Extra-high-voltage power grid visual high-altitude operation safety monitoring system and method

By deploying a visual aerial operation safety monitoring system at the aerial operation site of the ultra-high voltage power grid, real-time monitoring and analysis of image data, the safety of aerial operation is solved, timely correction of violations and prediction of safety risks is achieved, and operation safety is improved.

CN119068255APending Publication Date: 2024-12-03NINGXIA POWER TRANSMISSION ENG CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411182063.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Safety problems in high-altitude operations of ultra-high voltage power grids are becoming increasingly prominent, and accidents occur frequently, so it is difficult for existing technologies to effectively ensure the safety of high-altitude operations.

Method used

Design a visual high-altitude operation safety monitoring system for ultra-high voltage power grids, including image acquisition module, backend database and backend server. Through real-time image acquisition, processing and identification, combined with data analysis and alarm units, we can judge the safety status of the high-altitude operation site and timely Correct violations.

Benefits of technology

Real-time security monitoring of high-altitude operation sites and timely alarms of abnormal situations, predict safety risks, visualize safety conditions through data, help managers adjust their strategies, and ensure the safety of high-altitude operation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068255A_ABST
    Figure CN119068255A_ABST
Patent Text Reader

Abstract

The system comprises an image acquisition module, a background database, a background server and a data analysis and optimization unit, wherein the image acquisition module is used for acquiring real-time images during high-altitude operation; the data analysis and optimization unit is used for collecting violation behavior identification data confirmed by the background server and judging the safety condition of a high-altitude operation site; the background database comprises an image database used for storing various violation images collected in the early stage and a data table database used for storing safety regulations matched and corresponding to the violation images; the background server comprises a storage module used for storing images acquired by the image acquisition module in real time, an image processing module used for processing the acquired images, and an image identification module used for identifying the images processed by the image processing module and various violation graphs stored in a background database. The invention discloses a visual high-altitude operation safety monitoring method for an extra-high voltage power grid, and the method comprises the steps: recognizing whether the operation process breaks rules or not through obtaining a real-time operation image, and judging the safety condition of high-altitude operation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electric power operation, and in particular to a visual high-altitude operation safety monitoring system and method for ultra-high voltage power grids. Background Art

[0002] With the rapid economic development and urbanization process in my country, the pace of UHV power grid construction has accelerated. Accident hazards and safety precautions focus on high-altitude operation scenes. During the operation, safety issues have become increasingly prominent, and accidents have occurred frequently, causing huge economic losses and life safety hazards to enterprises and individuals. Therefore, it is particularly important to research and develop a technical means that can effectively ensure the safety of high-altitude operations. Summary of the invention

[0003] In order to solve the technical problems existing in the above-mentioned technologies, it is necessary to provide a visual high-altitude operation safety monitoring system for ultra-high voltage power grids.

[0004] A visual high-altitude operation safety monitoring system for ultra-high voltage power grids, comprising an image acquisition module, a background database, a background server, and a data analysis and optimization unit; The image acquisition module is installed at the work site to obtain real-time images of personnel working at high altitudes; The backend database includes an image database and a data table library; the image database is used to store various violation images collected in the early stage, and the data table library is used to store the safety regulations corresponding to the violation images; The backend server includes a storage module, an image processing module, and an image recognition module; the storage module is used to store images collected in real time by the image collection module, the image processing module is used to process the collected images, and the image recognition module is used to identify the images processed by the image processing module with various types of violation graphics stored in the backend database; The data analysis and optimization unit is used to collect the violation behavior identification data confirmed by the background server, and judge the safety status of the aerial work site and whether the safety operation process is followed.

[0005] Preferably, it also includes an alarm unit, and the alarm unit establishes a wireless communication connection with the background server.

[0006] It is also necessary to provide a visual safety monitoring method for high-altitude operations in UHV power grids.

[0007] A method for visualizing high-altitude work safety monitoring for ultra-high voltage power grids, using the above-mentioned visualizing high-altitude work safety monitoring system for ultra-high voltage power grids, comprises the following steps: S1: Build an image database based on the images of illegal scenes of high-altitude operations collected in the early stage; S2: Based on the constructed image database, a data table library is established according to the definitions of safety regulations and various violations, photo materials and judgment criteria; S3: Acquire real-time images of personnel working at high altitudes at the work site; S4: feature extraction and classification of the preprocessed image; S5: Based on the extracted image features, the extracted image features are compared with the photos in the image database to determine whether a traffic violation has occurred; if so, proceed to the next step; if not, no action is taken; S6: Compare the violation photo image with the data table library to determine the type of violation behavior, and remind the violator through the alarm unit to correct the violation behavior; S7: Obtain relevant images and violation reports related to the violation, conduct statistical analysis according to the violation type, construct violation behavior charts, evaluate the safety status of the aerial work site, and adjust the management strategy in a timely manner.

[0008] Preferably, in step S4, the step of using the SSD algorithm to extract features and classify the preprocessed image includes the following steps: S41: Image Preprocessing Adjust the input image to the fixed size required by the network; normalize the image pixel values; S42: Feature extraction Use a deep convolutional neural network as the base network to extract deep features of the image; add auxiliary feature layers at different levels of the network to capture features of different scales; S43: Generate default box Generate a series of default boxes on each cell of the auxiliary feature map, with different aspect ratios and scales; S44: Calculate confidence and class probability For each default box, the network predicts two main outputs: one is the confidence (the probability of the object being present or not), and the other is the probability distribution of the object category; S45: Non-maximum suppression Sort the predicted bounding boxes and apply non-maximum suppression (NMS) to remove boxes with large overlaps to ensure that each object is detected only once; S46: Output the final test result According to the confidence and category probability, the final detection results are then filtered out through non-maximum suppression (NMS).

[0009] Preferably, in step S5, when the extracted image features are compared with the photos in the image database, the following method is adopted: S51: Using SIFT algorithm to process key points in the extracted image features; S52: Using SURF algorithm to process key points in the extracted image features; S53: using the ORB algorithm to process the key points in the extracted image features; S54: Use deep learning algorithms for optimization and improvement.

[0010] Preferably, when the key points in the extracted image features are processed using the SIFT algorithm, the following method is adopted: First, key points are detected in the image, and Gaussian difference function (DoG) is used to detect key points in different scale spaces; For each detected key point, the Taylor expansion of the area around the key point is fitted to remove the deviation of the edge response; Assign one or more directions to each key point, and determine the main direction by calculating the gradient direction histogram in the neighborhood of the key point; For each keypoint, the algorithm generates a 128-dimensional feature descriptor.

[0011] Preferably, when the SURF algorithm is used to process the key points in the extracted image features, the following method is adopted: Use the determinant of the Hessian matrix to determine the location and scale of key points in the image, and construct the Hessian matrix by applying box filters of different scales to the image; Accurately locate the position and scale of key points and remove unstable key points; Assign one or more directions to each key point, and determine the main direction by calculating the Haar wavelet response in the neighborhood of the key point; For each keypoint, a 64-dimensional feature descriptor is generated.

[0012] Preferably, when the ORBF algorithm is used to process the key points in the extracted image features, the following method is adopted: Perform appropriate preprocessing of the image; Detect candidate key points. For each pixel in the image, compare its brightness with that of the surrounding pixels. If there are N consecutive pixels that are brighter or darker than the center point, the point is marked as a candidate key point. Perform non-maximum suppression on the detected candidate key points to remove smaller candidate points near larger candidate points. Use machine learning classifiers to filter out true key points and remove falsely detected points. For each keypoint, calculate the centroid of the pixels in its neighborhood and use the direction between the centroid and the keypoint as the direction of the keypoint; A series of point pairs are randomly selected in the neighborhood of the key point, which are used to generate binary descriptors; for each point pair, their brightness values ​​are compared, if the brightness of the first point is greater than the second point, the result is 1, otherwise it is 0; the point pair is rotated according to the direction of the key point to ensure that the descriptor is rotationally invariant; The similarity between two ORB descriptors is compared using the Hamming distance, which calculates the number of different bits between two binary strings. The best matching points are selected based on the Hamming distance, and the ratio test is used to further remove false matches.

[0013] Preferably, when using deep learning algorithm for optimization and improvement, the following method is adopted: Data Collection Collect enough training data and possibly validation / test data to ensure data quality and diversity; Data preprocessing Data cleaning: remove noise and irrelevant data; Data enhancement: increase data diversity through rotation, scaling, cropping, etc.; Data standardization / normalization: adjust the data scale to make it suitable for model training; Data segmentation: divide the data into training set, validation set and test set; Model design Choose the appropriate network architecture, defining the number of layers, number of neurons, activation function, loss function, and optimizer of the model; Model Training Adjust the model structure and parameters based on the evaluation results, use regularization techniques to reduce overfitting, and perform hyperparameter tuning; Model Saving Save the trained model weights and architecture for subsequent use; Model maintenance Regularly retrain your model with new data to keep it accurate and relevant.

[0014] Compared with the prior art, the visual high-altitude operation safety monitoring system and method for ultra-high voltage power grid provided by the present invention has the following advantages: Conduct real-time analysis of monitoring images and promptly issue alarms with the alarm system when any abnormal situation is detected to ensure the safety of aerial work.

[0015] Combine historical data to predict the safety risks of working at heights and take preventive measures in advance.

[0016] (3) Realize data visualization and present the analysis results to management personnel in the form of charts, reports, etc., so as to facilitate understanding of the safety status of the aerial work site and timely adjust the management strategy; realize multi-level linkage. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0018] Figure 1 It is a structural schematic diagram of the present invention.

[0019] Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0021] Please see Figure 1 , In one embodiment, the present invention provides a visual high-altitude operation safety monitoring system for a UHV power grid, including an image acquisition module, a background database, a background server, and a data analysis and optimization unit; The image acquisition module is installed at the work site to obtain real-time images of personnel working at high altitudes; The backend database includes an image database and a data table library; the image database is used to store various violation images collected in the early stage, and the data table library is used to store the safety regulations corresponding to the violation images; The backend server includes a storage module, an image processing module, and an image recognition module; the storage module is used to store the images collected by the image acquisition module in real time, the image processing module is used to process the collected images, and the image recognition module is used to identify the images processed by the image processing module with various violation graphics stored in the backend database; The data analysis and optimization unit is used to collect the violation identification data confirmed by the background server, judge the safety status of the aerial work site, and whether the safe operation process is followed.

[0022] In order to be able to promptly remind on-site workers of violations, an alarm unit is added. The alarm unit establishes a wireless communication connection with the background server, which can issue warnings in time through the background server to stop or remind workers.

[0023] Please see Figure 2In one embodiment, the present invention provides a method for visualizing high-altitude work safety monitoring of a UHV power grid, and using the visualizing high-altitude work safety monitoring system of a UHV power grid, the method includes the following steps: S1: Build an image database based on the images of illegal scenes of high-altitude operations collected in the early stage; S2: Based on the constructed image database, a data table library is established according to the definitions of safety regulations and various violations, photo materials and judgment criteria; S3: Acquire real-time images of personnel working at high altitudes at the work site; S4: feature extraction and classification of the preprocessed image; S5: Based on the extracted image features, the extracted image features are compared with the photos in the image database to determine whether a traffic violation has occurred; if so, proceed to the next step; if not, no action is taken; S6: Compare the violation photo image with the data table library to determine the type of violation behavior, and remind the violator through the alarm unit to correct the violation behavior; S7: Obtain relevant images and violation reports related to the violation, conduct statistical analysis according to the violation type, construct violation behavior charts, evaluate the safety status of the aerial work site, and adjust the management strategy in a timely manner.

[0024] In one embodiment, the step of extracting features and classifying the preprocessed image using the SSD algorithm includes the following steps: Image preprocessing Adjust the input image to the fixed size required by the network; normalize the image pixel values; Feature extraction Use a deep convolutional neural network as the base network to extract deep features of the image; add auxiliary feature layers at different levels of the network to capture features of different scales; Generate Default Box Generate a series of default boxes on each cell of the auxiliary feature map, with different aspect ratios and scales; Calculate confidence and class probabilities For each default box, the network predicts two main outputs: one is the confidence (the probability of the object being present or not), and the other is the probability distribution of the object category; Non-maximum suppression Sort the predicted bounding boxes and apply non-maximum suppression (NMS) to remove boxes with large overlaps to ensure that each object is detected only once; Output the final test results According to the confidence and category probability, the final detection results are then filtered out through non-maximum suppression (NMS).

[0025] In one embodiment, the extracted image features are compared with photos in the image database in the following manner: The SIFT algorithm is used to process the key points in the extracted image features; The SURF algorithm is used to process the key points in the extracted image features; The ORB algorithm is used to process the key points in the extracted image features; Use deep learning algorithms for optimization and improvement.

[0026] Specifically, when the SIFT algorithm is used to process the key points in the extracted image features, the following method is used: First, key points are detected in the image. These key points are usually corners, edge points or other significant areas in the image. The Gaussian difference function (DoG) is used to detect key points in different scale spaces. DoG is obtained by subtracting continuous Gaussian blurred images and can detect features of different scales. For each detected key point, the Taylor expansion of the area around the key point is fitted to remove the deviation of the edge response; Assign one or more directions to each key point, which makes the feature invariant to image rotation; determine the main direction by calculating the gradient direction histogram in the neighborhood of the key point; For each key point, the algorithm generates a 128-dimensional feature descriptor; the descriptor is generated by considering the gradient direction and magnitude of the area around the key point, which makes the feature description invariant to lighting changes, image deformation, etc.

[0027] Specifically, when the SURF algorithm is used to process the key points in the extracted image features, the following method is used: Use the determinant of the Hessian matrix to determine the location and scale of key points in the image, and construct the Hessian matrix by applying box filters of different scales to the image; Accurately locate the position and scale of key points and remove unstable key points; Assign one or more directions to each key point so that the feature is invariant to image rotation; determine the main direction by calculating the Haar wavelet response in the neighborhood of the key point; For each keypoint, a 64-dimensional feature descriptor is generated.

[0028] Preferably, when the ORBF algorithm is used to process the key points in the extracted image features, the following method is adopted: Perform appropriate preprocessing on the image, such as grayscale conversion; Detect candidate key points. For each pixel in the image, compare its brightness with that of the surrounding pixels. If there are N consecutive pixels that are brighter or darker than the center point (for example, N=12), the point is marked as a candidate key point. Perform non-maximum suppression on the detected candidate key points to remove smaller candidate points near larger candidate points. Use machine learning classifiers (such as support vector machines) to filter out true key points and remove false positives. For each key point, calculate the centroid of the pixels in its neighborhood, and use the direction between the centroid and the key point as the direction of the key point. To increase robustness, multiple directions can be calculated and the most frequently occurring direction can be selected. A series of point pairs (e.g., 256 pairs) are randomly selected in the neighborhood of the key point, and these point pairs are used to generate binary descriptors; for each point pair, their brightness values ​​are compared, and if the brightness of the first point is greater than that of the second point, the result is 1, otherwise it is 0; these comparison results are combined into a binary string, namely the BRIEF descriptor; the point pairs are rotated according to the orientation of the key point to ensure that the descriptor is rotationally invariant; The similarity between two ORB descriptors is compared using the Hamming distance, which calculates the number of different bits between two binary strings. The best matching points are selected based on the Hamming distance, and a ratio test (such as NNDR, nearest neighbor distance ratio) is used to further remove false matches.

[0029] Specifically, when using deep learning algorithms for optimization and improvement, the following methods are used: Step 1: Problem Definition Determine the type of problem you want to solve (e.g., classification, regression, clustering, generation, etc.).

[0030] Clarify the target variable and prediction task.

[0031] Step 2: Data Collection Collect enough training data and possibly validation / test data.

[0032] Ensure the quality and diversity of data.

[0033] Step 3: Data Preprocessing Data cleaning: remove noise and irrelevant data.

[0034] Data enhancement: Increase data diversity through rotation, scaling, cropping, etc.

[0035] Data standardization / normalization: adjust the data scale to make it suitable for model training.

[0036] Data splitting: Divide the data into training set, validation set and test set.

[0037] Step 4: Model design Choose an appropriate network architecture (for example, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), etc.).

[0038] Define the model's number of layers, number of neurons, activation function, loss function, and optimizer.

[0039] Step 5: Model training Use the training data to train the model.

[0040] Tune hyperparameters (such as learning rate, batch size, number of iterations, etc.).

[0041] Use a validation set to monitor the training process and prevent overfitting.

[0042] Step 6: Model Evaluation Use the test set to evaluate the performance of the model.

[0043] Use appropriate evaluation metrics (such as precision, recall, F1 score, mean square error, etc.).

[0044] Step 7: Model Optimization Adjust the model structure and parameters based on the evaluation results.

[0045] Use regularization techniques (such as L1 / L2 regularization, dropout, etc.) to reduce overfitting.

[0046] Perform hyperparameter tuning.

[0047] Step 8: Save the model Save the trained model weights and architecture for subsequent use.

[0048] Step 9: Model deployment Integrate the model into real-world applications.

[0049] Provide model prediction services through API.

[0050] Monitor model performance and perform necessary maintenance and updates.

[0051] Step 10: Model maintenance Regularly retrain your model with new data to keep it accurate and relevant.

[0052] Update models to accommodate new business requirements or data patterns.

[0053] In addition, these steps can be adjusted according to specific project needs. Deep learning is an iterative process, and you may need to go back to the previous steps several times for adjustment and optimization.

[0054] What is disclosed above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A visual high-altitude operation safety monitoring system for ultra-high voltage power grid, characterized by: It includes image acquisition module, backend database, backend server, and data analysis and optimization unit; The image acquisition module is installed at the work site to obtain real-time images of personnel working at high altitudes; The backend database includes an image database and a data table library; the image database is used to store various violation images collected in the early stage, and the data table library is used to store the safety regulations corresponding to the violation images; The backend server includes a storage module, an image processing module, and an image recognition module; the storage module is used to store images collected in real time by the image collection module, the image processing module is used to process the collected images, and the image recognition module is used to identify the images processed by the image processing module with various types of violation graphics stored in the backend database; The data analysis and optimization unit is used to collect the violation behavior identification data confirmed by the background server, and judge the safety status of the aerial work site and whether the safety operation process is followed.

2. The visual high-altitude operation safety monitoring system for ultra-high voltage power grid according to claim 1 is characterized by: It also includes an alarm unit, which establishes a wireless communication connection with a background server.

3. A method for visualizing high-altitude work safety monitoring of a UHV power grid, using the visualizing high-altitude work safety monitoring system of a UHV power grid according to any one of claims 1 to 2, characterized in that: The following steps are included: S1: Build an image database based on the images of illegal scenes of high-altitude operations collected in the early stage; S2: Based on the constructed image database, a data table library is established according to the definitions of safety regulations and various violations, photo materials and judgment criteria; S3: Acquire real-time images of personnel working at high altitudes at the work site; S4: feature extraction and classification of the preprocessed image; S5: comparing the extracted image features with the photos in the image database to determine whether a traffic violation has occurred; If yes, proceed to the next step; if no, do nothing; S6: Compare the violation photo image with the data table library to determine the type of violation behavior, and remind the violator through the alarm unit to correct the violation behavior; S7: Obtain relevant images and violation reports related to the violation, conduct statistical analysis according to the violation type, construct violation behavior charts, evaluate the safety status of the aerial work site, and adjust the management strategy in a timely manner.

4. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 3 is characterized by: In step S4, the step of using the SSD algorithm to extract features and classify the preprocessed image includes the following steps: S41: Image Preprocessing Adjust the input image to the fixed size required by the network; normalize the image pixel values; S42: Feature extraction Use a deep convolutional neural network as the base network to extract deep features of the image; add auxiliary feature layers at different levels of the network to capture features of different scales; S43: Generate default box Generate a series of default boxes on each cell of the auxiliary feature map, with different aspect ratios and scales; S44: Calculate confidence and class probability For each default box, the network predicts two main outputs: one is the confidence (the probability of the object being present or not), and the other is the probability distribution of the object category; S45: Non-maximum suppression Sort the predicted bounding boxes and apply non-maximum suppression (NMS) to remove boxes with large overlaps to ensure that each object is detected only once; S46: Output the final test result According to the confidence and category probability, the final detection results are then filtered out through non-maximum suppression (NMS).

5. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 3 is characterized by: In step S5, the extracted image features are compared with the photos in the image database in the following manner: S51: Using SIFT algorithm to process key points in the extracted image features; S52: Using SURF algorithm to process key points in the extracted image features; S53: using the ORB algorithm to process the key points in the extracted image features; S54: Use deep learning algorithms for optimization and improvement.

6. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 5 is characterized by: When the SIFT algorithm is used to process the key points in the extracted image features, the following method is used: First, key points are detected in the image, and Gaussian difference function (DoG) is used to detect key points in different scale spaces; For each detected key point, the Taylor expansion of the area around the key point is fitted to remove the deviation of the edge response; Assign one or more directions to each key point, and determine the main direction by calculating the gradient direction histogram in the neighborhood of the key point; For each keypoint, the algorithm generates a 128-dimensional feature descriptor.

7. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 5 is characterized by: When the SURF algorithm is used to process the key points in the extracted image features, the following method is used: Use the determinant of the Hessian matrix to determine the location and scale of key points in the image, and construct the Hessian matrix by applying box filters of different scales to the image; Accurately locate the position and scale of key points and remove unstable key points; Assign one or more directions to each key point, and determine the main direction by calculating the Haar wavelet response in the neighborhood of the key point; For each keypoint, a 64-dimensional feature descriptor is generated.

8. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 5 is characterized by: When the ORBF algorithm is used to process the key points in the extracted image features, the following method is used: Perform appropriate preprocessing of the image; Detect candidate key points. For each pixel in the image, compare its brightness with that of the surrounding pixels. If there are N consecutive pixels that are brighter or darker than the center point, the point is marked as a candidate key point. Perform non-maximum suppression on the detected candidate key points to remove smaller candidate points near larger candidate points. Use machine learning classifiers to filter out real key points and remove false positives; For each keypoint, calculate the centroid of the pixels in its neighborhood and use the direction between the centroid and the keypoint as the direction of the keypoint; A series of point pairs are randomly selected in the neighborhood of the key point, which are used to generate binary descriptors; for each point pair, their brightness values ​​are compared, if the brightness of the first point is greater than the second point, the result is 1, otherwise it is 0; the point pair is rotated according to the direction of the key point to ensure that the descriptor is rotationally invariant; The similarity between two ORB descriptors is compared using the Hamming distance, which calculates the number of different bits between two binary strings. The best matching points are selected based on the Hamming distance, and the ratio test is used to further remove false matches.

9. The visual high-altitude operation safety monitoring method for UHV power grid according to claim 5 is characterized by: When using deep learning algorithms for optimization and improvement, the following methods are used: Data Collection Collect enough training data and possibly validation / test data to ensure data quality and diversity; Data preprocessing Data cleaning: remove noise and irrelevant data; Data enhancement: increase data diversity through rotation, scaling, cropping, etc.; Data standardization / normalization: adjust the data scale to make it suitable for model training; Data segmentation: divide the data into training set, validation set and test set; Model design Choose the appropriate network architecture, including the number of layers, number of neurons, activation function, loss function, and optimizer of the model; Model Training Adjust the model structure and parameters based on the evaluation results, use regularization techniques to reduce overfitting, and perform hyperparameter tuning; Model Saving Save the trained model weights and architecture for subsequent use; Model maintenance Regularly retrain your model with new data to keep it accurate and relevant.

Citation Information

Cited By

  • Wrong taking and placing early warning method and system suitable for unmanned aerial vehicle accessory storehouse

    CN119294976A