Power supply remote monitoring method and system based on image recognition

Abstract

The invention relates to the technical field of power supply remote monitoring, in particular to a power supply remote monitoring method and system based on image recognition, and the method comprises the steps: obtaining original image data of power equipment, carrying out the preprocessing, extracting a multi-scale feature map, and achieving the dynamic target detection and semantic segmentation through a multi-branch convolutional network. And generating an equipment state distribution diagram, constructing a health assessment model in combination with the historical operation data, and outputting an equipment health index to generate an abnormal early warning signal. The intelligent degree, the real-time performance and the adaptability in a complex scene can be improved, the defects in the aspects of image processing, environment adaptability and equipment depth monitoring in the prior art are overcome, and reliable guarantee is provided for safe operation of power supply equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power supply remote monitoring, and specifically relates to a power supply remote monitoring method and system based on image recognition. Background Art

[0002] With the rapid development of informatization and digitalization, the power supply remote monitoring method and system based on image recognition has become an important tool for power equipment operation and maintenance and energy management. It realizes real-time monitoring of the operating status of power equipment through image acquisition and analysis, which is of great significance in improving operation and maintenance efficiency and safety. However, the existing technical solutions still show certain limitations in terms of intelligence, real-time performance, and adaptability to complex scenarios, which affects the overall performance of the system.

[0003] After searching, the patent with publication number CN105811578B proposed a transmission line monitoring platform and its power monitoring algorithm and image warning algorithm, and the publication date was August 31, 2018. This technology uses a power module to power the monitoring module and the signal transmission module, and analyzes the operating status and potential threats of the transmission line through an image warning algorithm. However, this solution mainly relies on traditional image processing algorithms, and has limited ability to recognize dynamic changes of multiple targets in complex scenes, resulting in its accuracy and response speed in diversified environments to be improved. In addition, the system's monitoring of power supply equipment only stays at the basic status feedback level, and fails to achieve deep intelligent management and predictive analysis of potential faults.

[0004] Another patent with publication number CN105049786B involves an infrared thermal imager test auxiliary system and method, and the publication date is June 8, 2018. This technology uses a remote monitoring system to program the power supply and electrical information of multiple infrared thermal imagers in real time, and transmits the collected images to the remote end. However, the image acquisition and transmission process of this solution is highly dependent on the stability of hardware equipment and the quality of transmission channels. In complex electromagnetic environments or poor network conditions, there may be risks of data loss or delay. At the same time, the system does not fully combine advanced image recognition technology to conduct in-depth analysis of collected data, and it is difficult to meet the needs of modern remote monitoring for intelligent decision-making support.

[0005] The above problems show that the existing image recognition-based power supply remote monitoring technology still has room for improvement in intelligent image processing, adaptability to complex environments, and in-depth monitoring and fault prediction of power supply equipment. Therefore, the present invention aims to improve the real-time, accuracy and reliability of the system by introducing advanced image recognition algorithms and intelligent power supply management strategies, so as to better meet the actual needs of remote monitoring of power equipment. Summary of the invention

[0006] The present invention provides a power supply remote monitoring method and system based on image recognition, and its main purpose is to solve the deficiencies existing in the existing power supply remote monitoring technology in terms of intelligence level, real-time performance, and adaptability to complex scenarios.

[0007] To achieve the above object, a power supply remote monitoring method based on image recognition provided by the present invention includes: acquiring original image data in the operating state of power equipment, preprocessing the original image data to obtain normalized image data; performing multi-scale feature extraction on the normalized image data to obtain multi-scale feature maps; performing dynamic object detection on the multi-scale feature maps through a multi-branch convolutional network to obtain object detection results; performing semantic segmentation on the object detection results to generate a device state distribution map; constructing a device health assessment model by combining the device state distribution map and historical operation data, and outputting a device health index; using the device health assessment model to perform anomaly analysis on the current device operating state to generate an anomaly warning signal.

[0008] Optionally, preprocessing the original image data to obtain normalized image data includes: performing grayscale processing on the original image data to obtain a grayscale image; performing histogram equalization processing on the grayscale image to enhance the image contrast; performing normalization processing on the enhanced image to adjust the pixel value range to a preset interval to obtain normalized image data.

[0009] Optionally, performing multi-scale feature extraction on the normalized image data to obtain multi-scale feature maps includes: designing a set of convolutional kernels with different receptive fields, and performing multi-scale convolutional operations on the normalized image data; performing non-linear activation processing on the results of the convolutional operations to obtain preliminary feature maps; performing pooling operations on the preliminary feature maps to reduce the data dimension and retain key information to obtain multi-scale feature maps.

[0010] Optionally, performing dynamic object detection on the multi-scale feature maps through a multi-branch convolutional network to obtain object detection results includes: inputting the multi-scale feature maps into the multi-branch convolutional network, where each branch corresponds to a specific object category; gradually extracting high-dimensional features of the object through the convolutional layers of each branch; using a region proposal network to generate candidate boxes from the high-dimensional features, and determining the object category and position information within the candidate boxes through a classifier to obtain object detection results.

[0011] Optionally, performing semantic segmentation on the object detection results to generate a device state distribution map includes: classifying and marking each pixel point in the object detection results to distinguish different device components and their states; generating a mask map of the device components according to the classification marks; superimposing the mask map on the original image to construct a device state distribution map.

[0012] Optionally, before constructing a device health assessment model by combining the device status distribution map and historical operation data and outputting the device health index, the method further includes: collecting the historical operation data of the device, including parameters such as temperature, current, and voltage; performing time series analysis on the historical operation data to extract the periodic change rule; combining the periodic change rule with the spatial characteristics in the device status distribution map to construct an initial health assessment model; and optimizing the initial health assessment model by introducing an adaptive weight mechanism to obtain the device health assessment model.

[0013] Optionally, use the device health assessment model to perform anomaly analysis on the current device operation status and generate an anomaly warning signal, including: inputting the current device operation status into the device health assessment model to calculate the device health index; judging whether there is an anomaly according to the comparison result between the device health index and the preset threshold; if there is an anomaly, generate an anomaly warning signal and mark the specific location and type of the anomaly.

[0014] To solve the above problems, the present invention also provides a power supply remote monitoring system based on image recognition. The system includes: an image acquisition module for acquiring the original image data of the power equipment in the operation state; an image preprocessing module for preprocessing the original image data to obtain normalized image data; a feature extraction module for performing multi-scale feature extraction on the normalized image data to obtain a multi-scale feature map; a target detection module for performing dynamic target detection on the multi-scale feature map through a multi-branch convolutional network to obtain a target detection result; a semantic segmentation module for performing semantic segmentation on the target detection result to generate a device status distribution map; a health assessment module for constructing a device health assessment model by combining the device status distribution map and historical operation data and outputting the device health index; and an anomaly analysis module for using the device health assessment model to perform anomaly analysis on the current device operation status and generate an anomaly warning signal.

[0015] In the embodiments of the present invention, by acquiring the original image data of the power equipment in the operating state, preprocessing the original image data to obtain the normalized image data, a high-quality data basis is provided for subsequent feature extraction; by performing multi-scale feature extraction on the normalized image data to obtain multi-scale feature maps, the recognition ability of multi-object dynamic changes in complex scenes is improved; by using a multi-branch convolutional network to perform dynamic object detection on the multi-scale feature maps to obtain the object detection results, the real-time performance and accuracy of the system are enhanced; by performing semantic segmentation on the object detection results to generate the equipment status distribution map, a refined description of the equipment components and their statuses is realized; by combining the equipment status distribution map and historical operation data to construct an equipment health assessment model and output the equipment health index, support is provided for the in-depth intelligent management of the equipment; by using the equipment health assessment model to perform anomaly analysis on the current operating state of the equipment to generate an anomaly warning signal, the reliability and safety of the system are improved. Therefore, the power supply remote monitoring method and system based on image recognition proposed by the present invention solve the deficiencies of the prior art in aspects such as intelligent image processing, complex environment adaptability, and in-depth monitoring and fault prediction of power supply equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 FIG. is a flowchart of a power supply remote monitoring method based on image recognition provided by an embodiment of the present invention, showing the overall process from the acquisition of original image data to the generation of an anomaly warning signal.

[0017] Figure 2 FIG. is a structural block diagram of a power supply remote monitoring system based on image recognition provided by an embodiment of the present invention, including an image acquisition module, an image preprocessing module, a feature extraction module, an object detection module, a semantic segmentation module, a health assessment module, and an anomaly analysis module.

[0018] Figure 3 FIG. is a schematic diagram of the generation of the equipment status distribution map in the embodiments of the present invention, showing the process from the object detection results to the marking of the equipment component status and the overlay of the mask map.

[0019] The reference numerals are as follows: 1. Image acquisition module; 2. Image preprocessing module; 3. Feature extraction module; 4. Object detection module; 5. Semantic segmentation module; 6. Health assessment module; 7. Anomaly analysis module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention provides a power supply remote monitoring method and system based on image recognition, and its specific implementation process is described in detail in conjunction with FIGS. Figure 1 to FIGS. Figure 3 The operation process of the entire system is as shown in Figure 1As shown, it includes the overall steps from obtaining the original image data to generating the anomaly warning signal. The system structure block diagram is as Figure 2 shown, which contains multiple modules and cooperates with each other in a specific order to complete tasks. The generation process of the device status distribution map is as Figure 3 shown, which demonstrates the specific implementation from the target detection results to the device component status marking and mask map superposition.

[0021] First, in the Figure 2 system structure block diagram shown, the image acquisition module 1 is the basic component of the entire system, which is used to obtain the original image data of the power equipment under operation. The image acquisition module 1 usually consists of high-resolution industrial cameras, which are installed at key positions of the power equipment, such as substations, distribution cabinets and other scenarios. To ensure the quality of image acquisition, the cameras need to have low-light adaptability and anti-interference ability, and at the same time transmit the acquired image data to the subsequent module through the network interface. The image acquisition module 1 is connected to the image preprocessing module 2 through a data transmission link, and this link can be a wired or wireless communication method to ensure real-time performance and stability.

[0022] The image preprocessing module 2 normalizes the original image data from the image acquisition module 1 to obtain high-quality normalized image data. Specifically, the image preprocessing module 2 first performs grayscale processing on the original image data, converting the color image into a grayscale image to reduce the subsequent calculation amount. Then, histogram equalization processing is performed on the grayscale image to enhance the image contrast, making the details in the image more clearly visible. Finally, normalization processing is performed on the enhanced image to adjust the pixel value range to a preset interval between 0 and 1, so as to provide a unified data format for subsequent feature extraction. The output of the image preprocessing module 2 is directly transmitted to the feature extraction module 3, and efficient data exchange is achieved between the two through the memory sharing mechanism.

[0023] The feature extraction module 3 is responsible for performing multi-scale feature extraction on the normalized image data to improve the recognition ability of multi-target dynamic changes in complex scenarios. As Figure 1 shown, the feature extraction module 3 designs a set of convolutional kernels with different receptive fields, and extracts the initial feature map through multi-scale convolutional operations on the input image. The design of the convolutional kernel set follows the principle from small to large, covering local details and global information in turn, ensuring that different-scale target features can be captured. Subsequently, non-linear activation processing is performed on the result of the convolutional operation, and the ReLU function is used to enhance the feature expression ability. To further reduce the data dimension and retain key information, the feature extraction module 3 uses max pooling operation to downsample the initial feature map, and finally obtains the multi-scale feature map. The output of the feature extraction module 3 is transmitted to the target detection module 4 through the data pipeline.

[0024] The core of the target detection module 4 lies in the design of a multi-branch convolutional network, whose function is to perform dynamic target detection on the multi-scale feature maps output by the feature extraction module 3. As Figure 2 shown, the target detection module 4 contains multiple branches internally, and each branch corresponds to a specific target category, such as transformers, circuit breakers, or cable joints in power equipment. Each branch gradually extracts high-dimensional features of the target through convolutional layers and uses the region proposal network to generate candidate boxes. The generation process of the candidate boxes combines the sliding window technique and the anchor mechanism, enabling the rapid localization of potential target regions in the image. Subsequently, the classifier determines the target category and location information within the candidate boxes and finally outputs the target detection results. The output of the target detection module 4 is transmitted to the semantic segmentation module 5 through the data flow channel.

[0025] The role of the semantic segmentation module 5 is to refine the target detection results and generate a device status distribution map. As Figure 3 shown, the semantic segmentation module 5 first classifies and labels each pixel point in the target detection results to distinguish different device components and their states. For example, for the transformer component, specific regions such as its casing, cooling device, and terminal block can be marked. Then, a mask map of the device components is generated based on the classification labels, and each pixel value in the mask map represents the category number of the corresponding component. Finally, the mask map is superimposed on the original image to construct the device status distribution map. This process realizes the visual description of the device components and their states, providing intuitive spatial feature support for subsequent health assessment. The output of the semantic segmentation module 5 is transmitted to the health assessment module 6 through the data interface.

[0026] The health assessment module 6 constructs an initial health assessment model by combining the device status distribution map and historical operation data and obtains the final device health assessment model through optimization. Specifically, the health assessment module 6 first collects the historical operation data of the device, including parameters such as temperature, current, and voltage. These data are collected in real time by sensors and stored in the database for subsequent analysis. Then, time series analysis is performed on the historical operation data to extract periodic change patterns, such as the fluctuation patterns or trends of certain parameters during the device operation. Combining these periodic change patterns with the spatial features in the device status distribution map, an initial health assessment model is constructed. To improve the accuracy of the model, the health assessment module 6 introduces an adaptive weight mechanism to optimize the initial model and finally obtains the device health assessment model. The optimization process is realized through an iterative training algorithm to ensure that the model can accurately reflect the actual operation state of the device. The output of the health assessment module 6 is transmitted to the anomaly analysis module 7 through the data link.

[0027] The anomaly analysis module 7 uses the device health assessment model to perform anomaly analysis on the current operating state of the device and generate anomaly warning signals. Specifically, the anomaly analysis module 7 inputs the current operating state of the device into the device health assessment model to calculate the device health index. The device health index is a comprehensive indicator that reflects the overall current operating condition of the device. Whether there is an anomaly is judged based on the comparison result between the device health index and the preset threshold. If the device health index is lower than the threshold, it is considered that the device has an anomaly. The anomaly analysis module 7 generates an anomaly warning signal and marks the specific location and type of the anomaly. For example, if the temperature of the wiring terminal of a certain transformer is too high, the anomaly analysis module 7 will mark this location and prompt the type of overheat fault. The anomaly warning signal is fed back to the operator in real time through the alarm system or the user interface so that measures can be taken in a timely manner.

[0028] During the operation of the entire system, each module achieves efficient cooperation through the coordinated work of hardware and software. For example, the data transmission between the image acquisition module 1 and the image preprocessing module 2 depends on the high-speed network interface, while the data exchange between the feature extraction module 3 and the target detection module 4 is completed through the memory sharing mechanism. In addition, the cooperation between the health assessment module 6 and the anomaly analysis module 7 depends on the database management system to ensure the seamless integration of historical data and real-time data. This modular design not only improves the scalability of the system but also enhances the robustness of the system, enabling it to operate stably in complex environments.

[0029] Through the above implementation manners, it can be seen that the power supply remote monitoring method and system based on image recognition proposed by the present invention can effectively solve the deficiencies of the prior art in terms of intelligence level, real-time performance, and adaptability to complex scenarios. Every step from image acquisition to the generation of anomaly warning signals is carefully designed to ensure the efficiency and reliability of the system.

[0030] In order to better enable relevant personnel in the technical field to fully understand and implement the present invention, the following supplements the specific implementation principle of the present invention in combination with a specific application scenario.

[0031] In the actual operation environment of a certain substation, the demand for remote monitoring of power equipment is particularly prominent. In this scenario, key equipment such as transformers, circuit breakers, and cable joints are distributed in multiple areas, and some equipment is in an environment with low light or strong electromagnetic interference. To ensure that the operating status of these devices can be monitored in real time and potential faults can be detected in a timely manner, the method and system proposed in the present invention are applied to this scenario. The following are the specific implementation steps and their principle supplements: First, the image acquisition module 1 obtains the original image data through high-resolution industrial cameras installed in the substation. These cameras are deployed at key positions near the transformer shell, circuit breaker operating mechanism, and cable joints to cover the main components of the equipment. To cope with the low-light environment, the cameras are equipped with infrared fill lights and are designed with anti-electromagnetic interference to ensure the stability of image acquisition. The acquired image data is transmitted to the image preprocessing module 2 through a wired network interface, and this process relies on a high-speed communication link to ensure the real-time nature of data transmission.

[0032] Subsequently, the image preprocessing module 2 performs normalization processing on the received original image data. Specifically, grayscale processing converts the color image into a grayscale image, thereby reducing the complexity of subsequent calculations; histogram equalization significantly enhances the contrast of the image, making the fine cracks on the transformer shell or the local overheating phenomenon of the cable joint more clearly visible; normalization processing further adjusts the pixel values to between 0 and 1, providing a unified data format for subsequent feature extraction. This series of operations effectively improves the image quality and lays a foundation for the accurate analysis of subsequent modules.

[0033] Next, the feature extraction module 3 performs multi-scale feature extraction on the normalized image data. The design of the convolutional kernel group in module 3 follows the principle from small to large, covering local details and global information in turn. For example, when detecting the transformer shell, the convolutional kernel with a small receptive field can capture fine features such as cracks, while the convolutional kernel with a large receptive field focuses on the changes in the overall structure. By performing ReLU activation processing on the convolutional results, module 3 enhances the feature expression ability; the max pooling operation further reduces the data dimension while retaining the key information. The finally generated multi-scale feature map can comprehensively reflect the dynamic changes of the equipment components.

[0034] The object detection module 4 receives the multi-scale feature map from the feature extraction module 3 and performs dynamic object detection through a multi-branch convolutional network. Each branch focuses on a specific object category, such as a transformer, a circuit breaker, or a cable joint. When detecting the circuit breaker, the region proposal network combines the sliding window technique and the anchor mechanism to quickly locate the operating mechanism area and generate candidate boxes. The classifier then determines the object category and location information within the candidate boxes and outputs accurate object detection results. This process ensures that the system can quickly identify multi-object devices in a complex scenario.

[0035] The semantic segmentation module 5 performs refined processing on the object detection results to generate a device status distribution map. Taking a transformer as an example, module 5 first classifies and labels each pixel point in the object detection results to distinguish specific areas such as the outer shell, cooling device, and terminal. Then, a mask map is generated according to the classification labels, where each pixel value corresponds to the category number of the component. Finally, the mask map is superimposed on the original image to construct a device status distribution map. This process realizes the visual description of the device components and their states, providing intuitive spatial feature support for subsequent health assessment.

[0036] The health assessment module 6 constructs an initial health assessment model by combining the device status distribution map and historical operation data. For example, when analyzing the operating status of a transformer, module 6 collects its historical temperature, current, and voltage data, and extracts the periodic fluctuation law through time series analysis. These laws are combined with the spatial features in the device status distribution map to construct an initial health assessment model. To further improve the accuracy, module 6 introduces an adaptive weight mechanism to optimize the model, and finally obtains a device health assessment model. This model can accurately reflect the actual operating status of the transformer and provide a reliable basis for anomaly analysis.

[0037] The anomaly analysis module 7 uses the device health assessment model to perform anomaly analysis on the current device operating status. For example, when the temperature of a terminal of a transformer exceeds the normal range, module 7 calculates its device health index and compares it with a preset threshold. If the health index is lower than the threshold, an anomaly warning signal is generated, and the specific location and type of the anomaly are marked. This signal is fed back to the operation and maintenance personnel in real time through the alarm system so that measures can be taken in a timely manner. For example, for the problem of overheating of the terminal, the operation and maintenance personnel can quickly arrange operations such as cooling or replacing components.

[0038] During the operation of the entire system, the modules cooperate efficiently through the coordinated work of hardware and software. For example, the data transmission between the image acquisition module 1 and the image preprocessing module 2 depends on a high-speed network interface, while the data exchange between the feature extraction module 3 and the object detection module 4 is completed through a memory sharing mechanism. In addition, the cooperation between the health assessment module 6 and the anomaly analysis module 7 depends on the database management system to ensure the seamless integration of historical data and real-time data. This modular design not only improves the scalability of the system but also enhances its robustness in complex environments.

[0039] It can be seen from the implementation steps of the above specific application scenarios that the power supply remote monitoring method and system based on image recognition proposed by the present invention can effectively solve the deficiencies of the prior art in terms of intelligence level, real-time performance, and adaptability to complex scenarios. Every step from image acquisition to the generation of anomaly warning signals is carefully designed to ensure the efficiency and reliability of the system.

Claims

1. A remote power supply monitoring method based on image recognition, characterized in that, The method includes the following steps: obtaining the original image data of the power equipment in the operating state; preprocessing the original image data to obtain normalized image data; performing multi-scale feature extraction on the normalized image data to obtain multi-scale feature maps; performing dynamic object detection on the multi-scale feature maps through a multi-branch convolutional network to obtain object detection results; performing semantic segmentation on the object detection results to generate a device status distribution map; constructing a device health assessment model by combining the device status distribution map and historical operation data, and outputting a device health index; using the device health assessment model to perform anomaly analysis on the current operating state of the device to generate an anomaly warning signal.

2. The power supply remote monitoring method based on image recognition according to claim 1, wherein The preprocessing of the original image data to obtain normalized image data includes the following steps: performing grayscale processing on the original image data to obtain a grayscale image; performing histogram equalization processing on the grayscale image to enhance the image contrast; performing normalization processing on the enhanced image to adjust the pixel value range to a preset interval to obtain normalized image data.

3. The power supply remote monitoring method based on image recognition according to claim 1, characterized in that The multi-scale feature extraction of the normalized image data to obtain multi-scale feature maps includes the following steps: designing a set of convolutional kernels with different receptive fields and performing multi-scale convolutional operations on the normalized image data; performing non-linear activation processing on the results of the convolutional operations to obtain preliminary feature maps; performing pooling operations on the preliminary feature maps to reduce the data dimension and retain key information to obtain multi-scale feature maps.

4. The power supply remote monitoring method based on image recognition according to claim 1, characterized in that, The dynamic object detection of the multi-scale feature maps through a multi-branch convolutional network to obtain object detection results includes the following steps: inputting the multi-scale feature maps into the multi-branch convolutional network, where each branch corresponds to a specific object category; gradually extracting high-dimensional features of the object through the convolutional layers of each branch; using a region proposal network to generate candidate boxes from the high-dimensional features, and determining the object category and location information within the candidate boxes through a classifier to obtain object detection results.

5. The power supply remote monitoring method based on image recognition according to claim 1, wherein The semantic segmentation of the object detection results to generate a device status distribution map includes the following steps: classifying and labeling each pixel point in the object detection results to distinguish different device components and their states; generating a mask map of the device components according to the classification labels; overlaying the mask map with the original image to construct a device status distribution map.

6. The power supply remote monitoring method based on image recognition according to claim 1, wherein, Before constructing the device health assessment model by combining the device status distribution map and historical operation data, the following steps are also included: collecting the historical operation data of the device, including temperature, current, and voltage parameters; performing time series analysis on the historical operation data to extract periodic change rules; combining the periodic change rules with the spatial features in the device status distribution map to construct an initial health assessment model; optimizing the initial health assessment model by introducing an adaptive weight mechanism to obtain the device health assessment model.

7. The power supply remote monitoring method based on image recognition according to claim 1, wherein The abnormal analysis of the current device operation status using the device health assessment model to generate an abnormal warning signal includes the following steps: input the current device operation status into the device health assessment model to calculate the device health index; determine whether there is an abnormality according to the comparison result between the device health index and the preset threshold; if there is an abnormality, generate an abnormal warning signal and mark the specific location and type of the abnormality.

8. A remote power supply monitoring system based on image recognition, characterized in that, The system includes: an image acquisition module (1) for acquiring original image data under the operation status of the power device; an image preprocessing module (2) for preprocessing the original image data to obtain normalized image data; a feature extraction module (3) for performing multi-scale feature extraction on the normalized image data to obtain a multi-scale feature map; a target detection module (4) for performing dynamic target detection on the multi-scale feature map through a multi-branch convolutional network to obtain a target detection result; a semantic segmentation module (5) for performing semantic segmentation on the target detection result to generate a device status distribution map; a health assessment module (6) for constructing a device health assessment model by combining the device status distribution map and historical operation data and outputting a device health index; an abnormal analysis module (7) for performing abnormal analysis on the current device operation status using the device health assessment model to generate an abnormal warning signal.

9. The power supply remote monitoring system based on image recognition according to claim 8, wherein The image acquisition module (1) consists of a high-resolution industrial camera, and the industrial camera is installed at key positions of the power device and transmits image data through a network interface.

10. The power supply remote monitoring system based on image recognition according to claim 8, wherein, The health assessment module (6) optimizes the initial health assessment model through an adaptive weight mechanism and improves the accuracy of the device health assessment model using an iterative training algorithm.

Images