Tunnel fire smoke diffusion detection method based on deep learning and image segmentation

Abstract

The invention discloses a tunnel fire smoke diffusion detection method based on deep learning and image segmentation, and belongs to the technical field of fire smoke diffusion detection. The method comprises the following steps: S1, data acquisition and processing: acquiring a monitoring video in a tunnel environment by using an image acquisition module, obtaining diversified smoke image data when a fire occurs in a tunnel, and storing the obtained smoke image data in a data storage module; s2, smoke image segmentation and enhancement: utilizing an image processing module to segment and enhance the smoke image data acquired in the S1; s3, smoke recognition: after image segmentation and enhancement are completed, smoke target detection is performed on the processed image by using a YOLO v5 deep learning model based on a smoke detection module; s4, result analysis and visualization: utilizing a result analysis and visualization module to display the smoke recognition result obtained in the S3 in real time; and S5, data storage: storing a result displayed in the S4 in real time in a data storage module. Compared with the prior art, fire smoke diffusion can be accurately monitored in real time, timely control and early warning are achieved, and the influence of smoke diffusion on the tunnel environment and the external environment is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fire smoke diffusion detection, and specifically to a tunnel fire smoke diffusion detection method based on deep learning and image segmentation. Background Art

[0002] Smoke is the most harmful part of tunnel fires, which poses a serious threat to personnel safety, fire fighting and rescue, tunnel structure and environment. Statistical results show that toxic and harmful smoke produced by incomplete combustion is the main factor leading to death in tunnel fires. Traditional fire detection methods are mainly based on smoke, temperature and light sensors. When a fire occurs, physical quantities such as smoke, heat and light radiation generated by combustion are converted into electrical signals, transmitted to the fire alarm controller, and then the fire alarm and fire extinguishing system are activated.

[0003] With the development of artificial intelligence and computer vision technology, high-definition cameras, image processing technology and deep learning algorithms (such as convolutional neural networks (CNN)) can be used to train image data, analyze features such as smoke and flames in video images, improve the recognition accuracy and anti-interference ability of fire smoke and flames, improve the timeliness and accuracy of fire monitoring, and realize remote and visual monitoring of fires.

[0004] Traditional fire detection methods mainly use smoke, temperature and light sensors. When a fire occurs, physical quantities such as smoke, heat and light radiation generated by combustion are converted into electrical signals, transmitted to the fire alarm controller, and then the fire alarm and fire extinguishing system are activated. Traditional fire detection technology can effectively detect the occurrence of fires to a certain extent and provide timely alarms, but it has limitations, such as high prices and the need for regular maintenance and inspection, which increases maintenance costs and workload.

[0005] Smoke visually appears dynamic, translucent, and irregular in shape, making it difficult to effectively identify using traditional image processing methods. Although threshold-based segmentation methods perform well when lighting conditions are stable and the difference between the target and the background is obvious, it is difficult to find a suitable global threshold under the condition of uneven lighting in the tunnel, resulting in inaccurate segmentation results. Edge detection methods segment by capturing pixel brightness changes, but the boundaries of smoke in tunnels are fuzzy and the lighting is unstable, so edge detection is easily disturbed and cannot provide reliable results. Deep learning methods, such as convolutional neural networks (CNNs), although they have shown powerful segmentation capabilities in highly complex scenes, require a large amount of labeled data for training, and sample data of tunnel smoke is often difficult to obtain. In addition, the high computational cost of deep learning models also limits their application in real-time smoke detection.

[0006] Based on the above content, the present invention proposes a tunnel fire smoke diffusion detection method based on deep learning and image segmentation. Summary of the invention

[0007] 1. Technical Problems to be Solved by the Present Invention

[0008] The purpose of the present invention is to propose a tunnel fire smoke diffusion detection method based on deep learning and image segmentation to solve the problems raised in the background technology.

[0009] 2. Technical solution

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] A tunnel fire smoke diffusion detection method based on deep learning and image segmentation is implemented based on a tunnel fire smoke diffusion detection system. The system consists of an image acquisition module, an image processing module, a smoke detection module, a result analysis and visualization module, and a data storage module, and specifically includes the following contents:

[0012] The image acquisition module uses a high-resolution camera and is installed at a key position in the tunnel to capture the smoke image generated by the fire in real time. The image acquisition module is connected to the image processing module to transmit video data in real time.

[0013] The image processing module is composed of an embedded computing device or a high-performance computer, and is used to receive and process real-time video data from the image acquisition module; the image processing module includes a frame difference method unit, a color filtering unit, and a background subtraction unit. The image processing module extracts the smoke area from the background through a built-in segmentation algorithm and further enhances the segmented image to improve the visibility of the smoke area, and transmits the result to the detection unit;

[0014] The smoke detection module uses the YOLO v5 deep learning model, which runs on the computing platform of the image processing module; the smoke detection module is used to analyze the enhanced image, identify the location and shape of the smoke, and output the bounding box information of the smoke;

[0015] The result analysis and visualization module is an external device connected to the image processing module and is used to display the diffusion position, speed and coverage of smoke in real time;

[0016] A data storage module is used to record all smoke monitoring data and images generated during the detection process; the storage unit of the data storage module adopts a local hard disk or a remote database to facilitate the retrieval of historical data for analysis after an accident;

[0017] The method comprises the following steps:

[0018] S1. Data acquisition and processing: using the image acquisition module to collect monitoring videos in the tunnel environment, obtaining diversified smoke image data when a fire occurs in the tunnel, and storing the obtained smoke image data in the data storage module;

[0019] S2, smoke image segmentation and enhancement: using the image processing module to segment and enhance the smoke image data obtained in S1;

[0020] S3, smoke recognition: After completing image segmentation and enhancement, based on the smoke detection module, the YOLO v5 deep learning model is used to perform smoke target detection on the processed image;

[0021] S4, result analysis and visualization: using the result analysis and visualization module to display the smoke identification results obtained in S3 in real time;

[0022] S5, data storage: the results displayed in real time in S4 are stored in the data storage module.

[0023] Preferably, S2 specifically includes the following contents:

[0024] S2.1, Frame difference method to detect smoke motion area:

[0025] The pixel difference between adjacent video frames is calculated by frame difference method to extract the moving area of ​​smoke, highlight the dynamic change part, and eliminate the interference of static background on smoke detection;

[0026] S2.2, Color Filter Enhanced Smoke Area:

[0027] Further enhance the smoke area through color filtering, specifically: using HSV or other color spaces, defining the color range of the smoke as the filtering condition, removing the non-smoke area, further filtering the interference area, and enhancing the visibility of the smoke;

[0028] S2.3, Background subtraction method to eliminate static interference:

[0029] First, multiple frames of images are averaged or other statistical operations are performed to generate a background model; then, each frame of image is differentiated from the background model, and the difference portion is retained, which is the smoke area, to eliminate the influence of static background on detection accuracy;

[0030] S2.4, comprehensive processing: obtain the image generated by S2.1 to S2.4, perform comprehensive processing on it, generate a more accurate smoke segmentation result, retain the main area of ​​smoke to the greatest extent, reduce the risk of false detection and missed detection, and use the final segmentation result to display the accurate outline of smoke in the entire video frame;

[0031] S2.5, Image labeling: The processed image data needs to be labeled to accurately mark the smoke area to supervise the training of the deep learning model; the labeling work includes labeling the smoke boundary, position, and area information, and provides high-quality training data for subsequent image segmentation and detection algorithms.

[0032] Preferably, S3 specifically includes the following contents:

[0033] The YOLO v5 deep learning model is used to monitor the presence of smoke in real time and generate accurate bounding boxes. The smoke diffusion path is determined by tracking the position of the smoke bounding box in consecutive frames. The smoke diffusion speed is calculated based on the frame rate and the change in the front smoke edge position in adjacent frames. By tracking the smoke diffusion position and speed, key data support is provided for the fire warning system.

[0034] 3. Beneficial effects

[0035] The present invention proposes a tunnel fire smoke diffusion detection method based on deep learning and image segmentation, which has the following beneficial effects compared with the prior art:

[0036] (1) Wider adaptability to smoke types: Image segmentation can detect smoke of different types and concentrations. The deep learning model can identify the characteristics of smoke in different environments and reduce dependence on light and environmental factors.

[0037] (2) Higher robustness in complex environments: Deep learning technology can train models through large amounts of data, enabling them to effectively detect smoke under complex lighting conditions and different background environments, overcoming the interference problems of traditional optical and ion detection methods.

[0038] (3) Real-time monitoring and dynamic response: Deep learning and image segmentation can continuously monitor the smoke in the video, detect the speed and location of smoke diffusion, and respond more sensitively.

[0039] (4) Higher detection accuracy: Compared with traditional optical and sensor-based detection methods, smoke detection technology based on image processing can improve detection accuracy through enhanced processing and image segmentation, especially in scenes with thin smoke or complex backgrounds.

[0040] In summary, the present invention realizes the real-time collection of smoke data through the camera, and adopts the image processing method combining the frame difference method, color filtering and background subtraction technology to accurately segment and enhance the smoke area, significantly improve the visibility of the smoke target, improve the monitoring quality, and be able to detect smoke diffusion earlier, thereby improving the timeliness and accuracy of monitoring. At the same time, through the YOLOv5 deep learning model, a large amount of image data can be efficiently processed in a short time, greatly improving efficiency. In practical applications, the system can process monitoring videos in real time, ensure dynamic monitoring and response to smoke, and improve the response speed in emergency situations. Compared with traditional smoke detection equipment, the present invention can monitor the spread of fire smoke in real time and accurately, control and warn in time, and reduce the impact of smoke diffusion on the tunnel environment and the external environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is an overall flow chart of the tunnel fire smoke diffusion detection method based on deep learning and image segmentation proposed by the present invention;

[0042] Figure 2 This is a flow chart of the image segmentation and enhancement method mentioned in Example 1 of the present invention;

[0043] Figure 3 This is a comparison diagram of the image enhancement effects before and after mentioned in Example 1 of the present invention. DETAILED DESCRIPTION

[0044] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0045] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0047] The tunnel fire smoke diffusion detection method based on deep learning and image segmentation proposed by the present invention is described below with reference to specific examples and drawings. The specific contents are as follows.

[0048] Embodiment 1:

[0049] A tunnel fire smoke diffusion detection method based on deep learning and image segmentation is implemented based on a tunnel fire smoke diffusion detection system. The system consists of an image acquisition module, an image processing module, a smoke detection module, a result analysis and visualization module, and a data storage module, and specifically includes the following contents:

[0050] The image acquisition module uses a high-resolution shooting device (high-resolution camera) and is installed at a key position in the tunnel to capture the smoke image generated by the fire in real time; the image acquisition module is data-connected with the image processing module to transmit video data in real time;

[0051] The image processing module is composed of an embedded computing device or a high-performance computer, including a central processing unit (CPU) or a graphics processing unit (GPU), and is used to receive and process real-time video data from the image acquisition module; the image processing module includes a frame difference method unit, a color filtering unit, and a background subtraction unit. The image processing module extracts the smoke area from the background through a built-in segmentation algorithm and further enhances the segmented image to improve the visibility of the smoke area, and transmits the result to the detection unit;

[0052] The smoke detection module uses the YOLO v5 deep learning model, which runs on the computing platform of the image processing module; the smoke detection module is used to analyze the enhanced image, identify the location and shape of the smoke, and output the bounding box information of the smoke;

[0053] The result analysis and visualization module is an external device connected to the image processing module and is used to display the diffusion position, speed and coverage of smoke in real time;

[0054] The data storage module is used to record all smoke monitoring data and images generated during the detection process; the storage unit of the data storage module uses a local hard disk or a remote database to facilitate the retrieval of historical data for analysis after an accident;

[0055] The above method comprises the following steps:

[0056] S1. Data acquisition and processing: using the image acquisition module to collect monitoring videos in the tunnel environment, obtaining diversified smoke image data when a fire occurs in the tunnel, and storing the obtained smoke image data in the data storage module;

[0057] S2. Smoke image segmentation and enhancement: Use the image processing module to segment and enhance the smoke image data obtained in S1; see Figure 2 , specifically including the following:

[0058] S2.1, Frame difference method to detect smoke motion area:

[0059] The pixel difference between adjacent video frames is calculated by frame difference method to extract the moving area of ​​smoke, highlight the dynamic change part, and eliminate the interference of static background on smoke detection;

[0060] S2.2, Color Filter Enhanced Smoke Area:

[0061] The smoke area is further enhanced by color filtering, specifically: using HSV or other color spaces, defining the color range of the smoke as the filtering condition, removing non-smoke areas, further filtering interference areas, and enhancing the visibility of the smoke (such as Figure 3 As shown, the upper part is the original image detected by YOLO, and the lower part is the enhanced image effect);

[0062] S2.3, Background subtraction method to eliminate static interference:

[0063] First, perform average or other statistical operations on multiple frames of images to generate a background model; then, perform a difference between each frame of image and the background model, retain the difference part, which is the smoke area, to eliminate the influence of static background on detection accuracy;

[0064] S2.4, comprehensive processing: obtain the image generated by S2.1 to S2.4, perform comprehensive processing on it, generate a more accurate smoke segmentation result, retain the main area of ​​smoke to the greatest extent, reduce the risk of false detection and missed detection, and use the final segmentation result to display the accurate outline of smoke in the entire video frame;

[0065] S2.5, Image labeling: The processed image data needs to be labeled and the smoke area is accurately marked to supervise the training of the deep learning model; the labeling work includes the labeling of smoke boundary, position, and area information, and provides high-quality training data for subsequent image segmentation and detection algorithms;

[0066] S3, smoke recognition: After completing image segmentation and enhancement, based on the smoke detection module, the YOLO v5 deep learning model is used to detect smoke targets on the processed images; specifically, the following contents are included:

[0067] Use the YOLO v5 deep learning model to monitor the presence of smoke in real time and generate accurate bounding boxes; determine the path of smoke diffusion by tracking the position of the smoke bounding box in consecutive frames; calculate the smoke diffusion speed based on the frame rate and the change in the front smoke edge position in adjacent frames; provide key data support for the fire warning system by tracking the smoke diffusion position and speed;

[0068] S4, result analysis and visualization: using the result analysis and visualization module to display the smoke identification results obtained in S3 in real time;

[0069] S5, data storage: the results displayed in real time in S4 are stored in the data storage module.

[0070] In summary, the present invention realizes the real-time collection of smoke data through the camera, and adopts the image processing method combining the frame difference method, color filtering and background subtraction technology to accurately segment and enhance the smoke area, significantly improve the visibility of the smoke target, improve the monitoring quality, and be able to detect smoke diffusion earlier, thereby improving the timeliness and accuracy of monitoring. At the same time, through the YOLOv5 deep learning model, a large amount of image data can be efficiently processed in a short time, greatly improving efficiency. In practical applications, the system can process monitoring videos in real time, ensure dynamic monitoring and response to smoke, and improve the response speed in emergency situations. Compared with traditional smoke detection equipment, the present invention can monitor the spread of fire smoke in real time and accurately, control and warn in time, and reduce the impact of smoke diffusion on the tunnel environment and the external environment.

[0071] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and improved concepts of the present invention within the technical scope disclosed by the present invention, and they should be covered by the protection scope of the present invention.

Claims

1. A tunnel fire smoke diffusion detection method based on deep learning and image segmentation, characterized in that: The method is implemented based on a tunnel fire smoke diffusion detection system, which consists of an image acquisition module, an image processing module, a smoke detection module, a result analysis and visualization module, and a data storage module, and specifically includes the following contents: The image acquisition module uses a high-resolution camera and is installed at a key position in the tunnel to capture the smoke image generated by the fire in real time. The image acquisition module is connected to the image processing module to transmit video data in real time. The image processing module is composed of an embedded computing device or a high-performance computer, and is used to receive and process real-time video data from the image acquisition module; the image processing module includes a frame difference method unit, a color filtering unit, and a background subtraction unit. The image processing module extracts the smoke area from the background through a built-in segmentation algorithm and further enhances the segmented image to improve the visibility of the smoke area, and transmits the result to the detection unit; The smoke detection module uses the YOLO v5 deep learning model and runs on the computing platform of the image processing module; the smoke detection module is used to analyze the enhanced image, identify the location and shape of the smoke, and output the bounding box information of the smoke; The result analysis and visualization module is an external device connected to the image processing module and is used to display the diffusion position, speed and coverage of smoke in real time; Data storage module, used to record all smoke monitoring data and images generated during the detection process; The storage unit of the data storage module adopts a local hard disk or a remote database, so as to retrieve historical data for analysis after an accident; The method comprises the following steps: S1. Data acquisition and processing: using the image acquisition module to collect monitoring videos in the tunnel environment, obtaining diversified smoke image data when a fire occurs in the tunnel, and storing the obtained smoke image data in the data storage module; S2, smoke image segmentation and enhancement: using the image processing module to segment and enhance the smoke image data obtained in S1; S3, smoke recognition: After completing image segmentation and enhancement, based on the smoke detection module, the YOLO v5 deep learning model is used to perform smoke target detection on the processed image; S4, result analysis and visualization: using the result analysis and visualization module to display the smoke identification results obtained in S3 in real time; S5, data storage: the results displayed in real time in S4 are stored in the data storage module.

2. The tunnel fire smoke diffusion detection method based on deep learning and image segmentation according to claim 1 is characterized in that: The S2 specifically includes the following contents: S2.1, Frame difference method to detect smoke motion area: The pixel difference between adjacent video frames is calculated by frame difference method to extract the moving area of ​​smoke, highlight the dynamic change part, and eliminate the interference of static background on smoke detection; S2.2, Color Filter Enhanced Smoke Area: Further enhance the smoke area through color filtering, specifically: using HSV or other color spaces, defining the color range of the smoke as the filtering condition, removing the non-smoke area, further filtering the interference area, and enhancing the visibility of the smoke; S2.3, Background subtraction method to eliminate static interference: First, multiple frames of images are averaged or other statistical operations are performed to generate a background model; then, each frame of image is differentiated from the background model, and the difference portion is retained, which is the smoke area, to eliminate the influence of static background on detection accuracy; S2.4, comprehensive processing: obtain the image generated by S2.1 to S2.4, perform comprehensive processing on it, generate a more accurate smoke segmentation result, retain the main area of ​​smoke to the greatest extent, reduce the risk of false detection and missed detection, and use the final segmentation result to display the accurate outline of smoke in the entire video frame; S2.5, Image labeling: The processed image data needs to be labeled to accurately mark the smoke area to supervise the training of the deep learning model; the labeling work includes labeling the smoke boundary, position, and area information, and provides high-quality training data for subsequent image segmentation and detection algorithms.

3. The tunnel fire smoke diffusion detection method based on deep learning and image segmentation according to claim 1 is characterized in that: The S3 specifically includes the following contents: The YOLO v5 deep learning model is used to monitor the presence of smoke in real time and generate accurate bounding boxes. The smoke diffusion path is determined by tracking the position of the smoke bounding box in consecutive frames. The smoke diffusion speed is calculated based on the frame rate and the change in the front smoke edge position in adjacent frames. By tracking the smoke diffusion position and speed, key data support is provided for the fire warning system.

Images