Ship fire source point positioning method
By combining infrared radiation images and visible light visual images, using the YOLOv5 network model, the automatic positioning and analysis of ship fire source points is achieved, and the problem of low efficiency of fire monitoring misjudgment and real-time patrol inspection in the existing technology is solved, and fire prevention is improved Capacity and ship safety.
Patent Information
- Application Number
- CN202410988619.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-08
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing ship fire monitoring system is prone to misjudgment in complex working conditions, and cannot achieve accurate fire monitoring and analysis, resulting in low efficiency in early fire information identification and inability to achieve real-time continuous patrol.
Using visual image data acquisition, data preprocessing, image weighted fusion, fire recognition image screening, fire source identification database and fire source detection model based on YOLOv5 network, the fire source detection model is realized through the combination of infrared radiation images and visible visual images. Automated positioning and analysis.
It realizes automated analysis and positioning of fire source points, reduces labor costs, improves fire prevention and prediction capabilities, enhances ship safety, and meets the needs of real-time fire monitoring and early warning throughout the ship.
Smart Images

Figure CN118918189A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of ship fire fighting systems, and in particular relates to a method for locating a ship fire source. Background Art
[0002] As the core equipment for water operations, ship equipment is an important guarantee for the development of marine resources and the maintenance of marine sovereignty. The development and utilization of marine resources and various operational tasks have continuously increased the requirements for the quality and safety of ships. As an important content to ensure the safety of ship use, ship fire is the top priority of daily ship disaster protection drills. Due to the limited resources that ships can carry, the resources that can be provided to ships for fire treatment are limited under the premise of ensuring the completion of various tasks. Therefore, the most effective solution is always to prevent problems before they happen. In the early stage of ship fires, effective positioning and analysis can be carried out to facilitate rapid response, early intervention, prevent the development of fire sources, and prevent the occurrence of large-scale fires. In order to solve the problems of low efficiency of early manual inspections, limited coverage of inspection areas, high personnel costs, and inability to achieve real-time continuous inspections, various fire monitoring equipment have been widely installed on important ships at present. Various high temperature, odor, smoke and other elements on the ship are detected to obtain fire source information. However, due to the complex working state of the ship, non-fire elements such as high temperature gas, smoke and odor may also be generated in different types of operational tasks, which will lead to misjudgment of fire information and fail to achieve accurate fire monitoring and analysis. Summary of the invention
[0003] The purpose of the present invention is to provide a ship fire source location method for monitoring ship fire hazards and being able to quickly locate and analyze fire source information when a fire occurs, based on actual needs.
[0004] To achieve the above purpose, the present invention adopts the following technical solution.
[0005] A method for locating a fire source on a ship comprises the following steps:
[0006] Step 1: Visual image data acquisition, wherein the visible light data includes: infrared radiation images and visible light visual images in the ship area acquired by the visual acquisition device;
[0007] Step 2: Data preprocessing, including preprocessing of infrared radiation images;
[0008] Step 3: weighted fusion of image data, including weighted fusion of infrared radiation image and visible light visual image of the same ship area to form a visible light fire identification image;
[0009] Step 4: Screening of fire recognition images, including analyzing pixel information in the image and the matching degree between the infrared radiation image, the visible light visual image and the recognition image to screen the recognition image;
[0010] Step 5: Establish a fire source recognition database based on recognition images; including extracting recognition images for different fire source points and different fire characteristics in the target ship area, and establishing a ship area infrared radiation image database, a ship area visible light visual image database, and a ship area recognition image database based on the extracted data. If there is historical data, establish a historical fire image database based on the historical image data of the corresponding ship area;
[0011] Step 6: Establishing a machine learning database, including: based on the data in the aforementioned fire source identification database, selecting no less than 10-20% of the images containing fire source information for manual or machine annotation to establish a test database, and using the remaining samples in the database as a training database;
[0012] Step 7: Establish a fire source detection model based on the YOLOv5 network, and train and optimize the fire source detection model using the aforementioned test database and training database;
[0013] Step 8: Use the fire source detection model trained with the fire image data to be analyzed, set model parameters, and perform fire source location and identification.
[0014] For further improvement or specific implementation steps of the aforementioned ship fire source location method, the data acquisition device in step one includes: a camera for acquiring visible light visual images, and an infrared radiation camera for acquiring infrared radiation images.
[0015] Further improvement or specific implementation steps of the above-mentioned ship fire source location method, the pre-processing in step 2 specifically includes:
[0016] B1. Grayscale processing of infrared radiation images, specifically, for any pixel a(i, j) in the infrared radiation image A, its grayscale image A(i, j) is obtained by channel space transformation, where:
[0017] A(i,j)=ω1R(i,j)+ω2G(i,j)+ω2B(i,j
[0018] Among them, R(i, j), G(i, j), B(i, j) refer to the three color channel values of the original image, ω1, ω2, ω2 are the three channel weights, and (i, j) refers to the pixel coordinates.
[0019] Further improvement or specific implementation steps of the above-mentioned ship fire source location method, step three specifically includes:
[0020] C1. Image resolution correction, specifically, refers to the unification of the resolution of the image data of the ship area collected by different devices according to the camera's own resolution and the distance of the collection area, so that the image of the ship area has a relatively consistent resolution;
[0021] C2. Weighted fusion of image data: The fused fire identification image C(i, j) = ωaA(i, j) + ωbB(i, j) is obtained through weighted fusion; wherein A(i, j) refers to the grayscale image of the infrared radiation image, B(i, j) refers to the visible light visual image, and ωa and ωb refer to the weighted weights of the infrared radiation image and the visible light visual image.
[0022] For further improvement or specific implementation steps of the aforementioned ship fire source location method, image resolution correction refers to unifying the resolution of other images based on the resolution of the visible light image obtained by the camera with the highest resolution or the camera closest to the fire ship area.
[0023] Further improvement or specific implementation steps of the above-mentioned ship fire source location method, step 4 specifically includes:
[0024] D1. Based on the fire identification image C(i, j) and its corresponding infrared radiation image grayscale image A(i, j) and visible light visual image B(i, j), obtain the corresponding pixel gradient map and edge histogram;
[0025] D2. Determine the clarity index based on the pixel gradient of the fire identification image
[0026]
[0027] Where kx is the pixel gradient value in the x direction of the fire recognition image, ky is the pixel gradient value in the y direction of the fire recognition image, and (x, y) is the pixel gradient map coordinate;
[0028] D3. Visibility index based on grayscale distribution of fire recognition images
[0029]
[0030] in It refers to the average gray value of the fire identification image;
[0031] D4, determining a similarity index f3=β1G1+β2G2+β3G3 based on the correlation between the fire identification image, the infrared radiation image and the visible light visual image;
[0032]
[0033] Among them G A (x,y),G B(x,y),G C (x, y) refer to the edge histograms of the visible light vision image, infrared radiation image, and fire recognition image, respectively;
[0034] D5. Based on the above evaluation indicators, a screening indicator F = μ1f1 + μ2f2 + μ3f3 is established;
[0035] Wherein μ1, μ2, and μ3 refer to the index weight coefficients given based on historical data or expert experience; D6, based on the screening index F, selects the fire recognition image with the best index.
[0036] Further improvement or specific implementation steps of the above-mentioned ship fire source positioning method, step B1, grayscale processing of the infrared radiation image also includes optimization processing of the grayscale image A(i, j), specifically referring to: using a Gaussian filter to filter A(i, j); using an edge detection operator to perform edge detection on the filtered image, and extract edge pixels, and eliminate non-edge pixel information.
[0037] Further improvement or specific implementation steps of the aforementioned ship fire source positioning method, the extraction of identification images for different fire source points and different fire characteristics in the target ship area means: full extraction of full-factor images of different environments, different time points, different fire source types, different causes of fire where fires may occur in the ship area, and different shooting distances and different resolutions of the acquisition equipment for the same ship area.
[0038] Its beneficial effects are:
[0039] The ship fire source location method of the present invention fully simulates the principle of manual judgment of fire source information, and provides a solution based on automatic image information recognition and extraction, which can automatically analyze whether there is a fire source in the image based on regional image data information combined with fire image information characteristics, extract fire source information, and then realize automatic analysis and location of the fire source. It can fully combine and utilize the resources of the existing ship monitoring system, effectively reduce the current labor cost of ship fire emergency response, improve the fire prevention and prediction capabilities, and further improve ship safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 It is a flow chart of the method for locating the source of ship fire. DETAILED DESCRIPTION
[0041] The present invention is described in detail below in conjunction with specific embodiments.
[0042] like Figure 1As shown, it is an implementation flow chart of a method for locating the source of a ship fire of the present invention. The present invention is mainly used to meet the needs of marine ship fire monitoring and cargo location analysis. In order to achieve more comprehensive automatic fire source identification and positioning as much as possible under the premise of giving priority to software and hardware costs, the present invention combines the image device already available in the existing ship monitoring system, optimizes the image data processing and feature fusion processing, effectively enhances the data volume and dimension in the single image data related to the fire source, and then visualizes the required fire source information and features in the image, greatly reduces the difficulty of running the detection network, reduces the system software and hardware costs and computing power costs, so that it can meet the needs of real-time and fully quantitative fire monitoring and early warning for the entire ship.
[0043] It specifically includes the following steps:
[0044] Step 1: Visual image data acquisition, wherein the visible light data includes: infrared radiation images and visible light visual images in the ship area acquired by the visual acquisition device;
[0045] The data acquisition device includes: a camera for acquiring visible light visual images and an infrared radiation camera for acquiring infrared radiation images.
[0046] Step 2: Data preprocessing, including preprocessing of infrared radiation images; the necessary preprocessing includes
[0047] B1. Grayscale processing of infrared radiation image. For any pixel a(i, j) in infrared radiation image A, its grayscale image A(i, j) is obtained by channel space transformation, where:
[0048] A(i,j)=ω1R(i,j)+ω2G(i,j(+ω2B(i,j
[0049] Among them, R(i, j), G(i, j), B(i, j) refer to the three color channel values of the original image, ω1, ω2, ω2 are the three channel weights, and (i, j) refers to the pixel coordinates.
[0050] B2. Optimizing the grayscale image A(i, j), specifically: filtering A(i, j) using a Gaussian filter; performing edge detection on the filtered image using an edge detection operator, extracting edge pixels, and eliminating non-edge pixel information.
[0051] Step 3: Weighted fusion of image data, including weighted fusion of infrared radiation image and visible light visual image of the same ship area to form a visible light fire identification image, specifically:
[0052] C1. Image resolution correction: The image data of the ship area collected by different devices are processed to a uniform resolution according to the camera's own resolution and the distance of the collection area, so that the image of the ship area has a relatively consistent resolution;
[0053] In actual implementation, in order to ensure that high-value and high-information visual images are retained as much as possible, when correcting the image resolution, it is best to unify the resolution of other images based on the resolution of the visible light image obtained by the camera with the highest resolution or the camera closest to the fire ship area, so as to ensure that high-quality images maintain the best resolution.
[0054] C2. Weighted fusion of image data: The fused fire identification image C(i, j) = ωaA(i, j) + ωbB(i, j) is obtained through weighted fusion; wherein A(i, j) refers to the grayscale image of the infrared radiation image, B(i, j) refers to the visible light visual image, and ωa and ωb refer to the weighted weights of the infrared radiation image and the visible light visual image.
[0055] Step 4: Screening of fire identification images, including analyzing pixel information in the image and the matching degree between the infrared radiation image, the visible light visual image and the identification image to screen the identification image; the following is a specific implementation method:
[0056] D1. Based on the fire identification image C(i, j) and its corresponding infrared radiation image grayscale image A(i, j) and visible light visual image B(i, j), obtain the corresponding pixel gradient map and edge histogram;
[0057] D2. Determine the clarity index based on the pixel gradient of the fire identification image
[0058]
[0059] Where kx is the pixel gradient value in the x direction of the fire recognition image, ky is the pixel gradient value in the y direction of the fire recognition image, and (x, y) is the pixel gradient map coordinate;
[0060] D3. Visibility index based on grayscale distribution of fire recognition images
[0061]
[0062] in It refers to the average gray value of the fire identification image;
[0063] D4, determining a similarity index f3=β1G1+β2G2+β3G3 based on the correlation between the fire identification image, the infrared radiation image and the visible light visual image;
[0064]
[0065] Among them G A (x,y),G B (x,y),G C (x, y) refer to the edge histograms of the visible light vision image, infrared radiation image, and fire recognition image, respectively;
[0066] D5. Based on the above evaluation indicators, a screening indicator F = μ1f1 + μ2f2 + μ3f3 is established;
[0067] Among them, μ1, μ2, and μ3 refer to the indicator weight coefficients based on historical data or expert experience;
[0068] D6. Select the fire recognition image with the best index based on the screening index F.
[0069] Step 5: Establish a fire source recognition database based on recognition images; including extracting recognition images for different fire source points and different fire characteristics in the target ship area.
[0070] In order to obtain full information on ship fires and to avoid omissions and misjudgments as much as possible, when extracting identification images for different fire sources and different fire characteristics in the target ship area, full extraction should be carried out for full-factor images of different environments, different time points, different fire source types, different causes of fires that may occur in the ship area, and different shooting distances and different resolutions of the acquisition equipment for the same ship area.
[0071] Based on the extracted data, establish a ship area infrared radiation image database, a ship area visible light visual image database, and a ship area recognition image database. If there is historical data, establish a historical fire image database based on the historical image data of the corresponding ship area;
[0072] Step 6: Establishing a machine learning database, including: based on the data in the aforementioned fire source identification database, selecting no less than 10-20% of the images containing fire source information for manual or machine annotation to establish a test database, and using the remaining samples in the database as a training database;
[0073] Step 7: Establish a fire source detection model based on the YOLOv5 network, and train and optimize the fire source detection model with the aforementioned test database and training database. The YOLOv5 network model is a mature and stable target detection model that can meet the needs of the present invention for real-time monitoring and real-time processing of ships.
[0074] Step 8: Use the fire source detection model trained with the fire image data to be analyzed, set model parameters, and perform fire source location and identification.
[0075] Through the aforementioned steps, the present invention establishes a mechanism for rapid positioning and identification in a large number of ship video surveillance images by performing data fusion processing and feature screening optimization on infrared radiation image data and visible light visual image data. In the later stage, through continuous updating and upgrading of data, continuous optimization training of the model can be achieved based on key features, thereby continuously improving the accuracy of fire judgment and fire source identification.
[0076] By inputting the fire identification images with optimized image data structure as identification and training elements, the efficiency of model training and recognition can be increased, while reducing the computing power cost required for the identification and establishment of image features, providing the software and hardware foundation for the rapid processing of the entire ship's full data.
[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.
Claims
1. A method for locating the source of a ship fire, characterized in that: The steps include: Step 1: Visual image data collection, wherein the visible light data includes: infrared radiation images and visible light visual images in the ship area acquired by the visual collection device; Step 2: Data preprocessing, including preprocessing of infrared radiation images; Step 3: weighted fusion of image data, including weighted fusion of infrared radiation image and visible light visual image of the same ship area to form a visible light fire identification image; Step 4: Screening of fire recognition images, including analyzing pixel information in the image and the matching degree between the infrared radiation image, the visible light visual image and the recognition image to screen the recognition image; Step 5: Establish a fire source recognition database based on recognition images; including extracting recognition images for different fire source points and different fire characteristics in the target ship area, and establishing a ship area infrared radiation image database, a ship area visible light visual image database, and a ship area recognition image database based on the extracted data. If there is historical data, establish a historical fire image database based on the historical image data of the corresponding ship area; Step 6: Establishing a machine learning database, including: based on the data in the aforementioned fire source identification database, selecting no less than 10-20% of the images containing fire source information for manual or machine annotation to establish a test database, and using the remaining samples in the database as a training database; Step 7: Establish a fire source detection model based on the YOLOv5 network, and train and optimize the fire source detection model using the aforementioned test database and training database; Step 8: Input the fire image data to be analyzed into the trained fire source detection model, set the model parameters, and perform fire source location and identification.
2. The method for locating the source of a ship fire according to claim 1, characterized in that: The data acquisition device in step one includes: a camera for acquiring visible light visual images and an infrared radiation camera for acquiring infrared radiation images.
3. The method for locating the source of a ship fire according to claim 1, characterized in that: The pretreatment in step 2 specifically includes: B1. Grayscale processing of infrared radiation images, specifically, for any pixel point a(i, j) in the infrared radiation image A, its grayscale image A(i, j) is obtained by channel space transformation, where: A(i,j)=ω1R(i,j)+ω2G(i,j)+ω2B(i,j) Among them, R(i,j), G(i,j), B(i,j) refer to the three color channel values of the original image, ω1, ω2, ω2 are the three channel weights, and (i,j) refers to the pixel coordinates.
4. The method for locating the source of a ship fire according to claim 3, characterized in that: The step three specifically includes: C1. Image resolution correction, specifically, refers to the unification of the resolution of the image data of the ship area collected by different devices according to the camera's own resolution and the distance of the collection area, so that the image of the ship area has a relatively consistent resolution; C2. Weighted fusion of image data: Obtain the fused fire identification image C(i,j)=ω through weighted fusion a A(i,j)+ω b B(i,j); where A(i,j) refers to the grayscale image of the infrared radiation image, B(i,j) refers to the visible light visual image, ω a and ω b It refers to the weighted weight of infrared radiation image and visible light visual image.
5. The method for locating the source of a ship fire according to claim 4, characterized in that: The image resolution correction refers to unifying the resolution of other images based on the resolution of the visible light image acquired by the camera with the highest resolution or the camera closest to the fire ship area.
6. The method for locating the source of a ship fire according to claim 4, characterized in that: The step 4 specifically includes: D1. Based on the fire identification image C(i, j) and its corresponding infrared radiation image grayscale image A(i, j) and visible light visual image B(i, j), obtain the corresponding pixel gradient map and edge histogram; D2. Determine the clarity index based on the pixel gradient of the fire identification image where k x is the pixel gradient value in the x direction of the fire recognition image, k y is the pixel gradient value in the y direction of the fire identification image, (x, y) is the pixel gradient map coordinate; D3. Visibility index based on grayscale distribution of fire recognition images in It refers to the average gray value of the fire identification image; D4, determining a similarity index f3=β1G1+β2G2+β3G3 based on the correlation between the fire identification image, the infrared radiation image and the visible light visual image; Among them G A (x,y),G B (x,y),G C (x, y) refer to the edge histograms of the visible light vision image, infrared radiation image, and fire recognition image, respectively; D5. Based on the above evaluation indicators, a screening indicator F = μ1f1 + μ2f2 + μ3f3 is established; Among them, μ1, μ2, and μ3 refer to the indicator weight coefficients based on historical data or expert experience; D6. Select the fire recognition image with the best index based on the screening index F.
7. The method for locating the source of a ship fire according to claim 1, characterized in that: Step B1, the grayscale processing of the infrared radiation image also includes optimizing the grayscale image A(i,j), specifically: filtering A(i,j) using a Gaussian filter; performing edge detection on the filtered image using an edge detection operator, extracting edge pixels, and eliminating non-edge pixel information.
8. The method for locating the source of a ship fire according to claim 1, characterized in that: The extraction of identification images for different fire sources and different fire characteristics in the target ship area means: full extraction of full-factor images of different shooting distances and different resolutions for different environments, different time points, different fire source types, different causes of fire where fires may occur in the ship area, and based on the acquisition equipment for the same ship area.
Citation Information
Patent Citations
Fire source detection, identification and early warning method for fire-fighting robot
CN114399882A
Fire positioning system based on infrared video monitoring
CN116229668A
Fire detection method and system based on infrared visible light image fusion
CN117689568A
Fire detection method and device, electronic equipment and storage medium
CN117876829A
Fire source identifying and positioning method based on infrared image and visible light image
CN117994490A
