Method for identifying flame characteristics in garbage incinerator grate based on image semantic segmentation

Abstract

The invention discloses a method for identifying flame characteristics in a garbage incinerator grate based on image semantic segmentation. The method comprises the following steps: acquiring a flame video of the garbage incinerator grate in a period of time through a high-temperature industrial camera in a hearth; frames are extracted from the flame video according to a certain interval, and a data set for a flame image semantic segmentation model is obtained after preprocessing; carrying out pixel-level manual annotation on the data set according to a flame region judgment standard; training a semantic segmentation deep learning network by using the labeled data set to obtain a semantic segmentation model; and the trained semantic segmentation model is applied to the garbage incinerator grate furnace to identify a flame region in the image, and fire line identification or preliminary judgment on some combustion conditions is carried out on the basis of the flame region. The method not only can be used for an intelligent combustion control system, but also can be used for preliminarily judging some combustion working conditions, and is beneficial to efficient and stable operation of the garbage incineration grate furnace.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control of waste incineration, and in particular to a method for identifying flame features in a waste incineration grate furnace based on image semantic segmentation, and in particular to a method for identifying flame features by performing pixel-level recognition on different areas of a flame image in a waste incineration grate furnace through image semantic segmentation technology. Background Art

[0002] By the end of 2023, the number of waste incinerators in operation in my country reached over 2,000, of which grate furnaces accounted for over 80%. To ensure stable and efficient operation of waste incinerators, furnace operators will focus on the flame characteristics within the furnace through flame images within the furnace, including the position of the fire line (flame front) and the combustion conditions characterized by the flame state. Human judgment mainly relies on observing flame images to determine flame characteristics, but compared with artificial intelligence judgment, manual judgment has low consistency, high lag, and weak anti-interference ability. Using artificial intelligence-related technologies to make judgments is more efficient, accurate, and stable.

[0003] Currently, research has been conducted on using furnace flame images to identify flame features. This research primarily utilizes traditional computer vision methods, such as grayscale conversion and binarization, to identify flame regions using grayscale thresholds. However, this method lacks differentiation between the different flame regions identified. Instead, it simply determines the gradient of grayscale values to extract bright and dark areas, identifying the bright areas as the main flame regions. Contour analysis algorithms are then used to extract the fire lines. While fast and simple, traditional computer vision-based methods are highly susceptible to external interference, often misidentifying non-flame areas as flame regions. Furthermore, the thresholds themselves require frequent adjustment. Consequently, the accuracy and stability of existing field methods are insufficient for long-term field use. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies in identifying flame features in waste incineration grates, this paper discloses a method for identifying flame features in waste incineration grates based on image semantic segmentation. Compared to existing flame feature recognition technologies based on traditional computer vision, this method offers advantages such as high accuracy, strong stability, and richer recognition information.

[0005] In order to identify fire lines and other flame features, the method of the present invention introduces image semantic segmentation technology based on deep learning, trains a semantic segmentation deep learning network through manually annotated data sets, and classifies flame images at the pixel level, thereby achieving rapid and accurate recognition of different areas of the flame image; at the same time, the fire line is identified according to the results of the semantic segmentation of the flame image (the fire line is located at the lower part of the outer contour of the main flame area), and other flame features are identified on this basis.

[0006] Based on the experience of experts and on-site boiler operators, the flame areas to be identified generally include: main flame area, tail flame area, background area, etc. The identification standards for each area are as follows:

[0007] Main flame area: refers to the main combustion area on the incineration grate, extending from the upper part of the combustion section to the end of the combustion section. The flame in this area basically covers the entire grate and is relatively bright and concentrated.

[0008] Tail fire area: refers to the residual flame below the main flame area on the incineration grate, located from the end of the combustion section to the burnout section. The flame in this area is unevenly dispersed, usually in sporadic or strip-like distribution, and is relatively dim.

[0009] Background area: furnace wall, burning section and other areas without flames.

[0010] The present invention achieves the purpose of pixel-level recognition of flame images in a garbage incineration grate furnace and identification of flame features through the following steps:

[0011] Firstly, a semantic segmentation model is used to perform pixel-level recognition of different areas in the furnace flame image. Then, the main flame area contour is determined based on the pixel-level recognition results. Based on the main flame area contour, the side contours and undercut contours are removed and the contour height is determined to identify the fire line. Finally, other flame features are identified based on the pixel-level recognition results and the fire line.

[0012] Preferably, the construction of the semantic segmentation model specifically includes the following steps:

[0013] S1: Dataset preparation stage: Obtain flame image video inside the waste incineration grate furnace;

[0014] S2: Dataset extraction stage: Flame image frames are extracted from the flame image video inside the garbage incineration grate furnace at a certain interval and preprocessed. The preprocessed flame image frames are randomly arranged to obtain a dataset;

[0015] S3: Training data set preparation stage: manually mark and divide the pixels of all flame image frames in the data set obtained in step S2 according to the flame area classification standard, and divide the divided data set into a training set and a validation set;

[0016] S4: semantic segmentation model training stage: using the training set to train the semantic segmentation network to obtain the semantic segmentation model; the input of the semantic segmentation model is the preprocessed flame image frame, and the output is the pixel-level recognition result of the flame area;

[0017] S5: Semantic segmentation model effect verification stage: Use the trained model to perform pixel-level recognition on the flame images of the verification set and calculate the accuracy of pixel-level recognition. If the accuracy meets the requirements, the semantic segmentation model is constructed.

[0018] Preferably, in step S2, flame image frames are extracted from the flame video at intervals of 20 seconds to 10 minutes, and the total number of flame image frames in the data set should be no less than 5000 frames.

[0019] Preferably, in step S3, the flame area is divided into a main flame area, a tail flame area, and a background area.

[0020] Preferably, in step S4, a pre-trained deep learning network including Deeplab v3+ network, Segnet network, U-net network, etc. can be selected as the backbone network of the semantic segmentation model.

[0021] Preferably, in step S5, the calculation formula of the accuracy rate P is: Among them, A i It represents the number of accurately identified pixels in the i-th image frame, n represents the total number of image frames, and Height and Width represent the pixel height and pixel width of each frame.

[0022] Using the trained semantic segmentation model, we can now identify the main flame region in a flame image at the pixel level. Furthermore, we can then quickly identify the fire line and other flame features using the following steps.

[0023] Preferably, the pixel-level recognition result is used to determine the main flame area contour, and based on the main flame area contour, the side contour and the undercut contour are removed and the contour height is determined in sequence, thereby identifying the fire line; the specific steps are as follows:

[0024] S01: Determine the outline of the main flame area: Based on the pixel-level recognition results of the furnace flame image, extract the main flame pixel area of the furnace flame image, binarize the main flame pixel area and perform connected domain analysis to obtain each connected domain of the main flame pixel area; use morphological expansion to eliminate holes in each connected domain, and designate the connected domain with the largest area as the main flame area, and then extract the outermost outline of the main flame area through connected domain outer contour recognition.

[0025] S02: Side contour removal: First, determine and cut off the left and right side ranges of the outermost contour of the main flame area: (minX, maxX). The range value is determined by the following formula: Where, X left and X right The leftmost and rightmost horizontal coordinates represent the outermost contour of the main flame area, respectively. and Represents the interception coefficient of the left and right sides, ranging from 0 to 0.5, with a representative value of 0.15. and The value can effectively remove the vertical rising parts on the left and right sides of the contour. max It is the maximum horizontal distance between the left and right sides of the outermost contour of the main flame area.

[0026] S03: Scoliosis contour removal: The remaining contour after the side contour is removed is divided into several continuous line segments through connected domain analysis. It is determined whether the leftmost and rightmost sides of each line segment reach minX and maxX at the same time. If not, it is a scoliosis contour segment and is removed.

[0027] S04: Determine the contour height: For the contour line segments after the undercut contour is removed, calculate the average ordinate value of each line segment. The positive direction of the ordinate is from top to bottom. The specific calculation formula is: Among them, y k Represents the average ordinate value of the kth line segment, n k Indicates the number of contour points of the kth line segment, y i Represents the ordinate value of the i-th point of the k-th line segment; the line segment with the largest average ordinate value is the desired fire line.

[0028] Preferably, if the main flame area accounts for too small a proportion or the background accounts for too large a proportion, it is judged to be an abnormal state (usually image blur caused by dust) and the fire line recognition is stopped; if the horizontal distance between the leftmost and rightmost sides of the outer contour of the main flame area is less than a certain threshold, the image is judged to be invalid and the fire line and other information are not recognized for this flame image.

[0029] Specifically, when the flame image in the furnace meets the following conditions at the same time, fire line recognition is performed:

[0030] R mainflame >T mainflame ; R background <T background

[0031] Δx max <a×Width

[0032] in,

[0033] Where R mainflame and R background Respectively represent the proportion of main flame pixels and background pixels in the current image; n mainflame 、n background Represents the number of main flame pixels and background pixels respectively, n total Represents the total number of image pixels; T mainflame and T backgroundRepresent the threshold of the proportion of main flame pixels and background pixels respectively; γ main Indicates the main flame ratio adjustment factor, ranging from 0 to 1, with a representative value of 0.8; γ back Represents the background ratio adjustment factor, ranging from 1 to 2, with a representative value of 1.2; n represents n images; R mainflame,i Represents the value of the main flame pixel ratio in the i-th image, R background,i Represents the percentage of background pixels in the i-th image; Width is the pixel width, a is the threshold, ranging from 0 to 1, and the representative value is 0.1.

[0034] Preferably, other flame characteristics include the distribution of the incineration grate, the biased burning situation, the uniformity of the combustion level, the total area, grayscale mean and coefficient of variation of the specified pixels in the target area, etc.

[0035] The method for identifying the distribution of incineration grates is as follows: some waste incineration grates, such as the Guangda grate, are fed from the top by several feeding trolleys arranged side by side. In order to determine whether the feeding speed or feeding stroke of each feeding trolley needs to be adjusted, the grate is divided into several vertical columns from left to right according to the number of feeding trolleys. The vertical coordinates of each point on the fire line in each column are added and averaged to obtain the average value of the vertical coordinates of the fire line in each column area. If the average value of the vertical coordinates of the fire line is in the burning section, it is judged that the feeding trolley stroke in the column is too large and the feeding speed is too fast. If the average value of the vertical coordinates of the fire line is in the drying section, it is judged that the feeding trolley stroke in the column is too small and the feeding speed is too slow.

[0036] The method for identifying biased burning is as follows: biased burning refers to the situation where the flame on the grate is obviously biased to one side. Whether the fire line is biased to one side can be used to make a preliminary judgment on the biased burning situation. Specifically, the vertical coordinate average value of the fire line on the left and right halves of the incineration grate is calculated using the identified fire line. If the difference between the two exceeds a certain threshold, it is judged that biased burning has occurred. The specific judgment formula is |Δy|=|y left -y right |>T partialburning where y left It represents the average value of the vertical coordinate of the fire line on the left half of the incinerator, y right It represents the average value of the vertical coordinate of the fire line on the right half of the incinerator grate. The left and right halves can be artificially determined. partialburning Indicates the threshold of partial burning, which is generally determined manually.

[0037] The method for identifying the uniformity of combustion level is as follows: The uniformity of combustion is generally related to whether the flame front on the grate is flush. The present invention reflects the uniformity of combustion here by calculating the coefficient of variation of the fire line. The specific formula is Where VC represents the coefficient of variation, σ represents the standard deviation of the vertical coordinate of the fire line pixels, and μ represents the average vertical coordinate of the fire line pixels. A larger coefficient of variation indicates a higher degree of dispersion of the fire line pixels and lower combustion uniformity.

[0038] The total area of the specified pixels in the target area is calculated as follows: In practical applications, it is sometimes necessary to pay attention to the combustion situation in a certain area. The semantic segmentation model of the present invention can be used to perform pixel-level recognition of any target area in the furnace flame image. Specifically, the total area of the specified pixels in the target area can be obtained by adding the areas of the pixels of the specified category in the target area. The specific formula is A k,c =∑a i,c , where A k,c represents the total area of pixels of category c in the k region, a i,c Represents the area of the i-th pixel in the specified category c in the k-region.

[0039] Grayscale mean and coefficient of variation of the specified pixels in the target area: The grayscale value of the specified category of pixels in the target area is calculated to obtain its mean and coefficient of variation. The specific formula for the grayscale mean of the specified pixels in the target area is: μ k,c =∑g i,c / n total,c , where μ k,c represents the average grayscale of the pixels of category c in the k region, g i,c Represents the grayscale value of the i-th pixel in the specified category c pixel of the k region; n total,c The total number of pixels of category c is specified for the k region; the coefficient of variation of the specified pixels in the target region is calculated as: Among them, VC k,c Represents the variation data of the grayscale of the pixel of the specified category c in the k region, σ k,c Represents the standard deviation of the grayscale values of pixels in a specified category c within region k. The flame intensity level and combustion uniformity can be determined by the grayscale mean and its coefficient of variation of the pixels in the specified category within the target region.

[0040] The beneficial effects of the present invention are:

[0041] The present invention processes the flame image inside the garbage incineration grate furnace to form a data set for classifying different flame areas. Based on the data set, a semantic segmentation model of the flame area inside the furnace is trained. Compared with traditional flame area recognition methods such as computer vision recognition methods, the model of the present invention can accurately and quickly perform pixel-level recognition of the flame area in the flame image. At the same time, the position of the fire line can also be identified through the recognition results, which is beneficial for boiler operators to judge key combustion conditions such as biased burning and combustion uniformity, thereby improving combustion stability and safety.

[0042] The method of the present invention has higher and wider recognition accuracy and identification information types of the flame area in the garbage incineration grate than traditional computer vision methods. The recognition results of the flame area can not only be used for the combustion intelligent control system, but also can make preliminary judgments on some combustion conditions, which is conducive to the efficient and stable operation of the garbage incineration grate. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 : Flowchart of a method for pixel-level recognition of different regions in flame images of a waste incineration grate using a semantic segmentation model.

[0044] Figure 2 : Flowchart of the method for identifying live wires.

[0045] Figure 3 : Schematic diagram of semantic segmentation of furnace flame image, where: 1: main flame area; 2: tail flame area; 3: background area; 4: flame outline.

[0046] Figure 4 : Schematic diagram of flame contour interception, where: 5: intercepted and removed flame contour (marked with red dotted line in the figure); 6: retained flame contour (marked with blue dotted line in the figure).

[0047] Figure 5 : Schematic diagram for judging the status of the feeding trolley, where: 7: the fire line in the area of the first row of feeding trolleys; 8: the fire line in the area of the second row of feeding trolleys; 9: the fire line in the area of the third row of feeding trolleys; 10: the fire line in the area of the fourth row of feeding trolleys.

[0048] Figure 6 : Schematic diagram for judging the partial burning condition, where: 11: left fire line; 12: right fire line. DETAILED DESCRIPTION

[0049] To make the purpose, technical solutions and advantages of the present invention clearer, the following describes the workflow and operation method of the present invention in detail with reference to specific embodiments, but the scope of protection of the present invention is not limited thereto:

[0050] Embodiment: A method for identifying flame features in a waste incineration grate furnace based on image semantic segmentation. The method comprises the following steps: first, a semantic segmentation model is used to perform pixel-level identification on different regions of the flame image in the furnace; then, the main flame area contour is determined using the pixel-level recognition results, and based on the main flame area contour, the side contours and undercut contours are removed and the contour height is determined in sequence to identify the fire line; finally, other flame features are identified based on the pixel-level recognition results and the fire line.

[0051] like Figure 1 As shown, based on the semantic segmentation model of the present invention, different areas in the waste incineration grate are identified in real time. The specific steps include:

[0052] S1: Dataset preparation phase: The flame image video data inside the waste incineration grate furnace is obtained over a period of time through on-site copying in the waste incineration plant electronic room.

[0053] S2: Dataset extraction stage: The obtained video is analyzed, and 8000 frames are extracted from the video at intervals of 2 minutes and preprocessed. Each frame is a JPG image with a resolution of 1920×1080. The extracted flame image frames are randomly arranged to obtain a dataset containing a total of 8000 flame image frames.

[0054] S3: Training data set preparation stage: For the data set obtained in step S2, use MATLAB's Image Labeler software or other similar software to mark the pixels of all flame image frames in the data set according to the flame area classification standard and divide them into three categories of pixels: main flame area, tail flame area and background area. The divided data set is divided into training set and validation set according to 8:2.

[0055] S4: Semantic segmentation model training phase: The Deeplab v3+ network was selected as the semantic segmentation backbone network and trained using the training set. It was then used to identify different areas in flame images within a waste incineration furnace. During training, the images in the dataset were compressed to a size of 224×224 to speed up training. Ultimately, the model achieved an accuracy of over 95% on the training set.

[0056] S5: Effect verification stage: The trained model is used to identify flame images in the verification set. Ultimately, the accuracy of the model on the verification set reaches over 90%, meeting the requirements for actual field deployment.

[0057] S6: Practical application effect verification stage: The trained model is applied to the real-time recognition of different areas in the actual waste incineration grate to observe the accuracy and stability of the semantic segmentation model.

[0058] like Figure 3 As shown, the identified flame area generally includes: main flame area 1, tail flame area 2, and background area 3. 4 is the flame outline of the main flame area 1.

[0059] The pixel-level recognition result is used to determine the main flame area contour, and based on the main flame area contour, the side contour and the undercut contour are removed and the contour height is determined in sequence, thereby identifying the fire line; Figure 2 The specific steps are as follows:

[0060] S01: Determine the outline of the main flame area: Based on the pixel-level recognition results of the furnace flame image, extract the main flame pixel area of the furnace flame image, binarize the main flame pixel area and perform connected domain analysis to obtain each connected domain of the main flame pixel area; use morphological expansion to eliminate holes in each connected domain, and designate the connected domain with the largest area as the main flame area, and then extract the outermost outline of the main flame area through connected domain outer contour recognition.

[0061] S02: Side contour removal ( Figure 4 ): First, determine and intercept the left and right side ranges of the outermost contour of the main flame area: (minX, maxX), and the range value is determined by the following formula: Where, X left and X right The leftmost and rightmost horizontal coordinates represent the outermost contour of the main flame area, respectively. and Represents the interception coefficient of the left and right sides, ranging from 0 to 0.5, with a representative value of 0.15. and The value can effectively remove the vertical rising parts on the left and right sides of the contour. max It is the maximum horizontal distance between the left and right sides of the outermost contour of the main flame area.

[0062] S03: Scoliosis contour removal: The remaining contour after the side contour is removed is divided into several continuous line segments through connected domain analysis. It is determined whether the leftmost and rightmost sides of each line segment reach minX and maxX at the same time. If not, it is a scoliosis contour segment and is removed.

[0063] S04: Determine the contour height: For the contour line segments after the undercut contour is removed, calculate the average ordinate value of each line segment. The positive direction of the ordinate is from top to bottom. The specific calculation formula is: Among them, y k Represents the average ordinate value of the kth line segment, n k Indicates the number of contour points of the kth line segment, y i Represents the ordinate value of the i-th point of the k-th line segment; the line segment with the largest average ordinate value is the desired fire line.

[0064] The flame characteristics include the distribution of the incineration grate, the biased burning situation, the uniformity of the combustion level, the total area of the specified pixels in the target area, the grayscale mean and the coefficient of variation, etc.

[0065] The identification method of the incineration grate distribution situation is as follows: the grate of some garbage incineration grates, such as the Guangda grate, is fed by several feeding trolleys (such as Figure 5) feed the grate from the top side by side. To determine whether the feeding speed or feeding stroke of each feeding trolley needs to be adjusted, the grate is evenly divided into several vertical columns from left to right according to the number of feeding trolleys. The vertical coordinates of each point on the fire line of each column are added and averaged to obtain the average value of the vertical coordinate of the fire line in each column area. If the average value of the vertical coordinate of the fire line is in the burning section, it is judged that the feeding trolley stroke of the column is too large and the feeding speed is too fast. If the average value of the vertical coordinate of the fire line is in the drying section, it is judged that the feeding trolley stroke of the column is too small and the feeding speed is too slow.

[0066] Partial burning (such as Figure 6 ) is identified as follows:

[0067] The identified fire lines are used to calculate the average vertical coordinates of the fire lines on the left and right sides of the incinerator. If the difference between the two exceeds a certain threshold, it is determined that partial burning occurs. The specific judgment formula for partial burning is:

[0068] |Δy|=|y left -y right |>T partialburning

[0069] Among them, y left It represents the average value of the vertical coordinate of the fire line on the left half of the incinerator, y right It represents the average value of the vertical coordinate of the fire line on the right half of the incineration grate, T partialburning Indicates the threshold of partial burning, which is generally determined manually.

Claims

1. A method for identifying flame features in a garbage incineration grate based on image semantic segmentation, characterized in that: Firstly, a semantic segmentation model is used to perform pixel-level recognition of different areas in the furnace flame image. Then, the main flame area contour is determined based on the pixel-level recognition results. Based on the main flame area contour, the side contours and undercut contours are removed and the contour height is determined to identify the fire line. Finally, other flame features are identified based on the pixel-level recognition results and the fire line.

2. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 1 is characterized in that: The construction of the semantic segmentation model specifically includes the following steps: S1: Dataset preparation stage: Obtain flame image video inside the waste incineration grate furnace; S2: Dataset extraction stage: Flame image frames are extracted from the flame image video inside the garbage incineration grate furnace at a certain interval and preprocessed. The preprocessed flame image frames are randomly arranged to obtain a dataset; S3: Training data set preparation stage: manually mark and divide the pixels of all flame image frames in the data set obtained in step S2 according to the flame area classification standard, and divide the divided data set into a training set and a validation set; S4: semantic segmentation model training stage: using the training set to train the semantic segmentation network to obtain the semantic segmentation model; the input of the semantic segmentation model is the preprocessed flame image frame, and the output is the pixel-level recognition result of the flame area; S5: Semantic segmentation model effect verification stage: Use the trained model to perform pixel-level recognition on the flame images of the verification set and calculate the accuracy of pixel-level recognition. If the accuracy meets the requirements, the semantic segmentation model is constructed.

3. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 2 is characterized in that: In step S2, the total number of flame image frames in the data set is not less than 5000 frames.

4. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 2 is characterized in that: In step S4, deep learning pre-trained networks including Deeplab v3+ network, Segnet network, and U-net network can be selected as the backbone network of the semantic segmentation model.

5. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 2 is characterized in that: In step S5, the calculation formula of the accuracy rate P is: Among them, A i It represents the number of accurately identified pixels in the i-th image frame, n represents the total number of image frames, and Height and Width represent the pixel height and pixel width of each frame.

6. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 1, characterized in that: The pixel-level recognition result is used to determine the main flame area contour, and based on the main flame area contour, the side contour and the undercut contour are removed and the contour height is determined in sequence, thereby identifying the fire line. The specific steps are as follows: S01: Determining the main flame connected domain outline: Based on the pixel-level recognition result of the furnace flame image, extracting the main flame pixel area of the furnace flame image, binarizing the main flame pixel area and performing connected domain analysis to obtain each connected domain of the main flame pixel area; The holes in each connected domain are eliminated by morphological expansion, and the connected domain with the largest area is designated as the main flame region. Then, the outer contour of the main flame region is extracted by identifying the outer contour of the connected domain. S02: Side contour removal: First, the left and right side ranges of the outermost contour of the main flame area are determined and intercepted: (minX, maxX). The range value is determined by the following formula: Where, X left and X right The leftmost and rightmost abscissas represent the outermost contour of the main flame area respectively; and Respectively represent the interception coefficients of the left and right sides, ranging from 0 to 0.5; Δx max The maximum horizontal distance between the left and right sides of the outermost contour of the main flame area; S03: Scoliosis contour removal: The remaining contour after the side contour removal is divided into several continuous line segments through connected domain analysis. It is determined whether the leftmost and rightmost sides of each line segment reach minX and maxX at the same time. If not, it is a scoliosis contour segment and is removed. S04: Determine the contour height: For the contour line segments after the undercut contour is removed, calculate the average ordinate value of each line segment. The positive direction of the ordinate is from top to bottom. The specific calculation formula is: Among them, y k Represents the average ordinate value of the kth line segment, n k Indicates the number of contour points of the kth line segment, y i Represents the ordinate value of the i-th point of the k-th line segment; the line segment with the largest average ordinate value is the desired fire line.

7. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 6, characterized in that: If the flame image in the furnace meets the following conditions, fire line recognition is performed: R mainflame >T mainflame ;R background <T background Δx max <a×Width in Where R mainflame and R background Respectively represent the proportion of main flame pixels and background pixels in the current image; n mainflame 、n background Represents the number of main flame pixels and background pixels respectively, n total Represents the total number of image pixels; T mainflame and T background Represent the threshold of the proportion of main flame pixels and background pixels respectively; γ main Indicates the main flame ratio adjustment factor, ranging from 0 to 1; γ back Represents the background ratio adjustment factor, ranging from 1 to 2; n represents n images, R mainflame,i Represents the value of the main flame pixel ratio in the i-th image, R background,i Represents the value of the background pixel ratio in the i-th image; Width is the pixel width, a is the threshold value, ranging from 0 to 1; If the above conditions are not met, no live wire identification will be performed.

8. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 6 is characterized in that: The other flame characteristics specifically include the distribution of the incineration grate, the biased burning situation, the uniformity of the combustion level, the total area, grayscale mean and coefficient of variation of the specified pixels in the target area.

9. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 8, characterized in that: The method for identifying the distribution condition of the incineration grate is as follows: the incineration grate is evenly divided into a number of vertical columns from left to right according to the number of feeding trolleys, the vertical coordinates of each point of the fire line of each vertical column are added and averaged to obtain the average value of the vertical coordinates of the fire line in each vertical column area; if the average value of the vertical coordinates of the fire line is in the burning section, it is judged that the feeding trolley of the vertical column has too much travel and the feeding speed is too fast; if the average value of the vertical coordinates of the fire line is in the drying section, it is judged that the feeding trolley of the vertical column has too little travel and the feeding speed is too slow; The method for identifying the partial burning condition is as follows: using the identified fire line, the average value of the vertical coordinates of the fire line on the left and right halves of the incineration grate is calculated respectively. If the difference between the two exceeds a certain threshold, it is determined that the partial burning condition has occurred. The specific judgment formula for the partial burning condition is: |Δy|=|y left -y right |>T partialburning Among them, y keft It represents the average value of the vertical coordinate of the fire line on the left half of the incinerator, y right It represents the average value of the vertical coordinate of the fire line on the right half of the incineration grate, T partialburning Indicates the threshold value of partial burning; The method for identifying the combustion level uniformity is to reflect the combustion level uniformity by calculating the coefficient of variation of the fire line. The specific formula is: Wherein, VC represents the coefficient of variation, σ represents the standard deviation of the vertical coordinate of the fire line pixel points, and μ represents the average vertical coordinate of the fire line pixel points.

10. The method for identifying flame features in a garbage incineration grate based on image semantic segmentation according to claim 8, characterized in that: The total area of the designated pixels in the target area is calculated as follows: the total area of the designated pixels in the target area is obtained by adding the areas of the pixels of the designated category in the target area. The specific formula is: A k,c =∑a i,c Among them, A k,c represents the total area of pixels of category c in the k region, a i,c represents the area of the i-th pixel in the specified category c in the k-region; The grayscale mean value of the specified pixel in the target area is calculated as follows: m k,c =∑g i,c / n total,c Among them, μ k,c represents the average grayscale of the pixels of category c in the k region, g i,c Represents the grayscale value of the i-th pixel in the specified category c pixel of the k region; n total,c Assign the total number of pixels of category c to region k; The coefficient of variation of a specified pixel in the target area is calculated as follows: Among them, VC k,c Represents the variation data of the grayscale of the pixel of the specified category c in the k region, σ k,c Represents the standard deviation of the grayscale of pixels of the specified category c in region k.

Images