Lightweight signal lamp fault detection method and system in complex scene

Abstract

The invention relates to the technical field of intelligent traffic, and discloses a lightweight signal lamp fault detection method and system in a complex scene, and the detection method comprises the following steps: 1, configuring a detection task; step 2, data preprocessing; step 3, signal lamp panel detection; 4, classifying lamp holders of the signal lamps; step 5, fault judgment; and step 6, alarming and recording. The signal lamp fault detection method achieves the beneficial effect of improving the signal lamp fault detection effect in a complex scene in the whole day period through automatic marking of the ROI region, daytime and night detection distinguishing, data set sample optimization under an extreme weather condition and an environmental factor judgment method under a poor illumination condition.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a lightweight signal light fault detection method and system in complex scenarios. Background Art

[0002] As an important system of urban road traffic, the traffic signal control system plays a vital role in traffic control. Traffic lights are prone to malfunction due to environmental factors, and traffic lights are far away from traffic supervision centers. Operation and maintenance personnel cannot promptly learn about the working conditions of traffic lights at intersections, and traffic light failures may cause traffic congestion or traffic accidents, posing a huge safety hazard.

[0003] At present, traffic signal fault detection mainly relies on electronic circuit detection equipment combined with manual observation to find faults, which has high maintenance costs and complexity. The detection method using deep learning is also used to a certain extent. The existing technical problems mainly exist in the following aspects: Existing technologies focus more on method research, and there is less optimization and improvement of detection effects in complex scenarios based on actual applications. In real scenes, due to the influence of complex background, extreme weather, illumination at different times, and environmental factors such as occlusion, the detection effect in some areas or time periods is poor. Existing methods have limitations in fault detection effects in complex situations such as the above; The existing methods are relatively rough in the classification of signal lights, mainly the classification of signal light colors and the classification of common national standard lamp heads, and the type coverage is not comprehensive enough. With the continuous improvement of traffic intelligence, the number of non-national standard signal lights that adapt to the characteristics of urban traffic is also increasing, becoming a major factor affecting urban traffic. Therefore, non-national standard signal lights: multi-lamp head signal lights (number of lamp heads>3), straight left turn, countdown and other light states that are not within the scope of national standard signal lights also need to be included in the scope of detection and management; Existing traffic light fault detection systems usually only focus on reporting fault information and lack an effective closed-loop mechanism for continuous improvement and optimization of detection results. Summary of the invention

[0004] Technical issues solved: In view of the shortcomings of the prior art, the present invention provides a lightweight signal light fault detection method and system, which can automatically adapt to complex environments, accurately locate the lamp panel, identify the lamp head status, and determine multiple types of faults. The lamp panel detection is realized by using the YOLO algorithm, and the structure-optimized convolutional neural network RepVGG-A is used to classify the lamp head image, thereby enhancing the system's adaptability under complex lighting conditions. The improved convolutional network adds hole convolution in the low layer to expand the receptive field, which is suitable for low-resolution and blurred scenes, ensuring the classification accuracy.

[0005] Technical Solution

[0006] To achieve the above purpose, the present invention provides a lightweight signal light fault detection method and system suitable for complex lighting and environment, aiming to solve the problem of insufficient detection accuracy of the prior art under conditions of lighting changes, environmental complexity and signal light diversity. The method and system of the present invention can automatically adapt to different lighting conditions, and through a specific combination of detection and classification models, efficient lamp panel detection, lamp head status classification and fault determination are achieved, achieving accurate monitoring and alarm functions of signal lights in complex scenes.

[0007] The present invention provides the following technical solution: a lightweight signal light fault detection method in a complex scenario, characterized in that the detection method comprises the following steps: Step 1: Configure the detection task: obtain real-time images through the camera, generate a two-dimensional coordinate system based on the image, and set the ROI area of ​​the monitoring reference frame; Step 2: Data preprocessing: cropping, normalizing, and enhancing the ROI area to generate an image containing the traffic light panel; Step 3: Traffic light panel detection: Use the YOLO algorithm combined with the daytime and nighttime separation detection models, and automatically switch models when lighting conditions change to ensure accurate positioning of the light panel under different lighting conditions; Step 4: Traffic light lamp head classification: divide the lamp panel into equal parts according to the number of lamp heads, and classify each lamp head image through the structure-optimized convolutional neural network RepVGG-A to obtain the lamp head's on and off, color, shape and other status information. The convolutional neural network adds a hole convolution branch in the low layer to expand the receptive field and improve the classification accuracy of low-light and blurred images; Step 5: Fault determination: Combined with the working status of the lamp holder, a variety of combined fault logics are used to determine yellow flash, lamp panel off, lamp holder offset and occlusion faults, and fault verification is performed based on continuous frame detection to improve the accuracy of fault identification; Step 6: Alarm and record: If the fault type meets the preset conditions, a fault record is generated and an alarm is issued.

[0008] Preferably, the YOLO algorithm used in the signal light panel detection step includes a daytime YOLO model and a nighttime YOLO model based on switching of lighting conditions, which are used to adapt to target detection in different lighting environments.

[0009] Preferably, the convolutional neural network is a RepVGG-A model, and a hole convolution branch is added to the low-level network structure to improve classification accuracy in scenes such as low resolution and blurred lighting.

[0010] Preferably, the classification models of the signal lamp caps include national standard and non-national standard classification models for daytime and nighttime, so as to adapt to the type and style changes of signal lamps with different standards.

[0011] Preferably, the fault types include: combined faults such as simultaneous lighting, extinguishing, offset and occlusion faults. The fault judgment logic is based on multi-frame detection and lamp head status combination analysis. The lamp head status combination is comprehensively analyzed through the detection results of multiple consecutive frames, thereby improving the accuracy of fault judgment and reducing false alarms.

[0012] A signal light fault detection system for executing the method according to any one of claims 1 to 5, characterized in that the system comprises: Fault detection platform, used to configure detection tasks and process image data; Fault detection engine, including data preprocessing, detection model, and fault judgment module; The fault information recording module is used to record the fault location, time, type and related video data.

[0013] Preferably, the fault detection engine determines whether the light intensity meets the preset threshold condition by calculating the grayscale mean of the ROI area, and switches to the daytime or nighttime monitoring mode according to the lighting conditions.

[0014] Preferably, the fault information recording module further includes continuous video data of a set time before and after fault detection to facilitate fault analysis and review.

[0015] Preferably, the system automatically sends out an alarm signal when an obstruction or unrecognizable target is detected, prompting manual intervention.

[0016] Preferably, the fault detection platform supports automatic analysis and false alarm optimization functions, and continuously updates the detection model based on historical data.

[0017] Preferably, the system also includes a signal light fault detection algorithm warehouse, which is used to store and manage daytime light panel detection models, nighttime light panel detection models, daytime national standard signal light classification models, nighttime national standard signal light classification models, daytime non-national standard signal light classification models and nighttime non-national standard signal light classification models to adapt to the detection requirements of different lighting conditions and signal light types.

[0018] Technical effects: 1. Adaptability to complex lighting environments The present invention introduces an automatic switching mechanism between daytime and nighttime detection models in signal light detection, and automatically selects the most suitable model in combination with lighting conditions to ensure accurate signal light detection in different lighting environments such as daytime, nighttime, strong light or low light. This separation model design greatly improves the stability and adaptability of detection, allowing the system to maintain high accuracy in extreme lighting environments.

[0019] 2. Efficient data preprocessing and ROI area annotation The ROI automatic annotation function adopted by the present invention supports the generation of a stable monitoring reference frame around the traffic light. Even if the camera moves slightly or the position of the traffic light is slightly offset, the system can still lock the monitoring area. In addition, the cropping, normalization and enhancement processing in the data preprocessing step further improves the image input quality, provides a standardized and stable input source for subsequent lamp panel detection and lamp head classification, and ensures the reliability of the detection results.

[0020] 3. Improved classification accuracy brought by improved convolutional neural network structure The present invention adds a dilated convolution branch to the lower layer of the RepVGG-A convolutional neural network to expand the receptive field and improve the feature extraction capability in complex environments such as low resolution and blurred lighting. Compared with the traditional convolutional network structure, the optimization of the present invention can effectively improve the classification accuracy in low-light and blurred images, ensure the accurate recognition of the on / off state, color, shape and other features of the lamp head, and enhance the fault tolerance of the system.

[0021] like Figure 3 As shown in the figure, the application of dilated convolution: The top convolution block in the figure (marked as "3x3 rate=2 stride=1") indicates the application of dilated convolution, that is, the convolution kernel is 3x3, the dilation rate is set to 2, and the stride is set to 1. This setting will cause the convolution kernel to skip a certain number of pixels on the feature map, thereby expanding the receptive field. Multi-branch structure: Both 3x3 convolution and 1x1 convolution kernels are used in each convolution layer. This multi-branch structure design can enhance the expressiveness of features. Through the combination of convolution kernels of different sizes, the network can extract features of different scales, thereby improving the accuracy of classification. Feature fusion (Residual Connection): After each convolution layer, a feature fusion operation (Residual Connection) is used, that is, the output of the previous layer and the output of the current layer are added to retain more low-level feature information. This residual connection helps the transfer of gradients, avoids gradient disappearance, and improves the stability of training. ReLU activation function: After each convolution operation, the ReLU activation function is used to make the network.

[0022] like Figure 4 As shown in the figure, the re-parameterization process of the improved RepVGG network structure of the present invention is as follows. Specifically, re-parameterization is a technique to simplify a complex network structure in the inference stage, thereby improving the inference efficiency while retaining the high expressive power in the training stage. The following is an explanation of each stage in the figure: Multi-branch structure in the training phase (left part): During the training phase, the RepVGG network adopts a multi-branch structure, including: Dilated convolution branch (3x3 convolution, rate=2, stride=1): This branch uses dilated convolution to expand the receptive field.

[0023] Standard convolution branch (3x3 convolution, stride=2 and stride=1): Different standard convolution layers extract features with different strides.

[0024] 1x1 convolution branch: used to extract local information and supplement the information of smaller receptive field.

[0025] This multi-branch structure can capture multi-scale and multi-level feature information during the training phase, improving the network's expressive power and feature extraction effect.

[0026] Simplified structure after reparameterization (right part): During the inference phase, these multi-branch structures are merged into a single 3x3 convolutional layer. Through reparameterization, the weights and biases of all branches can be integrated into an equivalent convolutional layer. This simplified structure can significantly improve computational efficiency during inference and reduce computation and memory requirements.

[0027] Effect of reparameterization: This method combines the advantages of the multi-branch structure in the training phase and the single-branch structure in the inference phase. The reparameterized network can maintain the performance during training while achieving efficient real-time computing in the inference phase.

[0028] 4.Combination decision logic for multiple fault types Based on the detection results of multiple consecutive frames, the present invention realizes accurate identification of various complex faults through combined fault logic. The system can identify various combined fault types such as simultaneous lighting, extinguishing, offset and occlusion, and reduce false alarms through multi-frame combined judgment of lamp holder status. The combined fault judgment logic improves the accuracy and robustness of the system, and is particularly suitable for complex signal lamp configurations such as multiple lamp holders and non-national standard signal lamps.

[0029] 5. Efficiency of closed-loop optimization mechanism The present invention has the functions of fault recording and automatic false alarm analysis, and stores the video clips before and after fault detection in the fault information recording module for subsequent analysis. At the same time, the system can optimize the model for false alarms and balance the data set samples based on the detection history to achieve self-optimization and continuous improvement of the model, thereby maintaining efficient and stable detection effects in practical applications.

[0030] 6. Lightweight design enables real-time and efficient detection The design of the present invention takes into account the lightweight requirements of the system, and adopts lightweight models such as YOLO and RepVGG-A optimization network to ensure that it can still run efficiently in edge computing devices or resource-constrained environments. Compared with systems that rely on cloud computing, the present invention achieves high detection accuracy while greatly improving real-time performance, making the system suitable for wide application in actual traffic monitoring scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of the architecture of an embodiment of a signal light fault detection system of the present invention; Figure 2 A flowchart of an embodiment of a signal light fault detection system of the present invention; Figure 3 This is a schematic diagram of the improved REVGG training network structure of the present invention; Figure 4 Schematic diagram of the improved REVGG network structure reparameterization process of the present invention. DETAILED DESCRIPTION

[0032] 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.

[0033] Embodiment 1: The overall architecture of the signal light fault detection system of this embodiment is divided into three main modules: signal light fault detection platform, signal light fault detection engine, and signal light fault detection algorithm warehouse. The functions of each module are as follows: Traffic light fault detection platform: Task configuration: responsible for configuring the detection task, including setting the camera's shooting parameters, day and night mode switching time, etc.

[0034] Fault warning: Generates alarm information after detecting a signal light failure and sends the alarm to the monitoring personnel.

[0035] Analysis report: Generate reports based on detected fault data to analyze fault types, frequency of occurrence, etc., to help monitoring personnel understand the operating status of traffic lights.

[0036] Traffic light fault detection engine: Result processing and analysis: Process and analyze the test results to generate the final fault judgment results.

[0037] Environmental factor determination: Detect the lighting conditions in the current environment, such as strong light, low light, etc., so as to adjust the detection strategy according to the environmental conditions.

[0038] Fault logic judgment: Based on the combination of detection results and lamp status, determine whether there is a fault type (such as red and green at the same time, offset, obstruction, etc.).

[0039] Video access: access real-time video stream from the camera for use by the detection system.

[0040] Data preprocessing: preprocess the received image data, including cropping, normalization, and enhancement.

[0041] Model reasoning: Call the detection and classification models in the algorithm warehouse to identify the status of the traffic light panel and lamp head.

[0042] Traffic light fault detection algorithm warehouse: Day and night light panel detection model: used for light panel detection under day and night conditions to ensure accurate positioning of light panels under different lighting conditions.

[0043] Daytime international standard classification model: used to classify the status of daytime signal lamp heads that meet international standards.

[0044] Daytime non-international standard classification model: used to classify the status of daytime signal lamp heads that do not meet international standards.

[0045] Nighttime international standard classification model: used to classify the status of nighttime signal light heads that meet international standards.

[0046] Nighttime non-international standard classification model: used to classify the status of nighttime signal light heads that do not meet international standards.

[0047] The operation process of this embodiment is as follows: 1. Configure the detection task The system first configures the detection task, sets the camera's shooting angle, focal length and other parameters, delineates the monitoring area of ​​the traffic light, and generates the ROI (Region of Interest) of the area for subsequent detection. During the configuration process, users can set the time for switching between day and night modes and the fault alarm threshold.

[0048] 2. Data Preprocessing The system preprocesses the real-time images captured by the camera, including image cropping, normalization and enhancement operations, to ensure high-quality input images, thereby improving the stability of detection.

[0049] 3. Select the detection mode according to the time period The system detects the current time and determines whether it is a daytime or nighttime time period. If it is daytime, the system calls the daytime detection mode; if it is nighttime, the system calls the nighttime detection mode. In different detection modes, the system will use different detection models to adapt to changes in lighting conditions.

[0050] 4. Traffic light panel inspection Depending on the selected daytime or nighttime mode, the system calls the corresponding YOLO detection model to locate the traffic light panel in the image. The system identifies the position of the traffic light panel in the ROI area based on the preprocessed image information and marks the coordinate information of the panel.

[0051] 5. Determination of signal light types and classification of lamp holders After the system detects the lamp panel, it selects different classification models to classify the lamp head status. The specific operations are as follows: Daytime national standard signal lights: Call the daytime national standard signal light classification model to classify the on and off status, color and shape of each lamp head.

[0052] Daytime non-national standard signal lights: call the daytime non-national standard signal light classification model to identify the on and off status and color of the lamp head.

[0053] National standard traffic lights at night: Call the national standard traffic lights classification model at night to classify the lamp status to meet the detection needs in low-light environments at night.

[0054] Nighttime non-national standard traffic lights: Call the classification model for nighttime non-national standard traffic lights to classify the status of the lamp heads and ensure detection accuracy in complex environments at night.

[0055] 6. Reasoning result processing The system processes the status information of each lamp head according to the inference results of the lamp head classification model. The lamp head status information includes features such as on and off status, color and shape, which will serve as the basis for subsequent fault judgment.

[0056] 7. Environmental testing The system determines whether there is a strong light or low light environment based on the lighting conditions of the current monitoring environment. If strong light (such as direct sunlight) is detected, which may cause detection interference, the system will adaptively adjust the detection results to reduce false detections under abnormal lighting conditions.

[0057] 8. Fault type determination Based on the combination of the inference results and the lamp status, the system determines whether the signal light is faulty, including the following fault types: Lighting up simultaneously: including red, yellow and green, red and green, red and yellow, and yellow and green. When multiple consecutive frames are detected to be lit up simultaneously, it is judged as a fault; Deviation: If the number of signal light panels that deviate from the preset monitoring area in the picture exceeds the threshold, it is judged as a fault; Blockage: In a single lamp panel, if the number of lamp heads identified as blocked exceeds the threshold, it is judged as a fault; Off: When all the lamp heads in the lamp panel are detected to be off for multiple consecutive frames, it is judged as a fault.

[0058] The system confirms the fault through multi-frame continuous detection to ensure the accuracy of fault judgment.

[0059] 9. Generate alarm information After determining that there is a fault, the system generates fault alarm information and stores the fault location, time, type and video clips in the fault information recording module. The alarm information can be sent to the monitoring platform in real time to remind the monitoring personnel to deal with the fault in time.

[0060] 10. Analysis report generation The system generates analysis reports based on the stored fault data, including the frequency of occurrence of fault types, fault distribution, etc., to facilitate long-term monitoring and optimization of the status of traffic lights. The analysis reports can be used to improve the detection model, improve the accuracy of fault detection and the stability of the system.

[0061] Embodiment 2: The difference between this embodiment and embodiment 1 is that the YOLO algorithm used in the signal light panel detection step includes a daytime YOLO model and a nighttime YOLO model based on the switching of lighting conditions, which are used to adapt to target detection in different lighting environments. The convolutional neural network is a RepVGG-A model, and a hole convolution branch is added to the low-level network structure to improve classification accuracy in low-resolution, blurred lighting and other scenes. The classification model of the signal light head includes national standard and non-national standard classification models for daytime and nighttime to adapt to the type and style changes of signal lights of different standards.

[0062] In this embodiment, a hole convolution branch is added to the low-level network structure of RepVGG to expand the receptive field. At the same time, adding parallel branches can enhance the representation ability of the model, thereby improving the feature extraction and classification capabilities of large targets in classified images.

[0063] The improved training network structure is as follows Figure 3 As shown in the figure, the blue part is the dilated convolution branch with ratedilation rate=2 and stride=1 added to the downsampling blocks of the low-level stage0 and stage1, and the non-downsampling block does not add branches. The receptive field of the feature map of the initial stage of the two low-level stages is expanded to improve the internal and overall representation capabilities of the stage.

[0064] Figure 4 The structural reparameterization process of the downsampling blocks in stage 0 and stage 1 is described. In order to retain the characteristics of the relationship between the pixels of the input feature map and not increase the computational complexity, the newly added branches are not fused. Since the branches are only added in the downsampling blocks in the initial stage, the inference model obtained after structural reparameterization is still a lightweight model. The improved network improves the confusion of some classification results caused by low resolution of deployment points in some areas of the actual scene, blurred images, and blurred outlines of lamp caps due to lighting effects to a certain extent, and improves the classification accuracy.

[0065] Embodiment 3: The difference between this embodiment and embodiment 2 is that the fault types include combined faults such as simultaneous lighting, extinguishing, offset and occlusion faults, and the fault judgment logic is based on multi-frame detection and lamp head status combination analysis. The lamp head status combination is comprehensively analyzed through the detection results of multiple consecutive frames, thereby improving the accuracy of fault judgment and reducing false alarms.

[0066] The fault detection engine determines whether the light intensity meets the preset threshold condition by calculating the grayscale mean of the ROI area, and switches to daytime or nighttime monitoring mode according to the lighting conditions.

[0067] Wherein, the fault information recording module further includes continuous video data of a set time before and after the fault detection to facilitate fault analysis and review.

[0068] The system automatically sends out an alarm signal when detecting an obstructed or unrecognizable target, prompting human intervention.

[0069] Among them, the fault detection platform supports automatic analysis and false alarm optimization functions, and continuously updates the detection model based on historical data.

[0070] Among them, the system also includes a signal light fault detection algorithm warehouse, which is used to store and manage daytime light panel detection models, nighttime light panel detection models, daytime national standard signal light classification models, nighttime national standard signal light classification models, daytime non-national standard signal light classification models and nighttime non-national standard signal light classification models to adapt to the detection requirements of different lighting conditions and signal light types.

[0071] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the invention, the scope of the invention being defined by the appended claims and their equivalents.

Claims

1. A lightweight signal light fault detection method in complex scenarios, characterized in that: The detection method comprises the following steps: Step 1: Configure the detection task: obtain real-time images through the camera, generate a two-dimensional coordinate system based on the image, and set the ROI area of ​​the monitoring reference frame; Step 2: Data preprocessing: cropping, normalizing, and enhancing the ROI area to generate an image containing the traffic light panel; Step 3: Traffic light panel detection: Use the YOLO algorithm combined with the daytime and nighttime separation detection models, and automatically switch models when lighting conditions change to ensure accurate positioning of the light panel under different lighting conditions; Step 4: Traffic light lamp head classification: divide the lamp panel into equal parts according to the number of lamp heads, and classify each lamp head image through the structure-optimized convolutional neural network RepVGG-A to obtain the lamp head's on and off, color, shape and other status information. The convolutional neural network adds a hole convolution branch in the low layer to expand the receptive field and improve the classification accuracy of low-light and blurred images; Step 5: Fault determination: Combined with the working status of the lamp holder, a variety of combined fault logics are used to determine yellow flash, lamp panel off, lamp holder offset and occlusion faults, and fault verification is performed based on continuous frame detection to improve the accuracy of fault identification; Step 6: Alarm and record: If the fault type meets the preset conditions, a fault record is generated and an alarm is issued.

2. The lightweight signal light fault detection method in complex scenarios according to claim 1, wherein: The YOLO algorithm used in the signal light panel detection step includes a daytime YOLO model and a nighttime YOLO model based on lighting condition switching, which are used to adapt to target detection in different lighting environments.

3. The lightweight signal light fault detection method in complex scenarios according to claim 1, wherein: The convolutional neural network is a RepVGG-A model, and a hole convolution branch is added to the low-level network structure to improve the classification accuracy in scenes such as low resolution and blurred lighting.

4. The lightweight signal light fault detection method in complex scenarios according to claim 1, wherein: The classification models of the signal lamp caps include national standard and non-national standard classification models for daytime and nighttime use, so as to adapt to the type and style changes of signal lamps with different standards.

5. According to the lightweight signal light fault detection method in complex scenarios as claimed in claim 1, the fault types include: For combined faults such as simultaneous lighting, extinguishing, offset and occlusion faults, the fault judgment logic is based on multi-frame detection and lamp head status combination analysis. The lamp head status combination is comprehensively analyzed through the detection results of multiple consecutive frames, thereby improving the accuracy of fault judgment and reducing false alarms.

6. A signal light fault detection system for executing the method according to any one of claims 1 to 5, characterized in that: The system comprises: Fault detection platform, used to configure detection tasks and process image data; Fault detection engine, including data preprocessing, detection model, and fault judgment module; The fault information recording module is used to record the fault location, time, type and related video data.

7. The signal light failure detection system according to claim 6, wherein: The fault detection engine determines whether the light intensity meets the preset threshold condition by calculating the grayscale mean of the ROI area, and switches to daytime or nighttime monitoring mode according to the lighting conditions.

8. The signal light failure detection system according to claim 6, wherein: The fault information recording module further includes continuous video data of a set time before and after fault detection to facilitate fault analysis and review.

9. The signal light failure detection system according to claim 6, wherein: The system also includes a signal light fault detection algorithm warehouse, which is used to store and manage daytime light panel detection models, nighttime light panel detection models, daytime national standard signal light classification models, nighttime national standard signal light classification models, daytime non-national standard signal light classification models and nighttime non-national standard signal light classification models to meet the detection requirements of different lighting conditions and signal light types.

10. The signal light failure detection system according to claim 6, wherein: The system automatically sends out an alarm signal when an obstruction or unrecognizable target is detected, prompting manual intervention. The fault detection platform supports automatic analysis and false alarm optimization functions, and continuously updates the detection model based on historical data.

Images