Breaker switch intelligent detection method and system based on deep learning, medium and equipment

Abstract

The invention discloses a circuit breaker switch intelligent detection method and system based on deep learning, a medium and equipment. The method comprises the following steps: acquiring a circuit breaker image and forming a training set; based on the YOLOV8 model, replacing a convolution module of the YOLOV8 with a depth separable asymmetric convolution module to obtain a YOLOV8m model; and training by using the training set to obtain a circuit breaker switch identification model. Inputting the training set into a circuit breaker switch identification model and outputting rectangular frame coordinate information of a switch, intercepting a switch image from an original circuit breaker image according to the output rectangular frame coordinate, and classifying according to a switch state to form a data set; replacing an SE module in the MoblieNetV3 network with a module fusing space attention and channel attention, and obtaining the MoblieNetV3 network fusing an attention mechanism; and training by adopting the data set to obtain a switch state recognition model. And converting the circuit breaker switch identification model and the switch state identification model into ONNX formats, and deploying the ONNX formats on equipment. According to the invention, efficient and accurate circuit breaker state detection on a mobile terminal or embedded equipment is realized, and the intelligent level and the detection efficiency of avionics experiments are improved.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent circuit breaker detection, specifically relating to a method, system, medium, and device for intelligent detection of circuit breaker switches based on deep learning. Background Technology

[0002] Traditional object detection networks, especially deep learning models, typically require powerful computing devices, such as workstations and servers equipped with high-performance graphics cards, due to their large number of network parameters. This makes it difficult to apply existing deep learning models to resource-constrained devices (such as embedded devices and mobile devices). Furthermore, the existence of various deep learning frameworks (such as TensorFlow, PyTorch, and Caffe) with significant differences in format and operation between them necessitates extensive adaptation and conversion work when deploying models across platforms or frameworks, hindering efficient model migration across different devices and platforms. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, medium, and device for intelligent detection of circuit breaker switches based on deep learning, which aims to solve the limitations of traditional methods in resource-constrained devices, and at the same time solve the compatibility problem between different deep learning frameworks.

[0004] This invention is mainly achieved through the following technical solutions:

[0005] A deep learning-based intelligent detection method for circuit breaker switches includes the following steps:

[0006] Step S1: Acquire circuit breaker images, label and preprocess them to form a training set;

[0007] Step S2: Based on the YOLOV8 model, a depthwise separable asymmetric convolutional module is used to replace the convolutional module of YOLOV8 to obtain the YOLOV8m model; the lightweight YOLOV8m model is trained using the training set to obtain the circuit breaker switch recognition model.

[0008] Step S3: Input the training set into the circuit breaker switch recognition model and output the rectangular frame coordinate information of the switch. Extract the switch image from the original circuit breaker image according to the output rectangular frame coordinates, and classify the data according to the switch state to form a dataset.

[0009] Step S4: Replace the SE module in the MobileNetV3 network with a module that integrates spatial attention and channel attention to obtain a MobileNetV3 network with an integrated attention mechanism; input the dataset into the MobileNetV3 network with an integrated attention mechanism for training to obtain the switch state recognition model.

[0010] Step S5: Convert the circuit breaker switch identification model in step S2 and the switch status identification model in step S4 into ONNX format and deploy them on the device to detect the circuit breaker switch status.

[0011] To better implement the present invention, further, in step S2, the processing of the depth-separable asymmetric convolution module includes the following steps:

[0012] Step A1: Perform channel-average segmentation on the input feature map, dividing it into two parts, and send them to the upper and lower branches of feature extraction for processing respectively;

[0013] Step A2: In the upper branch, first, a convolution operation with kernel size (1, 3) and stride 2 is performed; then, a convolution operation with kernel size (3, 1) and stride 2 is performed to obtain feature F1;

[0014] Step A3: In the next branch, first, a convolution operation with a kernel size of (3, 3) and a stride of 1 is performed; then, a convolution operation with a kernel size of (3, 3) and a stride of 2 is performed to obtain feature F2.

[0015] Step A4: Concatenate features F1 and F2 along the channel direction;

[0016] Step A5: Perform a pointwise convolution operation on the concatenated features with a 1×1 high-dimensional convolution kernel to fuse feature information and obtain feature F.

[0017] To better realize the present invention, step S3 further includes the following steps:

[0018] Step S31: Obtain the output of the circuit breaker switch identification model (x, y, w, h, confidence score, category, category probability);

[0019] Where (x, y) are the coordinates of the center of the bounding box;

[0020] w and h are the width and height of the bounding box, respectively;

[0021] Step S32: Transform the normalized coordinates (x, y) into the actual coordinates (x, y). actual y actual ):

[0022] x actual = x × 640,

[0023] y actual =y×640,

[0024] width actual =w×640,

[0025] height actual =h×640,

[0026] Among them, (x actual y actual () represents the actual center point coordinates of the rectangle;

[0027] width actual and height actual These are the actual width and height of the rectangle, respectively.

[0028] Step S33: Calculate the coordinates (x, y) of the top left corner of the rectangle. min y min ) and the coordinates of the lower right corner (x) max y max ):

[0029]

[0030]

[0031] Step S34: Extract an image from the original image based on the coordinates of the top left and bottom right corners of the rectangle:

[0032] cropped_image = image[y min :x min y max :x max ];

[0033] Where: cropped_image is the cropped image.

[0034] "image" refers to the original image.

[0035] To better realize the present invention, in step S3, the captured switch image is enlarged to 224×224 and classified according to its actual state to construct a binary classification dataset.

[0036] To better implement the present invention, further, in step S4, the processing of the module fusing spatial attention and channel attention includes the following steps:

[0037] Step B1: The input feature vector is fed into the channel attention mechanism branch and the spatial attention mechanism branch for processing, respectively;

[0038] Step B2: In the channel attention mechanism branch, the input feature F is first processed by global max pooling and global average pooling, then fed into two fully connected layers for training, and finally broadcast and added element by element to obtain the channel attention feature map α.

[0039] Step B3: In the spatial attention mechanism branch, the input feature F is first max pooled and average pooled along the channel direction to obtain a 2×H×W feature map, and then a spatial attention feature map β is generated by passing it through a 5×5 standard convolutional layer.

[0040] Step B4: Broadcast the channel attention feature map α and the spatial attention feature map β and multiply them element-wise to obtain the C×H×W feature map γ.

[0041] This invention is mainly achieved through the following technical solutions:

[0042] A deep learning-based intelligent circuit breaker detection system is provided to implement the aforementioned deep learning-based intelligent circuit breaker detection method. The system includes a camera module, a data preprocessing module, and a model processing module. The camera module is used to acquire images, and the data preprocessing module is used to preprocess the captured images. The model processing module uses an ONNX format circuit breaker recognition model and a switch status recognition model to analyze the preprocessed circuit breaker images and identify their switch status.

[0043] To better realize the present invention, it further includes a search module, a voice playback module, and a result display module. The search module is used to search and determine the experiment name and circuit breaker number so as to display the matching experiment template. The voice playback module is used to implement the voice broadcast function. The result display module is used to highlight the potential problem areas in the captured circuit breaker image.

[0044] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned intelligent detection method for circuit breakers based on deep learning.

[0045] An electronic device includes a memory and a processor; the memory stores a computer program; the processor is configured to execute the computer program in the memory to implement the aforementioned intelligent detection method for circuit breakers based on deep learning.

[0046] The beneficial effects of this invention are as follows:

[0047] (1) This invention utilizes the powerful feature extraction capabilities of deep learning models and combines them with the ONNX (Open Neural Network Exchange) standard format to achieve efficient and accurate circuit breaker status detection on mobile or embedded devices, thereby improving the intelligence level and detection efficiency of avionics experiments.

[0048] (2) This invention introduces a depthwise separable asymmetric convolution module to replace the original convolution module in YOLOv8, constructing a circuit breaker switch recognition model that reduces the number of parameters to 80% of the original. This improvement significantly reduces the computational complexity of the model while maintaining its feature extraction capabilities, enabling it to achieve higher efficiency on resource-constrained mobile or embedded devices. The recognition speed on the same device is improved from 10 seconds to 5 seconds, a 50% increase. Specifically, the depthwise separable asymmetric convolution module significantly reduces the number of parameters while maintaining the model's feature extraction capabilities through depthwise convolution in two directions. This not only optimizes the model structure and improves its efficiency on resource-constrained devices but also provides a new approach to lightweight design of deep learning models.

[0049] (3) Regarding switch status recognition, a MobileNetV3 network with an integrated attention mechanism was used to construct the switch status recognition model. The model can more accurately focus on the key areas of the switch, effectively improving the accuracy of status classification. On the test dataset, the accuracy of status classification increased from 85% to 99%, an improvement of 14 percentage points. This is attributed to the model's ability to better understand and utilize the feature information of the circuit breaker switch. Specifically, the module that integrates spatial attention and channel attention can more effectively capture and utilize the key feature information of the circuit breaker switch, significantly improving the accuracy of circuit breaker status classification. The spatial attention mechanism enables the model to focus on the key areas of the switch, while the channel attention mechanism enhances the model's efficiency in utilizing features. The combination of the two achieves accurate identification of the circuit breaker switch status.

[0050] (4) This invention solves the compatibility problem between different deep learning frameworks by using the ONNX unified format circuit breaker switch identification model and switch status identification model, enabling the model to run seamlessly on multiple devices and platforms, thus enhancing the system's compatibility and versatility. Testing has shown that the system can run on tablets and mobile phones with different Android versions and models. Specifically, this invention combines the lightweight design of YOLOV8m and MobileNetV3 to construct a highly efficient and accurate circuit breaker detection system. This system not only performs excellently in circuit breaker switch identification and status classification but also achieves efficient operation on resource-constrained mobile or embedded devices, meeting the dual requirements of real-time performance and accuracy in practical applications. Attached Figure Description

[0051] Figure 1 This is a flowchart of the intelligent detection method for circuit breakers based on deep learning according to the present invention;

[0052] Figure 2 This is a schematic diagram of the structure of a depth-separable asymmetric convolutional module.

[0053] Figure 3 A schematic diagram of the module that integrates spatial attention and channel attention;

[0054] Figure 4 This is a flowchart of the Android application's operation in Example 2. Detailed Implementation

[0055] Example 1:

[0056] A deep learning-based intelligent detection method for circuit breaker switches, such as... Figure 1 As shown, it includes the following steps:

[0057] Step S1: Acquire circuit breaker images, label and preprocess them to form a training set;

[0058] Step S11: Acquire circuit breaker images under different states and backgrounds, and preprocess them using contrast-limited adaptive histogram equalization (CLAHE) technology. Then, scale the preprocessed images proportionally to 640*640 pixels.

[0059] Step S12: Use the open-source annotation tool LabelImg to manually annotate the preprocessed image, accurately select the position of each circuit breaker switch, and assign it the corresponding category label. Save the annotation results as a YOLO format txt file.

[0060] Step S2: Based on the YOLOV8 model, a depthwise separable asymmetric convolutional module is used to replace the convolutional module of YOLOV8 to obtain the YOLOV8m model; the lightweight YOLOV8m model is trained using the training set to obtain the circuit breaker switch recognition model.

[0061] Step S21: This step optimizes and improves the YOLOv8m network structure by replacing the original standard convolutional modules with depthwise separable asymmetric convolutional modules, such as... Figure 2 As shown, the steps to build a depthwise separable asymmetric convolutional module are as follows:

[0062] Step A1: Perform channel-average segmentation on the input feature map, dividing it into two parts, and send them to two different feature extraction branches for processing;

[0063] Step A2: The upper branch first undergoes a convolution operation with a kernel size of (1, 3) and a stride of 2; then, it undergoes another convolution operation with a kernel size of (3, 1) and a stride of 2; the specific formula is as follows:

[0064]

[0065] in, and represents depthwise convolution with kernel size (3, 1) and stride 2 and depthwise convolution with kernel size (1, 3) and stride 2, respectively. split(·) represents feature map channel splitting operation, and F1 represents the output of the upper branch.

[0066] Step A3: The lower branch first undergoes a convolution operation with a kernel size of (3, 3) and a stride of 1; then, it undergoes another convolution operation with a kernel size of (3, 3) and a stride of 2, as shown in the following formula:

[0067]

[0068] in, and represents depthwise convolution with kernel size (3, 3) and stride 2 and depthwise convolution with kernel size (3, 3) and stride 1, respectively. split(·) represents feature map channel splitting operation, and F2 represents the output of the lower branch.

[0069] Step A4: Combine the results of Step A2 and Step S33 in the channel direction;

[0070] Step A5: Perform a pointwise convolution operation on the concatenated result from Step A4 with a 1×1 high-dimensional convolution kernel to further fuse feature information. The formula is as follows:

[0071]

[0072] in, This indicates a pointwise convolution with a 1×1 kernel, Cat(·) represents the feature map channel direction concatenation operation, and F” represents the output of the depth-separable asymmetric convolution module.

[0073] Step S22: Input the preprocessed image from step S11 and the label data obtained in step S12 into the optimized YOLOv8m network from step S21 for training. In this embodiment, YOLOv8m pre-trained weights on the COCO dataset are used.

[0074] Step S3: Using the circuit breaker switch recognition model trained in Step S2, inference is performed on the preprocessed image from Step S1. Based on the coordinate information of the detected switches in the model output, the image regions of all switches are cropped from the original image. The detailed steps for cropping the image of all switch regions in the original image in Step S3 are as follows:

[0075] Step S31: Obtain the input of YOLOv8m, which includes a total of 7 values: the x-coordinate of the bounding box center, the y-coordinate of the bounding box center, the bounding box width w, the bounding box height h, the confidence score, the class and the class probability.

[0076] Step S32: Convert the normalized coordinates into actual coordinates, using the following formula:

[0077] x actual = x × 640,

[0078] y actuall =y×640,

[0079] width actual =w×640,

[0080] height actual =h×640,

[0081] Where, x actual and y actual The actual center point coordinates of the rectangle, width actual and height actual 640 represents the actual width and height of the rectangle, and 640 represents the size of the original image.

[0082] Step S33: Calculate the coordinates of the top left and bottom right corners of the rectangle based on the center point coordinates and the width and height. The specific formula is as follows:

[0083]

[0084] Where, x min and y min The coordinates of the top-left corner of the rectangle are x max and y max The coordinates are the bottom right corner of the rectangle.

[0085] Step S34: Take a screenshot based on the coordinates of the top left and bottom right corners of the rectangle. The specific formula is as follows:

[0086] cropped_image = image[y min :x min ,y max :x max ];

[0087] Where cropped_image is the cropped image and image is the original image.

[0088] Step S35: Enlarge all the switch images extracted in step S34 to 224×224 and classify them according to their actual state (off or on) to construct a binary classification dataset.

[0089] Step S4: Replace the SE module in the MobileNetV3 network with a module that integrates spatial attention and channel attention to obtain a MobileNetV3 network with an integrated attention mechanism. Input the dataset into the MobileNetV3 network with the integrated attention mechanism for training to obtain a switch state recognition model for detecting the switch state of circuit breakers.

[0090] Step S41: This step innovatively improves the architecture of the classification network MobileNetV3. The original Squeezze-and-Excitation (SE) module is replaced with a new module that integrates spatial attention and channel attention mechanisms, such as... Figure 3 As shown, the steps to build a new module that integrates spatial attention and channel attention mechanisms are as follows:

[0091] Step B1: Feed the input feature vector into two different attention mechanism branches for processing;

[0092] Step B2: The upper branch uses a channel attention mechanism. First, the input feature F is processed by global max pooling and global average pooling, then fed into two fully connected layers for training. Finally, it is broadcast and summed element-wise to obtain α, as shown in the following formula:

[0093]

[0094] Among them, w (1) b (1) The first layer weight matrix and bias; w (2) b (2) Here, G represents the second-layer weight matrix and bias; G(·) denotes the ReLU activation function; P gm (·) and P ga (·) represent global max pooling and global average pooling, respectively; This indicates that the broadcast is added element by element.

[0095] Step B3: The next branch is a spatial attention mechanism. First, max pooling and average pooling are applied along the channels to obtain a 2×H×W feature map. Then, a 5×5 standard convolutional layer is used to generate a spatial attention feature map. The specific formula is as follows:

[0096]

[0097] Among them, P cm (·) and P ca (·) represents average pooling and max pooling along the channel direction, respectively; Cat(·) represents splicing along the channel direction; This indicates a standard convolution with a 5×5 kernel.

[0098] Step B4: Broadcast the results obtained from steps B2 and B3 and multiply them element-wise to obtain the feature map of C×H×W. The specific formula is as follows:

[0099] γ = α⊙β;

[0100] Here, ⊙ represents element-wise multiplication of broadcast data.

[0101] Step S42: Input the binary classification dataset constructed in step S35 into the optimized MobileNetV3 network in step S41 for training. To improve training efficiency and model performance, this embodiment uses the pre-trained weights of MobileNetV3 on the ImageNet dataset as initialization parameters, ultimately obtaining a switch state recognition model that can accurately identify switch states (open or closed).

[0102] Step S5: To facilitate deployment and operation on mobile devices, the circuit breaker switch recognition model trained in step S2 and the switch state recognition model trained in step S4 are converted to ONNX (OpenNeuralNetWorkExchange) format using the built-in function torch.onnx.export of PyTorch.

[0103] Example 2:

[0104] A deep learning-based intelligent detection system for circuit breaker switches, such as Figure 4 As shown, it includes a search module, a data preprocessing module, a voice playback module, a camera module, a model processing module, and a results display module.

[0105] ① The search module allows for step-by-step selection by project, batch, and flight. After clicking the "OK" button, you can further select a specific experiment name and circuit breaker number, and the system will display the matching experiment template.

[0106] ② The data preprocessing module uses the OpenCV for Android toolkit to preprocess the captured images, including contrast-limited adaptive histogram equalization and proportional scaling, to optimize image quality;

[0107] ③ The voice broadcast module utilizes offline text-to-speech technology, combined with the Android native text-to-speech interface and iFlytek offline speech engine, to achieve the voice broadcast function;

[0108] ④ The camera module provides high-quality image capture capabilities by calling the CameraX API;

[0109] ⑤ The model processing module uses the ONNX Runtime inference engine to run ONNX format models (circuit breaker switch recognition model and switch status recognition model) and perform in-depth analysis on the captured circuit breaker images to identify their switch status.

[0110] ⑥ The results display module can highlight the parts of the captured circuit breaker image that do not meet the preset requirements, and intuitively point out the potential problem areas by marking them with a conspicuous red box.

[0111] During use, after a photo is taken under certain conditions through the Android application interface, the system automatically runs a deep learning model in the background to analyze and predict the photo in real time, obtaining the prediction result. Subsequently, the system automatically compares this prediction result with a pre-set experimental template to evaluate whether the photographed object meets the requirements of the template. Figure 4 As shown, the operation of this system specifically includes the following steps:

[0112] Step T1: First, select the project, batch, and flight in sequence, and then click the "OK" button to complete the initial information input;

[0113] Step T2: Next, by selecting a specific experiment name and circuit breaker number, the system will automatically present the matching experiment template.

[0114] Step T3: This invention also provides a voice broadcast function. If needed, the user only needs to tap the "speaker" icon on the screen to broadcast the experimental template content in voice form;

[0115] Step T4: Enter shooting mode by clicking the "Open Camera" button. In low-light conditions, you can click the "Turn on Flash" button to activate the flash and ensure image quality.

[0116] Step T5: This invention designs an auxiliary shooting frame. Simply place the actual circuit breaker inside the frame, and the system will automatically focus. Then, click the "Take Photo" button.

[0117] Step T6: The system automatically runs the ONNX format models (circuit breaker switch identification model and switch status identification model) through the ONNX Runtime inference engine to perform in-depth analysis of the object captured in Step T5. After the analysis is completed, the system will automatically return to the initial page. At the same time, any parts that do not meet the preset requirements will be marked with a prominent red box.

[0118] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for intelligent detection of circuit breaker switches based on deep learning, characterized in that, Includes the following steps: Step S1: Acquire circuit breaker images, label and preprocess them to form a training set; Step S2: Based on the YOLOV8 model, a depthwise separable asymmetric convolutional module is used to replace the convolutional module of YOLOV8 to obtain the YOLOV8m model; the lightweight YOLOV8m model is trained using the training set to obtain the circuit breaker switch recognition model. Step S3: Input the training set into the circuit breaker switch recognition model and output the rectangular frame coordinate information of the switch. Extract the switch image from the original circuit breaker image according to the output rectangular frame coordinates, and classify the data according to the switch state to form a dataset. Step S4: Replace the SE module in the MobileNetV3 network with a module that integrates spatial attention and channel attention to obtain a MobileNetV3 network with an integrated attention mechanism; input the dataset into the MobileNetV3 network with an integrated attention mechanism for training to obtain the switch state recognition model. Step S5: Convert the circuit breaker switch identification model in step S2 and the switch status identification model in step S4 into ONNX format and deploy them on the device.

2. The intelligent detection method for circuit breakers based on deep learning according to claim 1, characterized in that, In step S2, the processing of the depth-separable asymmetric convolution module includes the following steps: Step A1: Perform channel-average segmentation on the input feature map, dividing it into two parts, and send them to the upper and lower branches of feature extraction for processing respectively; Step A2: In the upper branch, first, a convolution operation with kernel size (1, 3) and stride 2 is performed; then, a convolution operation with kernel size (3, 1) and stride 2 is performed to obtain feature F1; Step A3: In the next branch, first, a convolution operation with a kernel size of (3, 3) and a stride of 1 is performed; then, a convolution operation with a kernel size of (3, 3) and a stride of 2 is performed to obtain feature F2. Step A4: Concatenate features F1 and F2 along the channel direction; Step A5: Perform a pointwise convolution operation on the concatenated features with a 1×1 high-dimensional convolution kernel to fuse feature information and obtain feature F.

3. The intelligent detection method for circuit breakers based on deep learning according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Obtain the output of the circuit breaker switch identification model (x, y, w, h, confidence score, category, category probability); Where (x, y) are the coordinates of the center of the bounding box; w and h are the width and height of the bounding box, respectively; Step S32: Transform the normalized coordinates (x, y) into the actual coordinates (x, y). actual y actual ): x actual =x×640, and actual =y×640, width actual =w×640, height actual =h×640, Among them, (x actual y actual () represents the actual center point coordinates of the rectangle; width actual and height actual These are the actual width and height of the rectangle, respectively. Step S33: Calculate the coordinates (x, y) of the top left corner of the rectangle. min y min ) and the coordinates of the lower right corner (x) max y max ): Step S34: Extract an image from the original image based on the coordinates of the top left and bottom right corners of the rectangle: cropped_image=image[y min :x min ,y max :x max ]; Where: cropped_image is the cropped image. "image" refers to the original image.

4. The intelligent detection method for circuit breakers based on deep learning according to claim 3, characterized in that, In step S3, the captured switch image is enlarged to 224×224 and classified according to its actual state to construct a binary classification dataset.

5. The intelligent detection method for circuit breakers based on deep learning according to claim 1, characterized in that, In step S4, the processing of the module that fuses spatial attention and channel attention includes the following steps: Step B1: The input feature vector is fed into the channel attention mechanism branch and the spatial attention mechanism branch for processing, respectively; Step B2: In the channel attention mechanism branch, the input feature F is first processed by global max pooling and global average pooling, then fed into two fully connected layers for training, and finally broadcast and added element by element to obtain the channel attention feature map α. Step B3: In the spatial attention mechanism branch, the input feature F is first max pooled and average pooled along the channel direction to obtain a 2×H×W feature map, and then a spatial attention feature map β is generated by passing it through a 5×5 standard convolutional layer. Step B4: Broadcast the channel attention feature map α and the spatial attention feature map β and multiply them element-wise to obtain the C×H×W feature map γ.

6. A deep learning-based intelligent detection system for circuit breakers, used to implement the deep learning-based intelligent detection method for circuit breakers as described in any one of claims 1-5, characterized in that, It includes a camera module, a data preprocessing module, and a model processing module; the camera module is used to acquire images, the data preprocessing module is used to preprocess the captured images, and the model processing module is used to analyze the preprocessed circuit breaker images and identify their switching states using an ONNX format circuit breaker switch identification model and switch status identification model.

7. The intelligent detection system for circuit breakers based on deep learning according to claim 6, characterized in that, It also includes a search module, a voice playback module, and a result display module. The search module is used to search for and determine the experiment name and circuit breaker number so as to display the matching experiment template. The voice playback module is used to implement the voice broadcast function. The result display module is used to highlight potential problem areas in the captured circuit breaker image.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements a deep learning-based intelligent detection method for circuit breakers as described in any one of claims 1-5.

9. An electronic device, characterized in that, It includes a memory and a processor; the memory stores a computer program; the processor is used to execute the computer program in the memory to implement the intelligent detection method for circuit breakers based on deep learning according to any one of claims 1-5.

Images