Machine vision-based intelligent monitoring and risk early warning system for uninterrupted operation of distribution network

Abstract

The invention relates to the field of power distribution safety, and discloses a machine vision-based intelligent monitoring and risk early warning system for uninterrupted operation of a distribution network, which is used for solving the problem that an operator is in danger due to the fact that a protective device is not worn correctly during manual power distribution operation. The method comprises the following steps: acquiring a state image of a protector through a camera device, identifying an operator through an identification network model, performing corresponding information transmission according to an identification result, acquiring the state image of the protector worn by the operator through the camera device, analyzing the state image to obtain a position reasonability coefficient and a fastening coefficient, and transmitting the position reasonability coefficient and the fastening coefficient to the operator. And evaluating according to the position reasonable coefficient and the fastening coefficient to obtain a safety index, and performing corresponding early warning according to the safety index, thereby effectively improving the safety of the power distribution system.

Description

Technical Field

[0001] The present invention relates to the field of power distribution safety, and more specifically to a machine vision-based intelligent monitoring and risk warning system for non-stop power supply operations in distribution networks. Background Art

[0002] Machine vision is a technology that uses computers and image processing technology to simulate human vision. It obtains image or video data through cameras or sensors, and analyzes and understands these data through image processing algorithms, thereby realizing functions such as automated detection, recognition, measurement and control. Machine vision is widely used in industrial automation, quality inspection, robot navigation and other fields, providing high-precision and high-efficiency visual perception capabilities.

[0003] With the improvement of industrial safety standards, correctly wearing safety protection equipment has become an important factor in ensuring the safety of workers. However, traditional manual inspection methods have problems such as low efficiency and strong subjectivity. With the development of deep learning and computer vision technology, image-based automated detection has become a solution. In the existing technology, although there are many image detection and target recognition methods, there is still much room for improvement in accurately locating safety protection equipment and analyzing wearing conditions.

[0004] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent monitoring and risk warning system for non-stop power supply operations of distribution networks based on machine vision to solve the problems existing in the above-mentioned background technology.

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

[0007] The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision includes:

[0008] The protective gear recognition module is used to construct a recognition network model through a convolutional neural network, and to identify the operator through the recognition network model. If the recognition is successful, the recognition result is transmitted to the status data acquisition module; if the recognition is unsuccessful, the recognition result is transmitted to the safety warning module;

[0009] A status data acquisition module is used to acquire a status image of the protective gear through a camera device, wherein the status image is a status image of the operator wearing the protective gear, and transmit the status image to a data analysis module;

[0010] A data analysis module is used to receive the state image transmitted by the state data acquisition module, analyze the state image, and transmit the analysis result to the data evaluation module;

[0011] The data evaluation module is used to receive the analysis results transmitted by the data analysis module, evaluate the analysis results to obtain a safety index, and transmit the safety index to the safety warning module;

[0012] The safety warning module is used to receive the data transmitted by the protective gear identification module and the data evaluation module, and issue corresponding warnings.

[0013] Preferably, the step of constructing a recognition network model through a convolutional neural network is:

[0014] Collect a large amount of image data of workers and their safety protection equipment into the dataset, which should include a variety of different protective equipment, label each image, and pre-process the image;

[0015] Input image data into the input layer of the convolutional neural network with a size of H×W×C, where H and W represent the height and width of the image, and C is the number of channels;

[0016] In the convolutional layer of the convolutional neural network, multiple convolution kernels are applied to perform convolution operations on the input image to extract local features. The expression is: where y i,j,k It is represented as the value of the kth channel of the output feature map at position (i, j), x is the input image, w is the convolution kernel, and b k is the bias, M and N are the sizes of the convolution kernel;

[0017] Use the ReLU activation function to perform nonlinear transformation on the convolution result; flatten the feature map of the previous layer into a one-dimensional vector through maximum pooling, multiply it with the weight matrix, and output it to the classifier;

[0018] The cross entropy loss function is used to measure the difference between the model output and the true label, and the stochastic gradient descent method is used to optimize the algorithm;

[0019] The training data is input into the model, and the model parameters are optimized through multiple iterations to minimize the loss function. The validation set is used to evaluate the relevant performance of the model, which includes accuracy, precision, recall and F1-score indicators. The model performance is obtained by comprehensive evaluation of the relevant performance, and its expression is P = f(p1, p2, p3, p4). The model performance is compared with the threshold. If the model performance is greater than the threshold, the recognition network model is output. If the model performance is less than the threshold, training continues until the model performance is greater than the threshold.

[0020] Preferably, the step of collecting the status image of the protective gear is:

[0021] Use the camera to capture the image of the worker wearing protective gear and convert the color image into a grayscale image, which is expressed as I gray=0.2989×R+0.5870×G+0.1140×B, where R, G, and B are red, green, and blue channel values, respectively. Use Gaussian filtering to smooth the image.

[0022] Use the Canny edge detection algorithm to detect the edge of the protective gear and extract the contour information. The expression is: Among them I gradient (x, y) is the gradient amplitude of the image, Threshold is the threshold of edge detection;

[0023] The protective gear contour is extracted from the edge detection results.

[0024] Preferably, the step of analyzing the state image is:

[0025] According to the extracted protective gear contour, key feature points of the protective gear are determined, and key points on the protective gear and relevant key points of the human body are identified, wherein the key points of the human body are the parts of the human body where the protective gear is worn;

[0026] The wearing angle of the protective gear is calculated based on the key points. The calculation formula is: Where (x1, y1) and (x2, y2) are the coordinates of the key points of the protective gear, and θ is the calculated wearing angle;

[0027] The reasonable position coefficient is calculated based on the calculated wearing angle, and its calculation formula is: Where PR is the reasonable position coefficient, θ is the current wearing angle, AG id Indicates the ideal wearing angle, AG max Expressed as the maximum allowable deviation angle;

[0028] According to the extracted protective gear contour, relevant data of the protective gear elastic band are obtained, wherein the relevant data include the length and area data of the elastic band, and the tightening coefficient is calculated according to the relevant data of the elastic band.

[0029] Preferably, the step of calculating the tightening coefficient according to the relevant data of the elastic band is:

[0030] Calculate the length of the elastic band, the calculation formula is: Where (x3, y3) and (x4, y4) are the coordinates of the two ends of the elastic band, and L is the length of the elastic band;

[0031] Calculate the area of ​​the polygon covered by the elastic band, the calculation formula is: Where A is the area covered by the elastic band, (x i ,y i ) are the coordinates of the polygon vertices;

[0032] The tightening coefficient is calculated based on the length of the elastic band and the coverage area of ​​the elastic band. The calculation formula is: Where FT is the tightening factor, L max is the maximum length of the elastic band in the relaxed state, L current is the length of the elastic band currently measured, A current It is represented by the current coverage area of ​​the elastic band, A loose is the area of ​​the elastic band in the relaxed state.

[0033] Preferably, the step of evaluating the analysis results to obtain a safety index is:

[0034] The safety index is calculated by weighted summation of the reasonable position coefficient and the tightening coefficient. The calculation formula is SF=a1×PR+a2×FT, where SF represents the safety index, PR represents the reasonable position coefficient, FT represents the tightening coefficient, and a1 and a2 represent the weight coefficients of the reasonable position coefficient and the tightening coefficient.

[0035] Preferably, the steps of receiving the data transmitted by the protective gear identification module and the data evaluation module and performing corresponding early warning are:

[0036] If the information transmitted by the protective gear identification module is received, it is determined that the operator is not currently wearing protective gear, and an early warning is issued to the relevant management personnel to remind the operator to wear protective gear in time;

[0037] If the safety index transmitted by the data evaluation module is received, the safety index is compared with the preset threshold. If the safety index is greater than the preset threshold, it is determined that the current operator is wearing protective gear safely; if the safety index is less than the preset threshold, it is determined that the current operator is wearing protective gear in a dangerous situation, and an early warning is issued to the relevant management personnel to remind the operator to wear protective gear correctly. Technical effects and advantages of the present invention:

[0038] A recognition network model is constructed through a convolutional neural network, and the workers are identified through the recognition network model. Corresponding information is transmitted according to the recognition result, and a state image of the protective gear is collected through a camera device. The state image is a state image of the workers wearing the protective gear. The state image is analyzed to obtain a reasonable position coefficient and a tightening coefficient. The safety index is evaluated based on the reasonable position coefficient and the tightening coefficient, and corresponding warnings are issued based on the safety index, thereby effectively improving the safety of the power distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is the overall flow chart of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. In addition, the forms of the various structures recorded in the following embodiments are merely illustrative. The machine vision-based intelligent monitoring and risk warning system for non-stop power supply operations in the distribution network involved in the present invention is not limited to the various structures recorded in the following embodiments. All other implementations obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0041] The present invention provides a distribution network non-stop operation intelligent monitoring and risk warning system based on machine vision, the system comprising:

[0042] The protective gear recognition module is used to construct a recognition network model through a convolutional neural network, and to identify the operator through the recognition network model. If the recognition is successful, the recognition result is transmitted to the status data acquisition module; if the recognition is unsuccessful, the recognition result is transmitted to the safety warning module;

[0043] The convolutional neural network is a deep learning model specifically designed for processing image data. It automatically extracts the spatial features of an image through a series of convolutional layers, pooling layers, and fully connected layers. The convolutional layer applies filters in a sliding window manner to extract local features, while the pooling layer is used to reduce the dimension of the feature map and reduce the amount of calculation. CNN can effectively identify complex patterns and structures in images and is widely used in tasks such as image classification, object detection, and image segmentation.

[0044] The convolutional neural network model can automatically extract features from images and perform efficient pattern recognition. It has higher accuracy and robustness than traditional manual feature extraction methods. This ensures that the recognition system can accurately determine whether the operator is wearing protective gear; the convolutional neural network model can automatically process large amounts of image data and make judgments in real time, reducing the workload and delay of manual inspections, and improving monitoring efficiency and response speed; the convolutional neural network model can handle different image changes, such as different lighting conditions, angle changes, and background interference, and has strong adaptability. This enables the system to maintain stable performance in a variety of actual working environments; through an automated image recognition system, it can reduce recognition errors caused by human factors and improve the accuracy and consistency of safety inspections.

[0045] The convolutional neural network model can efficiently process and analyze large-scale image data. It is suitable for scenarios that require large-scale monitoring, such as construction sites or production lines. It can monitor the wearing of protective gear by multiple workers in real time. The system can instantly detect when protective gear is not worn, and issue warnings or feedback in time to help on-site managers take quick measures to ensure the safety of workers. The convolutional neural network model can gradually improve its recognition ability and accuracy through continuous training and data updates, and adapt to new types of protective gear or changing wearing standards.

[0046] In this embodiment, it should be specifically explained that the steps of constructing a recognition network model through a convolutional neural network are:

[0047] Collect a large amount of image data with workers and their safety protection equipment. The data set should include a variety of different protective equipment, such as helmets, goggles, etc., label each image to indicate whether the corresponding protective equipment is worn correctly, and preprocess the image, including resizing, normalization, data enhancement, etc., to improve the generalization ability of the model;

[0048] Input image data into the input layer of the convolutional neural network, usually of size H×W×C, where H and W represent the height and width of the image, and C is the number of channels, usually 3, i.e. RGB;

[0049] In the convolutional layer of the convolutional neural network, multiple convolution kernels are applied to perform convolution operations on the input image to extract local features. The expression is: where y i,j,k It is represented as the value of the kth channel of the output feature map at position (i, j), x is the input image, w is the convolution kernel, and b k is the bias, M and N are the sizes of the convolution kernel;

[0050] Use the ReLU activation function to perform a nonlinear transformation on the convolution result, which is expressed as ReLU(x)=max(0,x); through the maximum pooling or average pooling operation, reduce the dimension of the feature map while retaining the main features, which is expressed as y i,j,k =max(x i+p,j+q,k ), where p and q represent the size of the pooling window; the feature map of the previous layer is flattened into a one-dimensional vector, multiplied by the weight matrix, and output to the classifier, which is expressed as y = softmax (W x +b), where W is the weight matrix, b is the bias term, and the softmax function converts the output into a probability distribution;

[0051] The ReLU activation function is a commonly used nonlinear activation function that truncates negative values ​​in the input to zero and keeps positive values ​​unchanged. Specifically, the output of ReLU is the maximum value of the input value and zero. ReLU introduces nonlinear characteristics, allowing neural networks to learn more complex patterns, while having the advantages of simple calculation and fast convergence speed.

[0052] The cross entropy loss function is used to measure the difference between the model output and the true label, and its expression is Where N is the number of samples, C is the number of categories, and y ij is the true label, To predict the probability, the algorithm is optimized using stochastic gradient descent;

[0053] The cross entropy loss function is a function used to measure the difference between the probability distribution of the classification model output and the true label distribution. It calculates the entropy difference between the two probability distributions. Specifically, it is obtained by taking the negative of the logarithmic probability of the true label and summing them. The closer the predicted result is to the true label, the smaller the cross entropy loss value. Cross entropy loss is widely used in multi-classification tasks and is one of the commonly used loss functions in deep learning model training.

[0054] The training data is input into the model, and the model parameters are optimized through multiple iterations to minimize the loss function. The validation set is used to evaluate the relevant performance of the model, which includes indicators such as accuracy, precision, recall rate and F1-score. The model performance is obtained by comprehensive evaluation of the relevant performance, and its expression is P=f(p1, p2, p3, p4). The model performance is compared with the threshold. If the model performance is greater than the threshold, the recognition network model is output. If the model performance is less than the threshold, training continues until the model performance is greater than the threshold.

[0055] The accuracy is an indicator to measure the performance of the model, which indicates the proportion of the number of samples correctly predicted by the model to the total number of samples. Specifically, the accuracy is equal to the ratio of the number of correctly classified samples to the total number of all samples, reflecting the overall prediction ability of the model on the entire data set.

[0056] The precision is one of the indicators to measure the performance of the classification model. It indicates the proportion of samples predicted by the model to be positive that are actually positive. The precision is equal to the ratio of the number of samples correctly predicted to be positive to the total number of samples predicted by the model to be positive. It reflects the accuracy of the model in positive prediction. Especially when there are many false positive errors, the precision is an important evaluation indicator.

[0057] The recall rate indicates the proportion of samples that are actually positive that are correctly predicted as positive by the model. Specifically, the recall rate is equal to the ratio of the number of samples correctly predicted as positive to the total number of actual positive samples, reflecting the ability of the model in detecting positive samples. Especially when there are many false negative errors, the recall rate is an important evaluation indicator.

[0058] The F1-score is a comprehensive indicator of model performance. It is the harmonic mean of precision and recall, and aims to balance the relationship between the two. When both precision and recall perform well, the F1-score will be higher. The F1-score is particularly suitable for datasets with imbalanced categories because it comprehensively considers the impact of false positives and false negatives.

[0059] These indicators provide a multi-dimensional evaluation of model performance. Accuracy measures the correctness of the overall prediction, while precision, recall, and F1-score each focus on how the model performs when dealing with different categories (especially target categories). This comprehensive evaluation can provide a more comprehensive understanding of the effectiveness of the model in various situations; in the case of class imbalance (for example, the ratio of positive and negative samples in target detection is very different), using only accuracy may be misleading. Precision and recall better reflect the performance of the model on a few categories, especially when the target categories (such as the state of not wearing protective gear) are relatively small, and can provide more valuable information.

[0060] A status data acquisition module is used to acquire a status image of the protective gear through a camera device, wherein the status image is a status image of the operator wearing the protective gear, and transmit the status image to a data analysis module;

[0061] In this embodiment, it should be specifically explained that the step of collecting the status image of the protective gear is:

[0062] Use the camera to capture the image of the worker wearing protective gear and convert the color image into a grayscale image, which is expressed as I gray =0.2989×R+0.5870×G+0.1140×B, where R, G, and B are the red, green, and blue channel values, respectively. To simplify subsequent processing, use Gaussian filtering or median filtering to smooth the image and reduce noise;

[0063] Use the Canny edge detection algorithm to detect the edge of the protective gear and extract the contour information. Its expression is: Among them I gradient (x, y) is the gradient amplitude of the image, Threshold is the threshold of edge detection;

[0064] The protective gear contour is extracted from the edge detection results.

[0065] The Canny edge detection algorithm is a commonly used image processing technique for detecting edges in images. It is implemented through five steps: image smoothing (using a Gaussian filter to reduce noise), calculating image gradients (identifying possible edges), non-maximum suppression (eliminating non-edge pixels), dual threshold detection (distinguishing strong edges from weak edges), and edge connection (connecting weak edges to strong edges), effectively detecting continuous edges in images, with good noise suppression capabilities and edge positioning accuracy.

[0066] A data analysis module is used to receive the state image transmitted by the state data acquisition module, analyze the state image, and transmit the analysis result to the data evaluation module;

[0067] In this embodiment, it should be specifically explained that the steps of analyzing the state image are:

[0068] According to the extracted protective gear contour, determine the key feature points of the protective gear, such as the edge points and buckle centers of the protective gear, identify the key points on the protective gear and the relevant key points of the human body, where the key points of the human body are the parts of the human body where the protective gear is worn, such as the center of the head or the shoulder position;

[0069] The wearing angle of the protective gear is calculated based on the key points. The calculation formula is: Where (x1, y1) and (x2, y2) are the coordinates of the key points of the protective gear, and θ is the calculated wearing angle;

[0070] The reasonable position coefficient is calculated based on the calculated wearing angle, and its calculation formula is: Where PR is the reasonable position coefficient, θ is the current wearing angle, AG id Indicates the ideal wearing angle, AG max Expressed as the maximum allowable deviation angle;

[0071] According to the extracted protective gear contour, relevant data of the protective gear elastic band are obtained, wherein the relevant data include the length and area data of the elastic band, and the tightening coefficient is calculated according to the relevant data of the elastic band.

[0072] In this embodiment, it should be specifically explained that the step of calculating the tightening coefficient according to the relevant data of the elastic band is:

[0073] Calculate the length of the elastic band and determine whether the elastic band is tight. The calculation formula is: Where (x3, y3) and (x4, y4) are the coordinates of the two ends of the elastic band, and L is the length of the elastic band;

[0074] Calculate the area of ​​the polygonal region covered by the elastic band. After tightening the band, the area should be reduced. The calculation formula is: Where A is the area covered by the elastic band, (xi ,y i ) are the coordinates of the polygon vertices;

[0075] The tightening coefficient is calculated based on the length of the elastic band and the coverage area of ​​the elastic band. The calculation formula is: Where FT is the tightening factor, L max is the maximum length of the elastic band in the relaxed state, L current is the length of the elastic band currently measured, A current It is represented by the current coverage area of ​​the elastic band, A loose is the area of ​​the elastic band in the relaxed state.

[0076] The data evaluation module is used to receive the analysis results transmitted by the data analysis module, evaluate the analysis results to obtain a safety index, and transmit the safety index to the safety warning module;

[0077] By real-time monitoring and analyzing the wearing of protective gear, it is possible to quickly identify whether the workers are wearing protective gear correctly, such as helmets, goggles, earplugs, etc. This helps prevent safety accidents and injuries caused by improper wearing of protective gear; wearing protective gear correctly can effectively reduce the risk of injury, and by analyzing the wearing of protective gear, non-compliant wearing behaviors can be corrected in a timely manner, thereby reducing the probability of accidents; the safety analysis system can provide real-time wearing feedback to help safety managers identify problems and take measures in a timely manner to optimize on-site safety management; by accumulating protective gear wearing data, the safety management department can conduct data analysis, identify common wearing problems and trends, and make improvement measures based on the data.

[0078] In this embodiment, it should be specifically explained that the step of evaluating the analysis results to obtain the safety index is:

[0079] The safety index is calculated by weighted summation of the reasonable position coefficient and the fastening coefficient, and the calculation formula is SF = a1 × PR + a2 × FT, where SF is represented by the safety index and PR is represented by the reasonable position coefficient. When the reasonable position coefficient is high, it means that the wearing angle of the protective gear is close to the recommended ideal position, which can provide better protection and higher safety. On the contrary, a low reasonable position coefficient indicates that the wearing angle deviates from the standard, which may lead to a reduction in the protective function of the protective gear, and the safety index is also reduced. Therefore, by improving the reasonable position coefficient, the safety performance and wearing effect of the protective gear can be effectively improved. FT is represented by the fastening coefficient. When the fastening coefficient is high, it indicates that the protective gear is worn correctly and firmly, so that it can protect the wearer more effectively, and the safety index is also improved accordingly. On the contrary, a low fastening coefficient means that the protective gear may be loose or improperly worn, which will reduce its protective performance and cause the safety index to decrease. Therefore, by increasing the tightening coefficient, the protective ability of the protective gear can be significantly enhanced, thereby improving the overall safety. a1 and a2 are weight coefficients of the reasonable position coefficient and the tightening coefficient, and this embodiment does not perform specific calculations on the specific values ​​of a1 and a2.

[0080] The safety warning module is used to receive the data transmitted by the protective gear identification module and the data evaluation module, and issue corresponding warnings.

[0081] In this embodiment, it should be specifically explained that the steps of receiving the data transmitted by the protective gear identification module and the data evaluation module and performing corresponding early warning are:

[0082] If the information transmitted by the protective gear identification module is received, it is determined that the operator is not currently wearing protective gear, and an early warning is issued to the relevant management personnel to remind the operator to wear protective gear in time;

[0083] If the safety index transmitted by the data evaluation module is received, the safety index will be compared with the preset threshold. If the safety index is greater than the preset threshold, it is determined that the current operator is wearing protective gear in a relatively safe manner; if the safety index is less than the preset threshold, it is determined that the current operator is wearing protective gear in a relatively dangerous manner, and an early warning will be issued to the relevant management personnel to remind the operators to wear protective gear correctly.

[0084] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0085] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. Intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision, characterized by: The system comprises: The protective gear recognition module is used to construct a recognition network model through a convolutional neural network, and to identify the operator through the recognition network model. If the recognition is successful, the recognition result is transmitted to the status data acquisition module; if the recognition is unsuccessful, the recognition result is transmitted to the safety warning module; A status data acquisition module is used to acquire a status image of the protective gear through a camera device, wherein the status image is a status image of the operator wearing the protective gear, and transmit the status image to a data analysis module; A data analysis module is used to receive the state image transmitted by the state data acquisition module, analyze the state image, and transmit the analysis result to the data evaluation module; The data evaluation module is used to receive the analysis results transmitted by the data analysis module, evaluate the analysis results to obtain a safety index, and transmit the safety index to the safety warning module; The safety warning module is used to receive data transmitted by the protective gear identification module and the data evaluation module, and issue corresponding warnings.

2. The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision according to claim 1 is characterized by: The steps of constructing a recognition network model through a convolutional neural network are: Collect a large amount of image data of workers and their safety protection equipment into the dataset, which should include a variety of different protective equipment, label each image, and pre-process the image; Input image data into the input layer of the convolutional neural network with a size of H×W×C, where H and W represent the height and width of the image, and C is the number of channels; In the convolutional layer of the convolutional neural network, multiple convolution kernels are applied to perform convolution operations on the input image to extract local features. The expression is: where y i,j,k It is represented as the value of the kth channel of the output feature map at position (i, j), x is the input image, w is the convolution kernel, and b k is the bias, M and N are the sizes of the convolution kernel; Use the ReLU activation function to perform nonlinear transformation on the convolution result; flatten the feature map of the previous layer into a one-dimensional vector through maximum pooling, multiply it with the weight matrix, and output it to the classifier; The cross entropy loss function is used to measure the difference between the model output and the true label, and the stochastic gradient descent method is used to optimize the algorithm; The training data is input into the model, and the model parameters are optimized through multiple iterations to minimize the loss function. The validation set is used to evaluate the relevant performance of the model, which includes accuracy, precision, recall and F1-score indicators. The model performance is obtained by comprehensive evaluation of the relevant performance, and its expression is P = f(p1, p2, p3, p4). The model performance is compared with the threshold. If the model performance is greater than the threshold, the recognition network model is output. If the model performance is less than the threshold, training continues until the model performance is greater than the threshold.

3. The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision according to claim 1 is characterized by: The steps of collecting the status image of the protective gear are: Use the camera to capture the image of the worker wearing protective gear and convert the color image into a grayscale image, which is expressed as I gray =0.2989×R+0.5870×G+0.1140×B, where R, G, and B are red, green, and blue channel values, respectively. Use Gaussian filtering to smooth the image. Use the Canny edge detection algorithm to detect the edge of the protective gear and extract the contour information. The expression is: Among them I gradient (x, y) is the gradient amplitude of the image, Threshold is the threshold of edge detection; The protective gear contour is extracted from the edge detection results.

4. The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision according to claim 1 is characterized by: The steps of analyzing the state image are as follows: According to the extracted protective gear contour, key feature points of the protective gear are determined, and key points on the protective gear and relevant key points of the human body are identified, wherein the key points of the human body are the parts of the human body where the protective gear is worn; The wearing angle of the protective gear is calculated based on the key points. The calculation formula is: Where (x1, y1) and (x2, y2) are the coordinates of the key points of the protective gear, and θ is the calculated wearing angle; The reasonable position coefficient is calculated based on the calculated wearing angle, and its calculation formula is: Where PR is the reasonable position coefficient, θ is the current wearing angle, AG id Indicates the ideal wearing angle, AG max Expressed as the maximum allowable deviation angle; According to the extracted protective gear contour, relevant data of the protective gear elastic band are obtained, wherein the relevant data include the length and area data of the elastic band, and the tightening coefficient is calculated according to the relevant data of the elastic band.

5. The machine vision-based intelligent monitoring and risk warning system for non-stop power distribution network operations according to claim 4 is characterized by: The steps of calculating the tightening coefficient according to the relevant data of the elastic band are: Calculate the length of the elastic band, the calculation formula is: Where (x3, y3) and (x4, y4) are the coordinates of the two ends of the elastic band, and L is the length of the elastic band; Calculate the area of ​​the polygon covered by the elastic band, the calculation formula is: Where A is the area covered by the elastic band, (x i ,y i ) are the coordinates of the polygon vertices; The tightening coefficient is calculated based on the length of the elastic band and the coverage area of ​​the elastic band. The calculation formula is: Where FT is the tightening factor, L max is the maximum length of the elastic band in the relaxed state, L current is the length of the elastic band currently measured, A current Represents the current coverage area of ​​the elastic band, A loose is the area of ​​the elastic band in the relaxed state.

6. The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision according to claim 1 is characterized by: The steps of evaluating the analysis results to obtain the safety index are: The safety index is calculated by weighted summation of the reasonable position coefficient and the tightening coefficient. The calculation formula is SF=a1×PR+a2×FT, where SF represents the safety index, PR represents the reasonable position coefficient, FT represents the tightening coefficient, and a1 and a2 represent the weight coefficients of the reasonable position coefficient and the tightening coefficient.

7. The intelligent monitoring and risk warning system for non-stop power supply operation of distribution network based on machine vision according to claim 1 is characterized by: The steps of receiving the data transmitted by the protective gear identification module and the data evaluation module and performing corresponding early warning are as follows: If the information transmitted by the protective gear identification module is received, it is determined that the operator is not currently wearing protective gear, and an early warning is issued to the relevant management personnel to remind the operator to wear protective gear in time; If the safety index transmitted by the data evaluation module is received, the safety index will be compared with the preset threshold. If the safety index is greater than the preset threshold, it is determined that the current operator is wearing protective gear safely; if the safety index is less than the preset threshold, it is determined that the current operator is wearing protective gear in a dangerous situation, and an early warning is issued to the relevant management personnel to remind the operators to wear protective gear correctly.

Images