Vehicle charging area fire-fighting early warning system integrating AI identification and intelligent inspection

Abstract

The invention discloses a vehicle charging area fire-fighting early warning system fusing AI identification and intelligent inspection. The system comprises a front-end image acquisition module; the AI identification model is used for performing bimodal input, processing visible light images and thermal imaging data at the same time, and embedding a channel-space attention module based on an improved lightweight network; the alarm information distribution module is provided with a multi-level alarm strategy, and the multi-level alarm strategy comprises first-level alarm and second-level alarm; an intelligent inspection management module; and the alarm manual processing module is used for synchronizing all operation records to a Web platform log. The vehicle charging area fire-fighting early-warning system has the beneficial effects that a comprehensive, intelligent and efficient vehicle charging area fire-fighting early-warning solution is provided by fusing AI identification and intelligent inspection technologies, the problems of low identification rate, insufficient safety, backward inspection management and the like in the prior art are effectively solved, and the vehicle charging area fire-fighting early-warning system is suitable for popularization and application. And a powerful guarantee is provided for fire safety of places such as electric vehicle charging stations and the like.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and specifically to a vehicle charging area fire warning system that integrates AI recognition and intelligent inspection. Background Art

[0002] Similar AI fire alarm recognition products and equipment are not widely used in many industries. Most of them rely on simple cameras for monitoring and smoke alarms for warnings, rather than integrated applications. Even similar products suffer from low recognition rates and false positives due to factors such as the use of single model training parameters. Therefore, they also have some potential shortcomings:

[0003] 1. Low security: Relying on camera monitoring and smoke alarm reminders cannot respond to emergencies in a timely manner, which will cause greater threats to life and property safety.

[0004] 2. Low recognition rate: There are many interference factors in the identification of the vehicle charging area, and there are few trainable parameters, which often result in low recognition accuracy and insufficient reliability.

[0005] 3. High environmental requirements: It is sensitive to environmental factors such as light and lighting, and additional protection measures may be required to ensure the accuracy of the equipment.

[0006] 4. Issues with traditional inspection management: Relying on paper records, it is impossible to verify whether inspectors arrive at designated locations on time; historical data is difficult to trace, and there is a lack of correlation analysis between abnormal events and inspection records.

[0007] 5. Alarm processing defects: Existing systems mostly use one-way alarm push and lack manual confirmation and operation record-keeping mechanisms; there is no closed-loop record after the false alarm is closed, and the processing process cannot be traced. Summary of the Invention

[0008] The purpose of the present invention is to provide a vehicle charging area fire warning system that integrates AI recognition and intelligent inspection to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solutions: a vehicle charging area fire warning system integrating AI recognition and intelligent inspection, comprising:

[0010] Front-end image acquisition module: This includes several explosion-proof high-definition cameras with infrared night vision capabilities, deployed around the charging piles.

[0011] AI recognition model: Dual-modal input, processing visible light imagery and thermal imaging data simultaneously, embedding a channel-spatial attention module based on an improved lightweight network, and dynamically adjusting alarm trigger thresholds;

[0012] Alarm information distribution module: multi-level alarm strategy, including level one alarm and level two alarm, alarm information is distributed in JSON data packet format;

[0013] Intelligent inspection management module: Create inspection plans through the web platform, and the app verifies the location of inspectors through GPS geofencing and Bluetooth beacons to generate structured inspection reports;

[0014] Alarm manual processing module: After receiving the alarm information, the on-duty personnel must select the processing method within the specified time, and all operation records are synchronized to the Web platform log.

[0015] Preferably, the AI recognition model is based on the MobileNetV3 backbone network and embedded with a self-developed channel-spatial attention module (CSAM) to enhance sensitivity to small-scale smoke.

[0016] Preferably, the alarm information distribution module includes the following functions:

[0017] Level 1 alarm: triggers the on-site sound and light alarm;

[0018] Level 2 alarm: Push information to the on-duty personnel’s mobile app, send email to the management platform, and activate the fire sprinkler system.

[0019] Preferably, the intelligent inspection management module includes the following functions:

[0020] Create inspection plans on the Web platform;

[0021] The app uses GPS geofencing and Bluetooth beacon verification to ensure that inspectors reach the specified location.

[0022] Generate structured inspection reports and support exporting records by time, area, and device type.

[0023] Preferably, the alarm manual processing module includes the following functions:

[0024] After receiving the alarm information, the on-duty personnel must choose a handling method within 3 minutes;

[0025] All operation records (including operator, time, and on-site screenshots) are synchronized to the Web platform log.

[0026] Preferably, it also includes a sensor data auxiliary judgment module, which combines the smoke sensor and temperature sensor data and uses algorithm matching confidence to assist AI recognition.

[0027] Preferably, safety protection measures are also included, including App-side device offline reminder, front-end image acquisition program online detection, and sensor device online detection.

[0028] In the second aspect, a vehicle charging area fire warning method integrating AI recognition and intelligent inspection includes the following steps:

[0029] Step 1: Collect image data of the area surrounding the charging pile through the front-end image acquisition module;

[0030] Step 2: Input the collected image data into the AI recognition model for dual-modal input processing to identify potential fire risks;

[0031] Step 3: Based on the recognition results, the alarm information distribution module triggers the on-site sound and light alarm or pushes the alarm information to the on-duty personnel;

[0032] Step 4: Create an inspection plan through the intelligent inspection management module and verify that the inspection personnel arrive at the specified location;

[0033] Step 5: Record the alarm processing process through the alarm manual processing module and synchronize it to the Web platform log.

[0034] Preferably, in step 2, the AI recognition model is based on the MobileNetV3 backbone network, embedded with a self-developed channel-spatial attention module (CSAM), and dynamically adjusts the alarm trigger threshold.

[0035] Preferably, in step 3, the alarm information is distributed in JSON data packet format, including timestamp, camera ID, GPS coordinates, confidence level, and screenshot URL.

[0036] Compared with the existing technology, the beneficial effects of the present invention are: by integrating AI recognition and intelligent inspection technology, the present invention provides a comprehensive, intelligent and efficient fire warning solution for vehicle charging areas, effectively solving the problems of low recognition rate, insufficient security, backward inspection management and other problems existing in the existing technology, and providing strong protection for fire safety in places such as electric vehicle charging stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the system structure of an embodiment of the present invention;

[0038] Figure 2 Schematic diagram of the structure of the AI recognition model 200 according to an embodiment of the present invention;

[0039] Figure 3 is a flow chart of a method according to an embodiment of the present invention;

[0040] Figure 4 This is a flow chart of fire detection and alarm processing by the system of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0042] See also Figure 1 、 Figure 2 、 Figure 4 The present invention provides a technical solution: a vehicle charging area fire warning system that integrates AI recognition and intelligent inspection, including a front-end image acquisition module 100, an AI recognition model 200, an alarm information distribution module 300, an intelligent inspection management module 400 and an alarm manual processing module 500.

[0043] In an embodiment of the present invention, the front-end image acquisition module 100 includes several explosion-proof high-definition cameras that support infrared night vision function. The cameras are deployed in the area surrounding the charging pile, and the image acquisition frequency can be dynamically adjusted.

[0044] In this embodiment, a high-definition camera such as Hikvision DS-2CD3XXX has an image acquisition frequency of 5 frames per second by default, which is increased to 30 frames per second when an abnormality is detected.

[0045] In an embodiment of the present invention, the AI recognition model 200 supports dual-modal input, processes visible light images and thermal imaging data simultaneously, and dynamically adjusts the alarm trigger threshold based on an improved lightweight network embedding channel-spatial attention module.

[0046] In this embodiment, the AI recognition model 200 is based on the MobileNetV3 backbone network and is embedded with a self-developed channel-spatial attention module (CSAM) to enhance sensitivity to small-scale smoke.

[0047] In a specific embodiment of this embodiment, the AI recognition model 200 receives the front-end image acquisition module 100 specifically including:

[0048] Data preprocessing module 210. This module includes preprocessing of visible light image and thermal imaging data, wherein the processing process of visible light image data is as follows:

[0049] Normalize the pixel values of visible light images to the range of [0, 1] to reduce the impact of lighting changes on recognition.

[0050] Apply Gaussian filtering or median filtering to remove random noise in the image and improve image quality.

[0051] Resize the image to a fixed size required by the model input (e.g., 224×224 pixels) to ensure the consistency of the input data.

[0052] The process of preprocessing thermal imaging data is as follows:

[0053] The temperature values in thermal imaging data are mapped to a fixed range (such as [0,1]) to enhance the comparability under different environmental conditions.

[0054] Abnormal temperature area enhancement: Enhance areas with abnormally high temperatures to highlight potential fire risk areas.

[0055] Dual-modal data fusion module 220. The processing steps of the dual-modal data fusion processing module 220 are:

[0056] The pre-processed visible light image and thermal imaging data are input into the dual-modal fusion module. This module integrates the data of the two modalities during the feature extraction stage through feature-level fusion methods to enhance the model's adaptability to complex environments. It includes:

[0057] Feature extraction is performed on visible light images and thermal imaging data respectively to obtain their respective feature maps.

[0058] The feature maps of the two modalities are integrated into a comprehensive feature map through feature splicing or feature weighted fusion.

[0059] Feature extraction and enhancement module 230. Feature extraction and enhancement is based on an improved lightweight network (MobileNetV3). MobileNetV3 uses an inverted residual structure and depthwise separable convolution to significantly reduce computational complexity while maintaining high accuracy, making it suitable for resource-constrained edge computing scenarios. It includes:

[0060] Backbone network: The MobileNetV3 backbone network contains multiple depth-wise separable convolutional layers to gradually extract high-level features of the image.

[0061] Channel-Spatial Attention Module (CSAM): Embeds a self-developed channel-spatial attention module in key layers to enhance the model's sensitivity to key features such as small-scale smoke.

[0062] In this embodiment, the channel-spatial attention module (CSAM) specifically includes:

[0063] Channel attention: Global information of channel features is obtained through global average pooling and global maximum pooling, and then the weight of each channel is adjusted through a multi-layer perceptron (MLP) to highlight important feature channels.

[0064] Spatial attention: Based on channel attention, it further obtains the global information of spatial features through convolution operations, adjusts the weight of each spatial position, and enhances the features of key areas.

[0065] Classification and decision module 240. Includes:

[0066] Classifier: The extracted comprehensive features are input into the classifier, which outputs the probability distribution of different categories (such as normal, suspected fire, confirmed fire) based on the softmax function.

[0067] Decision-making mechanism: Make the final recognition decision based on the probability distribution of the classifier output and the dynamically adjusted threshold.

[0068] Dynamic threshold adjustment module 250, the module's work includes:

[0069] The alarm trigger threshold is dynamically adjusted based on factors such as ambient light intensity and the working status of the charging pile (such as charging power) to ensure high recognition rate and low false alarm rate under different conditions.

[0070] The camera's built-in light sensor monitors the ambient light intensity in real time.

[0071] Obtain current charging power and other status information through the communication interface with the charging pile.

[0072] Dynamically modify alarm trigger thresholds based on the monitored environment and device status through pre-trained adjustment strategies.

[0073] In an embodiment of the present invention, the alarm information distribution module 300 includes a multi-level alarm strategy, specifically including a first-level alarm and a second-level alarm, and the alarm information is distributed in a JSON data packet format.

[0074] In this embodiment, after receiving information from the AI recognition model 200, the alarm information distribution module 300 outputs the recognition results in a JSON data packet format, including information such as timestamp, camera ID, GPS coordinates, confidence level, and screenshot URL. This triggers an on-site audible and visual alarm or sends an alarm message to the on-duty personnel. The recognition results and related data are also recorded for continuous model optimization.

[0075] Among them, timestamp: the specific time when the identification occurs;

[0076] Camera ID: The camera ID that generates the recognition result;

[0077] GPS coordinates: the camera's geographic location information;

[0078] Confidence: The AI model’s confidence in the recognition result (0-100%);

[0079] Screenshot URL: the storage address of the on-site screenshot related to the recognition result;

[0080] Identify categories: such as "Normal", "Suspected Fire", and "Confirmed Fire".

[0081] In this embodiment, the alarm information distribution module 300 includes the following functions:

[0082] Level 1 alarm: triggers the on-site sound and light alarm; triggered when the identification category is "suspected fire" and the confidence level exceeds the preset threshold (such as 60%).

[0083] Level 2 Alarm: Push notifications to the on-duty personnel's mobile app, emails to the management platform, and triggers the fire sprinkler system. This is triggered when the identification category is "Confirmed Fire" and the confidence level exceeds a higher threshold (e.g., 80%).

[0084] In an embodiment of the present invention, the intelligent inspection management module 400 creates an inspection plan through a Web platform, and the App verifies the location of the inspection personnel through GPS geo-fences and Bluetooth beacons to generate a structured inspection report.

[0085] In this embodiment, the intelligent inspection module 400 is capable of creating inspection tasks, executing inspection tasks, recording and exporting inspection data, correlation analysis and feedback, alarm and inspection linkage, and automatic generation of inspection tasks.

[0086] In a specific embodiment of this embodiment, the creation of inspection tasks includes task planning and task creation. Determine the key areas and frequency of inspections based on the scale of the charging station, equipment distribution, and historical alarm data. Create an inspection plan through the Web platform, set the inspection time (such as 10:00 and 16:00 every day), inspection area (such as charging piles in Area A), and inspection content (such as equipment status inspection, environmental safety assessment). The administrator enters detailed information of the inspection plan on the Web platform, including inspection personnel, inspection routes, inspection points, etc. Assign the inspection task to the designated inspection personnel, and push the task notification through the App.

[0087] In a specific embodiment of this embodiment, the execution of the inspection task includes: the inspection personnel receive the inspection task notification through the mobile phone App and view the task details. The inspection personnel confirm the receipt of the task on the App side and prepare to execute it. When the inspection personnel arrive at the designated inspection point, the App side verifies whether their location is within the preset geographic fence through GPS positioning. Bluetooth beacons are deployed at key inspection points. When the inspection personnel arrive, the App side further verifies the accuracy of the location through Bluetooth signals. The inspection personnel check the equipment status, environmental safety, etc. according to the task requirements. The inspection results are recorded through the App side, including equipment operating status, environmental conditions, abnormal conditions, etc. Take pictures of key equipment and abnormal conditions, and the pictures are automatically uploaded to the system.

[0088] In a specific embodiment of this embodiment, inspection data recording and exporting includes: the system automatically generates a structured inspection report containing information such as inspection time, location, inspection personnel, equipment status, on-site photos, and abnormality markers. Inspection data is synchronized to the web platform in real time, ensuring that management personnel can view it promptly. Export function: The web platform supports exporting inspection records by time, region, equipment type, and other conditions, and export formats include Excel and PDF. All inspection data is stored in the system database, supporting historical data query and traceability.

[0089] In a specific embodiment of this embodiment, the correlation analysis and feedback includes: the system analyzes inspection data to identify abnormal conditions (such as equipment failures and environmental hazards); analyzes trends in inspection data to predict potential problems and provide a basis for preventive maintenance; generates inspection statistics reports, analyzes data such as inspection frequency and abnormal conditions; and proposes optimization suggestions based on the analysis results, such as adjusting inspection routes and increasing inspection frequency.

[0090] In a specific embodiment of this embodiment, the alarm and inspection linkage includes: when frequent alarms occur in a certain area (e.g., three or more times within 24 hours), the system automatically triggers an inspection task. If a serious abnormality is discovered during the inspection, the inspector can manually trigger the alarm through the app. The system automatically generates an inspection task, prompting the inspector to focus on checking the equipment status in the alarmed area. The inspection task is then pushed to the designated personnel through the app to ensure timely processing.

[0091] In a specific embodiment of this embodiment, automatic inspection task generation includes: the system analyzes historical inspection data and alarm data to identify high-risk areas and equipment. Based on machine learning algorithms, it predicts potential risk areas and generates preventive inspection tasks. The system automatically generates inspection tasks and pushes them to relevant personnel via the web platform and app. The execution of inspection tasks is monitored in real time to ensure timely completion.

[0092] Through the above working process, the intelligent inspection management module can manage inspection tasks efficiently and intelligently, ensure the comprehensiveness and timeliness of inspection work, and improve the fire safety management level of places such as charging stations.

[0093] In this embodiment, the alarm manual processing module 500: after receiving the alarm information, the on-duty personnel need to select a processing method within a specified time, and all operation records are synchronized to the Web platform log.

[0094] The alarm manual processing module includes the following functions:

[0095] After receiving the alarm information, the on-duty personnel must choose a handling method within 3 minutes;

[0096] All operation records (including operator, time, and on-site screenshots) are synchronized to the Web platform log.

[0097] In an optional embodiment of the present invention, the system further includes a sensor data auxiliary judgment module, which combines the smoke sensor and temperature sensor data and uses algorithm matching confidence to assist AI recognition.

[0098] In an optional embodiment of the present invention, the system also includes security protection measures, including App-end device offline reminder, front-end image acquisition program online detection, and sensor device online detection.

[0099] Also, see Figure 3 The embodiment of the present invention further provides a vehicle charging area fire warning method that integrates AI recognition and intelligent inspection, which is implemented in combination with the above embodiment and includes the following steps:

[0100] Step 1: Collect image data of the area surrounding the charging pile through the front-end image acquisition module;

[0101] Step 2: Input the collected image data into the AI recognition model for dual-modal input processing to identify potential fire risks;

[0102] Step 3: Based on the recognition results, the alarm information distribution module triggers the on-site sound and light alarm or pushes the alarm information to the on-duty personnel;

[0103] Step 4: Create an inspection plan through the intelligent inspection management module and verify that the inspection personnel arrive at the specified location;

[0104] Step 5: Record the alarm processing process through the alarm manual processing module and synchronize it to the Web platform log.

[0105] In this example, in step 2, the AI recognition model is based on the MobileNetV3 backbone network and embeds the self-developed channel-spatial attention module (CSAM) to dynamically adjust the alarm trigger threshold.

[0106] In this example, in step 3, the alarm information is distributed in the JSON data packet format, including timestamp, camera ID, GPS coordinates, confidence level, and screenshot URL.

[0107] While 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 variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A vehicle charging area fire warning system integrating AI recognition and intelligent inspection, characterized by: include: Front-end image acquisition module: This includes several explosion-proof high-definition cameras with infrared night vision capabilities, deployed around the charging piles. AI recognition model: Dual-modal input, processing visible light imagery and thermal imaging data simultaneously, embedding a channel-spatial attention module based on an improved lightweight network, and dynamically adjusting alarm trigger thresholds; Alarm information distribution module: multi-level alarm strategy, including level one alarm and level two alarm, alarm information is distributed in JSON data packet format; Intelligent inspection management module: Create inspection plans through the web platform, and the app verifies the location of inspectors through GPS geofencing and Bluetooth beacons to generate structured inspection reports; Alarm manual processing module: After receiving the alarm information, the on-duty personnel must select the processing method within the specified time, and all operation records are synchronized to the Web platform log.

2. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 1 is characterized by: The AI recognition model is based on the MobileNetV3 backbone network and is embedded with a self-developed channel-spatial attention module (CSAM) to improve sensitivity to small-scale smoke.

3. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 2 is characterized by: The alarm information distribution module includes the following functions: Level 1 alarm: triggers the on-site sound and light alarm; Level 2 alarm: Push information to the on-duty personnel’s mobile app, send email to the management platform, and activate the fire sprinkler system.

4. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 3 is characterized by: The intelligent inspection management module includes the following functions: Create inspection plans on the Web platform; The app uses GPS geofencing and Bluetooth beacon verification to ensure that inspectors reach the specified location. Generate structured inspection reports and support exporting records by time, area, and device type.

5. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 4 is characterized by: The alarm manual processing module includes the following functions: After receiving the alarm information, the on-duty personnel must choose a handling method within 3 minutes; All operation records (including operator, time, and on-site screenshots) are synchronized to the Web platform log.

6. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 5 is characterized by: It also includes a sensor data assisted judgment module, which combines smoke sensor and temperature sensor data and uses algorithm matching confidence to assist AI recognition.

7. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 6 is characterized by: It also includes security protection measures, including App-side device offline reminders, front-end image acquisition program online detection, and sensor device online detection.

8. A vehicle charging area fire warning method integrating AI recognition and intelligent inspection, characterized in that: The following steps are involved: Step 1: Collect image data of the area surrounding the charging pile through the front-end image acquisition module; Step 2: Input the collected image data into the AI recognition model for dual-modal input processing to identify potential fire risks; Step 3: Based on the recognition results, the alarm information distribution module triggers the on-site sound and light alarm or pushes the alarm information to the on-duty personnel; Step 4: Create an inspection plan through the intelligent inspection management module and verify that the inspection personnel arrive at the specified location; Step 5: Record the alarm processing process through the alarm manual processing module and synchronize it to the Web platform log.

9. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 8 is characterized by: In step 2, the AI recognition model is based on the MobileNetV3 backbone network, embedded with the self-developed channel-spatial attention module (CSAM), and dynamically adjusts the alarm trigger threshold.

10. The vehicle charging area fire warning system integrating AI recognition and intelligent inspection according to claim 9 is characterized by: In step 3, the alarm information is distributed in JSON data packet format, including timestamp, camera ID, GPS coordinates, confidence level, and screenshot URL.

Images