An explosion flame range extraction method based on Yolov5 artificial intelligence identification matching and YCbCr color space model

Abstract

The application discloses an explosion flame range extraction method based on a Yolov5 artificial intelligence recognition matching and a YCbCr color space model, and solves the problems that the flame range extracted by using the prior art has a large error with the original flame range under explosion conditions, and the flame edge processing is relatively rough. The application firstly determines the position of the flame area in the explosion process by using a Yolov5 detection algorithm, then accurately processes the flame range in the explosion process by using an explosion flame range extraction method based on the YCbCr color space model, and finally realizes flame parameter calculation according to the corresponding relationship of calibration, so as to realize the purpose of accurately extracting the explosion flame range.

Description

Technical Field

[0001] This invention relates to a method for extracting the range of explosion flames, specifically a method for extracting the range of explosion flames based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model. Background Technology

[0002] In the field of explosion mechanics, the range and propagation characteristics of flames have always been a key focus in studying the destructive effects of pyrotechnics. Accurately measuring the position and range of flames at different times, and clarifying the propagation patterns and characteristics of flames, is crucial for assessing the magnitude of the force field of pyrotechnics.

[0003] Early flame detection relied primarily on smoke detectors, temperature sensors, and gas monitors. Typically, a threshold was set based on the combustion characteristics and intermediate products of the flame; when these measuring devices detected parameters within the set-up scene exceeding this threshold, a flame was considered to be present. While this flame detection method was relatively simple and easy to implement, it could only detect the presence or absence of a flame within the scene; precise flame location and calculation of flame parameters remained quite challenging.

[0004] In recent years, artificial intelligence technology has been gradually applied to target detection in images. Artificial intelligence image recognition technologies such as YOLOv5 can accurately identify the location of flames. However, in practical applications and detection, this technology can only determine the existence of flames and their location in most cases, but cannot calculate the range and characteristics of flames in more detail.

[0005] With the development of image processing technology, the extraction of flame range from videos or images using relevant image processing algorithms has become widely used. Compared to sensors, optical measurement equipment acquires clearer images and can more accurately obtain information such as the flame's position and size. Currently, various techniques exist for processing flame regions, including image subtraction, threshold segmentation, image binarization, and color space models. However, most of these techniques are designed for flames in conventional scenes and are not very applicable to flame detection under explosion conditions. Furthermore, the processed flame region often deviates significantly from the original flame range, and the flame edges are often coarsely processed. Summary of the Invention

[0006] The purpose of this invention is to provide a method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model, in order to solve the technical problem that the flame range extracted by existing technologies under explosion conditions has a large error compared with the original flame range, and the flame edge processing is relatively rough.

[0007] The concept of this invention is as follows: First, the YOLOv5 detection algorithm is used to pre-determine the location of the flame area in the explosion process. Then, the explosion flame range extraction method based on the YCbCr color space model is used to accurately process the flame range in the explosion process. Finally, the flame parameters are calculated according to the calibrated correspondence to achieve the purpose of accurately extracting the explosion flame range.

[0008] To achieve the above objectives and complete the above inventive concept, the present invention adopts the following technical solution:

[0009] A method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model is characterized by the following steps:

[0010] Step 1: Read the visible light image of the flame during the explosion process, use the YOLOv5 detection algorithm to initially determine the flame area, then calibrate the ratio coefficient between the pixel area of ​​the flame area in the visible light image and the projected area of ​​the flame area in the actual geographic space, and calibrate the conversion relationship between the centroid of the flame area in the visible light image and the centroid of the flame area in the actual geographic space.

[0011] Step 2: Convert the visible light image from the RGB color space to the YCbCr color space, and calculate the average value Y of the luminance component Y in the image. mean The average value of the blue chromaticity component Cb. mean ;

[0012] Step 3: Use the following four-criteria model to determine each pixel within the flame region initially identified in Step 1, select the pixels that fit the four-criteria model, and group them into a new flame region:

[0013]

[0014] Where Y(x,y) represents the value of the luminance component at the (x,y) pixel in the processed image, Cb(x,y) represents the value of the blue chromaticity component at the (x,y) pixel in the processed image, and Cr(x,y) represents the value of the red chromaticity component at the (x,y) pixel in the processed image.

[0015] Step 4: Process the new flame area to obtain the complete flame area, and then determine the pixels contained in the complete flame area;

[0016] Step 5: Calculate the pixel area and centroid of the complete flame region based on the pixel points determined in Step 4;

[0017] Step 6: Calculate the projected area and centroid of the flame region in the actual geographic space based on the pixel area and centroid calculated in Step 5, as well as the scaling factor and conversion relationship calibrated in Step 1, to complete the extraction of the explosion flame range.

[0018] Furthermore, step 4 specifically involves:

[0019] The new flame region is filled with holes to obtain a complete flame region, and then the number of pixels contained in the complete flame region is determined.

[0020] Furthermore, step 1 specifically includes:

[0021] The explosion process is recorded on video. The recorded video is converted into a sequence of images. The i-th visible light image of the flame in the sequence of images is read, where i is a natural number. The YOLOv5 detection algorithm is used to initially determine the flame area. Then, the ratio coefficient between the pixel area of ​​the flame area in the visible light image and the projected area of ​​the flame area in the actual geographic space is calibrated. The conversion relationship between the centroid of the flame area in the visible light image and the centroid of the flame area in the actual geographic space is also calibrated.

[0022] Step 6 specifically involves: calculating the projected area and centroid of the flame region in the actual geographic space based on the pixel area and centroid calculated in Step 5, as well as the scaling factor and conversion relationship calibrated in Step 1, to complete the extraction of the explosion flame range at the corresponding moment of the i-th image.

[0023] Furthermore, it also includes step 7:

[0024] Return to step 1 and extract the explosion flame range corresponding to the (i+1)th image at the specified time, until the extraction of the explosion flame range of all images in the sequence is completed, and obtain the evolution characteristics of the flame parameters.

[0025] Furthermore, in step 2, the visible light image is specifically converted from the RGB color space to the YCbCr color space using the following formula:

[0026]

[0027] Where R, G, and B represent the values ​​of the red, green, and blue channels in the image within the RGB color space, respectively.

[0028] The beneficial effects of this invention are:

[0029] 1. The explosion flame range extraction method based on YOLOv5 artificial intelligence recognition and matching and YCbCr color space model proposed in this invention has the advantages of convenient processing, accurate results and wide adaptability.

[0030] 2. This invention uses YOLOv5 artificial intelligence recognition and matching to initially select the flame area during the explosion process, effectively avoiding the influence of other objects in the image on the subsequent extraction of the flame range. It is simple to operate, highly efficient, and has strong anti-interference capabilities.

[0031] 3. This invention optimizes the existing YCbCr color space model based on the characteristics of flames under strong explosion conditions. The proposed 4-criteria model can clearly extract the outline range of the explosion flame with high accuracy and obvious preservation of edge features. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of an embodiment of the method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model according to the present invention. Detailed Implementation

[0033] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] This invention extracts the complete outline of the flame during the explosion process and calculates key flame parameters and their evolution characteristics by pre-determining the flame position, initially extracting the flame, and repairing voids.

[0035] This invention provides a method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model. Figure 1 As shown, the method for extracting the range of an explosion flame includes the following steps:

[0036] Step 1: Record the explosion process on video, convert the recorded video into a sequence of images, read the i-th visible light image of the flame at a certain moment in the sequence of images (i is a natural number), use the YOLOv5 detection algorithm to preliminarily determine the approximate flame area, then calibrate the ratio coefficient between the pixel area of ​​the flame area in the visible light image and the projected area of ​​the flame area in the actual geographic space, and calibrate the conversion relationship between the centroid of the flame area in the visible light image and the centroid of the flame area in the actual geographic space.

[0037] This embodiment can also process each image in a continuous video, mainly using the YOLOv5 detection algorithm to initially determine the location and approximate area of ​​the flame, reducing the computational area for subsequent flame range extraction, and effectively avoiding interference from other objects within the entire image area. This process requires a large, custom-defined training set, using publicly available or experimentally tested explosion flame images for training, to improve the accuracy of the initial flame determination.

[0038] Step 2: Convert the visible light image from the RGB color space to the YCbCr color space, and calculate the average value Y of the luminance component Y in the image. mean The average value of the blue chromaticity component Cb. mean ;

[0039] The conversion formula for an image from the RGB color space to the YCbCr color space is shown below:

[0040] Y = 0.299R + 0.587G + 0.114B

[0041] Cb=-0.1687R-0.3313G+0.5B+128

[0042] Cr = 0.5R - 0.4187G - 0.0812B + 128

[0043] In the RGB color space, R, G, and B represent the values ​​of the red, green, and blue channels in the image, respectively; in the YCbCr color space, Y, Cb, and Cr represent the values ​​of the luminance, blue chromaticity, and red chromaticity components in the image, respectively.

[0044] Step 3: Use the following four-criteria model to determine each pixel within the flame region initially identified in Step 1, select the pixels that fit the four-criteria model, and group them into a new flame region:

[0045] Y(x,y)>Cb(x,y)

[0046] Cr(x,y)>Cb(x,y)

[0047] Y(x,y)>Y mean

[0048] Cb(x,y) <Cb mean ;

[0049] Where Y(x,y) represents the value of the luminance component at the (x,y) pixel in the processed image, Cb(x,y) represents the value of the blue chromaticity component at the (x,y) pixel in the processed image, and Cr(x,y) represents the value of the red chromaticity component at the (x,y) pixel in the processed image.

[0050] In the process of flame range extraction, a key step is to extract the specific flame range within the approximate flame region obtained from the initial flame determination through model matching. In the model matching stage, the flame regions obtained using traditional techniques such as image subtraction, thresholding, and image binarization are not accurate or complete enough, and their edge processing is relatively coarse, easily overlooking edge features. Previous studies have used the YCbCr color space model to extract flames in complex scenes. However, processing images from multiple explosion scenarios using the YCbCr color space model presented in the reported literature revealed that the proposed flame detection model is not applicable to all explosion scenarios. The generation process of everyday flames is relatively mild and slow, while the chemical reaction in an explosion is extremely intense and rapid, and the distribution of red concentration shift in the flame during an explosion is extremely uneven. Using the rules described in the reported literature to filter out red concentration shifts below the average value easily erases less intense areas of the flame during an explosion. To improve the flame range extraction effect, the criterion of red concentration shifts above the average value given in the model was removed, and only the above four criteria were used to re-detect the flame. Extensive experimental verification revealed that the new four-criteria model can clearly define the outline of the flame, achieving better extraction results compared to models reported in previous literature. Furthermore, the extracted flame range closely matches the flame range in the original image, and the detailed features of the flame are fully preserved.

[0051] Step 4: Perform hole filling processing on the new flame area to obtain a complete flame area, and then determine the number of pixels contained in the complete flame area.

[0052] Step 5: Calculate the pixel area and centroid of the complete flame region based on the pixel points determined in Step 4;

[0053] Step 6: Calculate the projected area and centroid of the flame region in the actual geographic space based on the pixel area and centroid calculated in Step 5 and the scaling factor and conversion relationship determined in Step 1, and complete the extraction of the explosion flame range at the corresponding time of the i-th image.

[0054] Step 7: Return to step 1 and extract the explosion flame range corresponding to the (i+1)th image at the specified time, until the extraction of the explosion flame range of all images in the sequence is completed, and obtain the evolution characteristics of the flame parameters.

[0055] That is, for continuous video or image sequence, by performing the above steps 1-6 on each image according to the time sequence, the evolution characteristics of flame parameters over time can be obtained. This method is simple, fast and highly accurate.

[0056] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model, characterized in that, Includes the following steps: Step 1: Read the visible light image of the flame during the explosion process, use the YOLOv5 detection algorithm to initially determine the flame area, then calibrate the ratio coefficient between the pixel area of ​​the flame area in the visible light image and the projected area of ​​the flame area in the actual geographic space, and calibrate the conversion relationship between the centroid of the flame area in the visible light image and the centroid of the flame area in the actual geographic space. Step 2: Convert the visible light image from the RGB color space to the YCbCr color space, and calculate the average value Y of the luminance component Y in the image. mean The average value of the blue chromaticity component Cb. mean ; Step 3: Use the following four-criteria model to determine each pixel within the flame region initially identified in Step 1, select the pixels that fit the four-criteria model, and group them into a new flame region: Where Y(x,y) represents the value of the luminance component at the (x,y) pixel in the processed image, Cb(x,y) represents the value of the blue chromaticity component at the (x,y) pixel in the processed image, and Cr(x,y) represents the value of the red chromaticity component at the (x,y) pixel in the processed image. Step 4: Process the new flame area to obtain the complete flame area, and then determine the pixels contained in the complete flame area; Step 5: Calculate the pixel area and centroid of the complete flame region based on the pixel points determined in Step 4; Step 6: Calculate the projected area and centroid of the flame region in the actual geographic space based on the pixel area and centroid calculated in Step 5, as well as the scaling factor and conversion relationship calibrated in Step 1, to complete the extraction of the explosion flame range.

2. The method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model according to claim 1, characterized in that, Step 4 is as follows: The new flame region is filled with holes to obtain a complete flame region, and then the number of pixels contained in the complete flame region is determined.

3. The method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model according to claim 1 or 2, characterized in that, Step 1 is as follows: The explosion process is recorded on video. The recorded video is converted into a sequence of images. The i-th visible light image of the flame in the sequence of images is read, where i is a natural number. The YOLOv5 detection algorithm is used to initially determine the flame area. Then, the ratio coefficient between the pixel area of ​​the flame area in the visible light image and the projected area of ​​the flame area in the actual geographic space is calibrated. The conversion relationship between the centroid of the flame area in the visible light image and the centroid of the flame area in the actual geographic space is also calibrated. Step 6 specifically involves: calculating the projected area and centroid of the flame region in the actual geographic space based on the pixel area and centroid calculated in Step 5, as well as the scaling factor and conversion relationship calibrated in Step 1, to complete the extraction of the explosion flame range at the corresponding moment of the i-th image.

4. The method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model according to claim 3, characterized in that, It also includes step 7: Return to step 1 and extract the explosion flame range corresponding to the (i+1)th image at the specified time, until the extraction of the explosion flame range of all images in the sequence is completed, and obtain the evolution characteristics of the flame parameters.

5. The method for extracting the range of an explosion flame based on YOLOv5 artificial intelligence recognition and matching and the YCbCr color space model according to claim 4, characterized in that, In step 2, the visible light image is converted from the RGB color space to the YCbCr color space using the following formula: Where R, G, and B represent the values ​​of the red, green, and blue channels in the image within the RGB color space, respectively.

Images