from each local region (such as a 2x2 or 3x3 window) of the input feature map, thereby reducing the size of the feature map while retaining the most significant features.

[0127] Swish is a smooth and non-monotonic activation function proposed by Google. Its formula is:

[0128]

[0129] Bilinear interpolation is a commonly used image upsampling method used to upscale low-resolution images or feature maps to higher resolutions. It calculates the target pixel value by taking the weighted average of the four nearest pixels, resulting in simple computation and smooth results.

[0130] Step 3.3: Perform multi-scale feature extraction and fusion on the feature map output by DASC-Conv.

[0131] In one embodiment of the present invention, the multi-scale feature processing module is a MSFE-Conv module, and its structure is as follows: Figure 6 As shown. The dimension of the output from DASC-Conv is Characteristic diagram of utility pole disconnector Input into the MSFE-Conv module for multi-scale feature extraction and fusion.

[0132] In the MSFE-Conv module, first The convolution operation with kernel size of 3 and step size of 1 and the convolution operation with kernel size of 1 and step size of 1 are performed respectively to obtain the characteristic graphs of the isolation switch. and disconnector characteristic diagram ,The main purpose of this step is to extract the local details and global structural features in the ,disconnector feature map; then and Perform channel concatenation (Concat) to obtain the characteristic diagram of the isolation switch The main purpose of this step is to fuse multi-scale features and enhance the diversity of features; then Perform global average pooling to obtain the isolation switch feature map The main purpose of this step is to compress the spatial dimension and extract global features; then Use the GELU activation function to perform activation operation and obtain the isolation switch feature map The main purpose of this step is to introduce nonlinearity and enhance the expressive power of features; then Perform convolution operation with kernel size of 1 and stride of 1, and adjust the number of channels to the same as The same, get the characteristic diagram of the disconnector , the main purpose of this step is to integrate multi-scale features; finally and Perform dot multiplication to obtain the output isolation switch characteristic diagram The main purpose of this step is to combine global features with original input features through dot product operations to enhance the model's sensitivity to key features. The MSFE-Conv module gradually extracts key features from the input isolation switch feature map through multi-scale feature extraction and feature fusion, and enhances the model's expressive power through nonlinear activation and feature fusion.

[0133] In a further embodiment of the present invention, 、 and The transformation formula is as follows:

[0134] (5)

[0135] (6)

[0136] (7)

[0137] Where, is the average pooling, GELU is the activation function, It is a convolution operation with a kernel size of 1 and a stride of 1. It is a convolution operation with a kernel size of 3 and a stride of 1.

[0138] It's important to note that GELU is an activation function that combines the ideas of ReLU and Dropout, enhancing the expressive power of neural networks by introducing the cumulative distribution function (CDF) of a Gaussian distribution. GELU has excellent performance in natural language processing (NLP) and computer vision (CV) tasks, and is particularly widely used in Transformer models such as BERT and GPT.

[0139] Step 3.4: Introduce the EA attention mechanism and detect it through the Detect module.

[0140] In one embodiment of the present invention, the state detection module is a Detect module, and its structure is as follows: Figure 7 shown.

[0141] Will Input to the EA module for attention mechanism feature enhancement to obtain feature map The disconnector state prediction module Detect outputs the open and close state detection results.

[0142] In the Detect module, first the isolation switch feature map Input to the global average pooling module (AvgPooling) to average the feature values ​​of each channel along the spatial dimension to generate the isolation switch feature map . The characteristic diagram of the isolating switch Through the fully connected layer, the channel dimension is mapped to the number of target categories (here 2, including the open and closed states of the disconnector), and the feature vector T is output; the Softmax function is applied to normalize T into a probability distribution ,in ,in, represents the probability that the disconnector is in the open state, represents the probability that the disconnector is in the closed state, The calculation process is as follows:

[0143] (8)

[0144] (9)

[0145] Where, is the average pooling operation, FC is the fully connected layer, is a normalized function, and P is the normalized probability distribution of the disconnector opening and closing states.

[0146] It should be noted that the Softmax function is a function that maps any real vector to a probability distribution. Given an input vector , the output of the Softmax function .

[0147] The Enhanced Attention Mechanism (EA) is an attention mechanism used to enhance the feature representation capabilities of deep learning models. By combining channel attention and spatial attention, it dynamically adjusts the importance of each channel and spatial position in the feature map, thereby enhancing key features and suppressing irrelevant background information. EA is widely used in computer vision tasks such as image classification, object detection, and semantic segmentation, significantly improving model performance.

[0148] In a specific embodiment, for an isolated switch image with an input dimension of 1024×1024×12 (width×height×channel), , first in the MSCA-Conv module, Perform a convolution operation with a kernel size of 1 and a stride of 1, and input it to the BN layer for normalization to obtain an isolation switch feature map with an output dimension of 1024×1024×12. , and input Perform a dilated convolution operation with a kernel size of 3 and a stride of 1 (dilated ratio = 2), and input it into the BN layer for normalization to obtain an isolating switch feature map with a dimension of 1024×1024×12. ; In addition, you will enter Perform a depth-wise separable convolution operation with a kernel size of 5 and a stride of 1, and input it to the BN layer for normalization to obtain an isolation switch feature map with a dimension of 1024×1024×12. .

[0149] Afterwards 、 、 Perform channel splicing to obtain the isolation switch feature map with a dimension of 1024×1024×36 ; then Perform a convolution operation with a kernel size of 1 and a stride of 1 to obtain an isolation switch feature map with a dimension of 1024×1024×12. , then Perform global average pooling to obtain the isolation switch feature map with a dimension of 1×1×12 .

[0150] Then The attention weights are generated through the fully connected layer and normalized using the Sigmoid function to obtain the isolation switch feature map with a dimension of 1×1×12. , and finally Characteristic diagram of input isolation switch Perform dot multiplication and output the isolation switch feature map with a dimension of 1024×1024×12 .

[0151] Then Input into the DASC-Conv module, first Perform a depth-wise separable convolution operation with a kernel size of 1 and a stride of 1 to obtain an isolation switch feature map with a dimension of 1024×1024×12. ; then Perform a dynamic convolution operation with a convolution kernel size of 3 and a step size of 1 to obtain the isolation switch feature map with a dimension of 1024×1024×12 ; Then Use MaxPool to perform the maximum pooling operation to obtain the isolation switch feature map with a dimension of 512×512×12 Then The isolating switch feature map with a dimension of 1024×1024×12 is obtained by upsampling through bilinear interpolation and activating through the Swish function. ; Finally, and Perform a dot multiplication operation to obtain the output isolation switch feature map with a dimension of 1024×1024×12 .

[0152] The feature map of the utility pole disconnector outputted from DASC-Conv with a dimension of 1024×1024×12 Input to the MSFE-Conv module for multi-scale feature extraction and fusion. In the MSFE-Conv module, first Convolution operations with kernel size of 3 and stride of 1 and convolution operations with kernel size of 1 and stride of 1 are performed respectively to obtain the isolation switch feature map with a dimension of 1024×1024×12. And the characteristic diagram of the disconnector with dimensions of 1024×1024×12 ; then and Perform channel concatenation (Concat) to obtain an isolation switch feature map with a dimension of 1024×1024×24. ; Then Perform global average pooling to obtain a 1×1×24 isolating switch feature map. ; then Use the GELU activation function to perform activation operation and obtain the isolation switch feature map with a dimension of 1×1×24 ; Then Perform convolution operation with kernel size of 1 and stride of 1, and adjust the number of channels to the same as The same, the characteristic map of the disconnector with the dimension of 1×1×12 is obtained. ; Finally, With isolating switch Perform a dot multiplication operation to obtain the isolation switch feature map with an output dimension of 1024×1024×12 .

[0153] Will Input to the ECA module for attention mechanism feature enhancement to obtain the isolation switch feature map with a dimension of 1024×1024×12 The disconnector state prediction module Detect outputs the open and close state detection results.

[0154] In the disconnector status prediction module Detect, first The input is sent to the global average pooling module (AvgPooling) to average the feature values ​​of each channel along the spatial dimension to generate an isolation switch feature map with a dimension of 1×1×12. . The characteristic diagram of the isolating switch Through the fully connected layer, the channel dimension is mapped to the number of target categories (here 2, including the open and closed states of the disconnector), and the output dimension is a feature vector T of 1×1×2. The Softmax function is applied to T for activation and normalization to a probability distribution. , The probability that the disconnector is in the open state is 0.85. The probability that the disconnector is in the closed state is 0.15, because , so the current state of the isolation switch is open.

[0155] S4: Load the training set and validation set into the network for training to update the network parameters, and then use the test set as input to verify the network effect.

[0156] During MDMENet network training, parameters are initialized based on the device's electrical characteristics. A multi-condition alternating optimization algorithm is then used to dynamically adjust the learning weights for clear and foggy weather samples during training. Model optimization is performed using a composite loss function (including state classification, switch profile, and arc signature losses). Training employs a fast alternating optimization strategy, minimizing the binary cross-entropy loss function to measure the difference between the model's predictions and the true labels. During the training phase, the preprocessed training data is fed into the network, with a batch size of 2 and 1000 iterations. Backpropagation is performed using the stochastic gradient descent algorithm, and the Adam optimizer is introduced to dynamically adjust the learning rate to accelerate network convergence and improve training stability.

[0157] After training, the validation set data was fed into the trained MDMENet network to calculate the accuracy of disconnector status detection to evaluate the model's performance on the validation set. Based on the validation results, network performance was further optimized, including adjusting hyperparameters such as the learning rate and regularization parameters, as well as fine-tuning the network structure (such as adding attention mechanism branches or introducing residual connections), thereby improving the model's detection accuracy and generalization capabilities.

[0158] The test set obtained in Step 2 is input into the MDMENet network to evaluate its performance in the disconnector status detection task, including indicators such as accuracy and stability. Based on the evaluation results, the model and system are further improved and optimized.

[0159] It should be noted that the composite loss function is an optimization objective function constructed by weightedly combining multiple basic loss functions. In disconnector status detection, it typically includes classification loss (to ensure accurate status judgment), positioning loss (to precisely locate the switch position), and edge loss (to enhance contour recognition). The contributions of each loss term are balanced using weight coefficients (such as 0.6, 0.3, and 0.1), ultimately forming a joint optimization objective that comprehensively optimizes model performance. This design maintains the discriminative power of the primary task while enhancing fine-grained feature learning through auxiliary loss terms. In practical applications, it can increase detection accuracy by over 5% while significantly reducing positioning error.

[0160] S5: Apply the obtained network to the disconnector opening and closing status detection task.

[0161] The MDMENet network is applied to the disconnector opening and closing state detection task. The disconnector image is captured from the static and dynamic images of the disconnector at regular intervals and then input into the MDMENet network. The MDMENet network processes the image and outputs the probability distribution of the disconnector. , and then get the opening and closing status of the disconnector. When the isolating switch is in the open state, it is determined that the isolating switch is in the closed state; otherwise, it is in the closed state.

[0162] Based on the same inventive concept, the embodiments of the present application also provide a deep learning-based disconnector open / close state detection device for implementing the aforementioned deep learning-based disconnector open / close state detection method. The implementation solution provided by this device is similar to the implementation solution described in the aforementioned method. Therefore, the specific limitations of the deep learning-based disconnector open / close state detection device embodiment provided below can be found in the limitations of the deep learning-based disconnector open / close state detection method above, and will not be repeated here.

[0163] See also Figure 8 , an embodiment of the present invention further provides a device for detecting the opening and closing state of an isolating switch based on deep learning, comprising:

[0164] An image acquisition module, used for acquiring an image of the disconnector;

[0165] A detection module is used to detect the disconnector image using a pre-trained disconnector detection network to obtain a detection result of the disconnector opening and closing state;

[0166] Among them, the isolating switch detection network uses the first feature extraction and processing module and the second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

[0167] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0168] Reference Figure 9 An embodiment of the present invention further provides a computer device, comprising: a memory and a processor and a computer program stored in the memory. When the computer program is executed on the processor, the method for detecting the opening and closing status of an isolating switch based on deep learning as described in any one of the above methods is implemented.

[0169] The computer device may be a desktop computer, notebook computer, PDA, cloud server or other computing device. The computer device may include, but is not limited to, a processor and a memory. It will be understood by those skilled in the art that Figure 9 This is merely an example of a computer device and does not constitute a limitation on the computer device. The computer device may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0170] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0171] In some embodiments, the memory may be an internal storage unit of the computer device, such as a hard drive or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped with the computer device. Furthermore, the memory may include both an internal storage unit of the computer device and an external storage device. The memory is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or is about to be output.

[0172] An embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the deep learning-based disconnector opening and closing state detection method as described in any one of the above methods.

[0173] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0174] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0175] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0176] In the embodiments disclosed in the present application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely schematic. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0177] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting the opening and closing status of an isolating switch based on deep learning, characterized in that: The steps include: Get the disconnector image; Detecting the isolating switch image using a pre-trained isolating switch detection network to obtain a detection result of the isolating switch opening and closing state; Among them, the isolating switch detection network uses a first feature extraction and processing module and a second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

2. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 1, characterized in that: The isolating switch detection network includes: Backbone feature extraction network, multi-scale feature processing module, attention enhancement module and state detection module; The backbone feature extraction network includes the first feature extraction and processing module and the second feature extraction and processing module; The first feature extraction and processing module is used to perform preliminary feature extraction on the input disconnector image. It extracts features from different angles through different convolutional layers in the multi-branch convolutional structure, and then splices the results of each branch to obtain a preliminary feature map while maintaining the spatial resolution of the image. The second feature extraction and processing module is used to perform deep feature mining on the preliminary features using a depthwise separable convolution layer and a dynamic convolution layer, thereby obtaining a key feature map of the disconnector image; The multi-scale feature processing module is used to perform multi-scale feature extraction and fusion on the key features to obtain a multi-scale feature map; The attention enhancement module is used to calculate and adjust the importance weights in the multi-scale feature map based on the EA attention mechanism to obtain a feature map that highlights key features; The state detection module is used to determine the final disconnector opening and closing state detection result based on the feature map highlighting the key features.

3. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 1 or 2, characterized in that: The first feature extraction and processing module is an MSCA-Conv module, and the image processing process of the MSCA-Conv module includes: A set of Conv modules are used to perform convolution operations, dilated convolution operations, and depthwise separable convolution operations on the input isolation switch feature map, thereby obtaining a second isolation switch feature map, a third isolation switch feature map, and a fourth isolation switch feature map; performing normalization processing on the second isolating switch characteristic graph, the third isolating switch characteristic graph, and the fourth isolating switch characteristic graph through a BN layer; pass The standardized isolating switch feature graphs are spliced ​​at a layer to obtain a fifth isolating switch feature graph; Performing a convolution operation on the fifth isolating switch feature map through a Conv module to obtain a sixth isolating switch feature map; Processing the sixth isolating switch characteristic map through a global average pooling module to obtain a seventh isolating switch characteristic map; Generating an attention weight for the seventh isolation switch feature map through a fully connected layer, and normalizing it using a Sigmoid function to obtain an eighth isolation switch feature map; A point product operation is performed on the eighth isolating switch characteristic map and the input isolating switch characteristic map to obtain a ninth isolating switch characteristic map, and the ninth isolating switch characteristic map serves as the final output of the MSCA-Conv module.

4. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 3 is characterized in that: The transformation formula of the fifth isolating switch characteristic diagram and the ninth isolating switch characteristic diagram is as follows: Where, represents the characteristic diagram of the fifth disconnector, Represents a splicing operation, Indicates standardized processing; 、 and They represent convolution operations with a kernel size of 1 and a stride of 1, dilated convolution operations with a kernel size of 3 and a stride of 1, and depthwise separable convolution operations with a kernel size of 5 and a stride of 1. Characteristic diagram of the disconnector representing the input; 4 shows the characteristic diagram of the ninth disconnector.

5. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 1 or 2, characterized in that: The second feature extraction and processing module is a DASC-Conv module, and the image processing process of the DASC-Conv module includes: Performing a convolution operation on the input isolation switch feature map through the first Conv module to obtain a tenth isolation switch feature map; Performing a convolution operation on the tenth isolating switch feature map through a second Conv module to obtain an eleventh isolating switch feature map; Performing maximum pooling processing on the eleventh isolating switch feature map through a maximum pooling module to obtain a twelfth isolating switch feature map; Performing bilinear interpolation processing on the twelfth isolating switch characteristic graph by a bilinear interpolation module to obtain a thirteenth isolating switch characteristic graph; The thirteenth isolating switch characteristic map is activated by a Swish function module, and a dot product operation is performed on the thirteenth isolating switch characteristic map with the tenth isolating switch characteristic map to obtain a fourteenth isolating switch characteristic map, and the fourteenth isolating switch characteristic map is used as the final output of the DASC-Conv module.

6. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 5, characterized in that: The transformation formula of the thirteenth isolating switch characteristic diagram and the fourteenth isolating switch characteristic diagram is as follows: Where, represents the characteristic diagram of the thirteenth disconnector, represents a bilinear interpolation operation, represents the maximum pooling operation, Represents a convolution operation with a kernel size of 3 and a stride of 1. Represents a convolution operation with a kernel size of 1 and a stride of 1. Characteristic diagram of the disconnector representing the input; express activation function, represents the characteristic diagram of the tenth disconnector, It represents the characteristic diagram of the fourteenth disconnector.

7. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 2, characterized in that: The multi-scale feature processing module is an MSFE-Conv module, and the image processing process of the MSFE-Conv module includes: Performing convolution operations on the input isolation switch feature map through two Conv modules of different sizes to obtain a fifteenth isolation switch feature map and a sixteenth isolation switch feature map; Performing a channel splicing operation on the fifteenth isolating switch characteristic diagram and the sixteenth isolating switch characteristic diagram by a channel splicing module to obtain a seventeenth isolating switch characteristic diagram; Performing a global average pooling operation on the seventeenth isolator feature map through a global average pooling module to obtain an eighteenth isolator feature map; Activating the eighteenth isolating switch characteristic graph through a GELU activation function to obtain a nineteenth isolating switch characteristic graph; Performing a convolution operation on the nineteenth isolating switch feature map through a Conv module to obtain a twentieth isolating switch feature map; A dot product operation is performed on the 20th isolating switch characteristic map and the input isolating switch characteristic map to obtain a 21st isolating switch characteristic map, and the 21st isolating switch characteristic map is used as the final output of the MSFE-Conv module.

8. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 7, characterized in that: The transformation formulas of the seventeenth isolating switch characteristic diagram, the twentieth isolating switch characteristic diagram, and the twenty-first isolating switch characteristic diagram are as follows: Where, represents the characteristic diagram of the seventeenth disconnector, Indicates channel splicing operation, Represents a convolution operation with a kernel size of 1 and a stride of 1. Represents a convolution operation with a kernel size of 3 and a stride of 1. Characteristic diagram of the disconnector representing the input; The characteristic diagram of the 20th disconnector is shown in FIG. express activation function, represents the global average pooling operation, Represents the twenty-first disconnector characteristic diagram.

9. The method for detecting the opening and closing state of an isolating switch based on deep learning according to claim 2, characterized in that: The state detection module is a Detect module, and the characteristic graph highlighting the key features is recorded as the twenty-second disconnector characteristic graph. Based on the characteristic graph highlighting the key features, determining the final disconnector opening and closing state detection result includes: averaging the feature values ​​of each channel in the twenty-second isolator feature map along the spatial dimension through a global average pooling module to generate a twenty-third isolator feature map; Pass the twenty-third isolation switch feature map through a fully connected layer, map the channel dimension to the number of target categories, and output a feature vector; Normalize the feature vector into a probability distribution through the Softmax function ,in ,in, represents the probability that the disconnector is in the open state, Represents the probability that the disconnector is in the closed state.

10. A device for detecting the opening and closing status of an isolating switch based on deep learning, characterized in that: include: An image acquisition module, used for acquiring an image of the disconnector; A detection module is used to detect the isolating switch image using a pre-trained isolating switch detection network to obtain a detection result of the isolating switch opening and closing state; Among them, the isolating switch detection network uses a first feature extraction and processing module and a second feature extraction and processing module to perform feature extraction; the first feature extraction and processing module is used to use a multi-branch convolution structure to perform preliminary feature extraction on the input isolating switch image, so as to obtain preliminary features while maintaining the spatial resolution of the image; the second feature extraction and processing module is used to use a depth-separable convolution layer and a dynamic convolution layer to perform deep feature mining on the preliminary features, so as to obtain the key features of the isolating switch image.

Images