Method, device, storage medium and computer equipment for determining spherical flame shape

The method improves spherical flame morphology determination by using image processing and a flame crack prediction model to automate the process, addressing subjectivity and overlap errors, enhancing accuracy and efficiency.

CN118918175BActive Publication Date: 2024-12-03NORTHEASTERN UNIV CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411400595.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-10-09
Publication Date
2024-12-03
Estimated Expiration
2044-10-09

AI Technical Summary

Technical Problem

Existing methods for determining spherical flame morphology are subjective and inefficient, leading to low accuracy due to manual threshold assignment and cellular structure overlap errors.

Method used

A method involving image processing and a flame crack prediction model is used to determine the actual flame radius and crack length, avoiding cellular structure overlap by using a local area image to predict the flame crack image, thereby improving accuracy and efficiency.

Benefits of technology

The method enhances the determination efficiency and accuracy of spherical flame morphology by automating the process and eliminating errors caused by cellular structure overlap, expanding the characterization range to include complex turbulent flames.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918175B_ABST
    Figure CN118918175B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device, storage medium and computer equipment for determining the spherical flame morphology, and relates to the technical field of flame morphology measurement, and mainly lies in being able to improve the accuracy and efficiency of determining the spherical flame morphology. The method comprises: in response to a flame morphology determination signal of a measured spherical flame, collecting a flame picture of the measured spherical flame; based on the flame picture, determining the actual flame radius of the measured spherical flame, and based on the actual flame radius, determining in the flame picture a local area image where there is no overlapping cellular structure of the measured spherical flame; inputting the local area image into a preset flame crack prediction model for crack prediction, and obtaining a flame crack image in the local area image; based on the flame crack image, determining the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determining the flame morphology of the measured spherical flame. The present invention is applied to flame morphology recognition scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of flame shape measurement, and in particular to a method, device, storage medium and computer equipment for determining a spherical flame shape. Background Art

[0002] Spherical flame is a classic flame mode of internal combustion engine combustion and premixed gas explosion. The evolution of the flame front morphology during its propagation is the dominant factor leading to flame acceleration. Therefore, quantitative characterization of spherical flame morphology is a key technology to guide internal combustion engine combustion organization and gas explosion risk assessment.

[0003] At present, the existing spherical flame morphology quantitative characterization technology is mainly based on the artificial threshold method, that is, the binary threshold is manually assigned according to the flame image. This method is highly subjective, resulting in low accuracy in determining the flame morphology. At the same time, the manual assignment method requires staff to consult relevant information in advance, resulting in low efficiency in determining the flame morphology. Summary of the invention

[0004] The present invention provides a method, device, storage medium and computer equipment for determining a spherical flame shape, which are mainly capable of improving the accuracy and efficiency of determining the spherical flame shape.

[0005] According to a first aspect of the present invention, there is provided a method for determining a spherical flame shape, comprising:

[0006] In response to a flame shape determination signal of the measured spherical flame, collecting a flame picture of the measured spherical flame;

[0007] Based on the flame image, determining an actual flame radius of the measured spherical flame, and based on the actual flame radius, determining in the flame image an image of a local area where no cellular structure overlaps with the measured spherical flame;

[0008] Inputting the local area image into a preset flame crack prediction model to perform crack prediction, thereby obtaining a flame crack image in the local area image;

[0009] Based on the flame crack image, the actual flame crack length of the measured spherical flame is determined, and based on the actual flame crack length and the actual flame radius, the flame morphology of the measured spherical flame is determined.

[0010] Optionally, determining the actual flame radius of the measured spherical flame based on the flame image includes:

[0011] Removing a background image not containing flames from the flame image to obtain a background-removed flame image;

[0012] Performing grayscale processing and binarization processing on the background-removed flame image to obtain a binarized flame image corresponding to the background-removed flame image;

[0013] Fitting the edge pixel points of the binary flame image to obtain an edge contour curve of the measured spherical flame in the binary flame image;

[0014] Based on the edge contour curve and the scaling ratio of the image acquisition device corresponding to the flame picture, the actual flame radius of the measured spherical flame is determined.

[0015] Optionally, the determining, based on the actual flame radius, in the flame image, a local area image where no cellular structure overlaps with the measured spherical flame, comprises:

[0016] Determining the radius of the intercepted area corresponding to the measured spherical flame based on the actual flame radius and the preset radius interception coefficient;

[0017] A spherical direction in which no cellular structure overlaps is determined in the measured spherical flame, and an arc-shaped flame region in which no cellular structure overlaps is determined in the measured spherical flame based on the spherical direction and the radius of the intercepted region, and a local area image corresponding to the arc-shaped flame region is determined in the flame image.

[0018] Optionally, before inputting the local area image into a preset flame crack prediction model to perform crack prediction and obtaining a flame crack image in the local area image, the method further includes:

[0019] constructing at least one preset initial flame crack prediction model;

[0020] Acquire a sample picture set, wherein the sample picture set includes a plurality of sample flame pictures with annotation information, the annotation information includes an annotation box and an annotation type of an object in the annotation box, and the annotation type includes flame cracks;

[0021] Based on the number of models of the preset initial flame crack prediction model, the sample image set is divided into multiple training sets and multiple test sets, each training set is used to train the corresponding preset initial flame crack prediction model, and each test set is used to test the preset initial flame crack prediction model after the corresponding training, and the preset initial flame crack prediction model that meets the test conditions is used as the preset flame crack prediction model.

[0022] Optionally, determining the actual flame crack length of the measured spherical flame based on the flame crack image includes:

[0023] Determine the pixel value of the reference pixel point in the preset neighborhood corresponding to each pixel point in the flame crack image, and sort the reference pixel points in the preset neighborhood based on the pixel value of the reference pixel point in the preset neighborhood to obtain each sorted pixel point in the preset neighborhood;

[0024] Determine a median pixel point among the sorted pixel points in the preset neighborhood, and use the median pixel point to replace each corresponding pixel point in the flame crack image to obtain a denoised flame crack image;

[0025] Determine a grayscale histogram H(x, y) corresponding to the denoised flame crack image, and based on the grayscale histogram H(x, y), determine a target evaluation crack image in the denoised flame crack image, and determine a binary crack image corresponding to the target evaluation crack image;

[0026] The binary crack image is skeletonized by using a preset skeleton algorithm to obtain a flame skeleton image with a single pixel width;

[0027] Based on the flame skeleton image, determining a reference flame crack length of the arc-shaped flame region corresponding to the local region image;

[0028] The actual flame crack length of the measured spherical flame is determined based on the flame radius of the arc-shaped flame area, the actual flame radius of the measured spherical flame, and the reference flame crack length.

[0029] Optionally, the grayscale histogram H(x, y) includes the grayscale histogram corresponding to each pixel in the denoised flame crack image, and any grayscale histogram is used to represent the grayscale value of each pixel and the number of pixels corresponding to each grayscale value;

[0030] The step of determining a target evaluation crack image in the denoised flame crack image based on the grayscale histogram H(x, y) includes:

[0031] Obtaining a peak value P0 of the number of pixels corresponding to each grayscale histogram in the grayscale histogram H(x, y), and calculating a first evaluation value corresponding to the grayscale histogram according to the peak value P0 of the number of pixels and a preset coefficient a, wherein the first evaluation value=a×P0;

[0032] Draw a first evaluation horizontal line L0 corresponding to the first evaluation value, and obtain two grayscale histograms X that intersect with the first evaluation horizontal line L0 and are closest to each other. l and X u , where X l Less than X u , the first evaluation level L0 is y=a×P0;

[0033] Set the gray value less than X l A first grayscale histogram is linearly fitted to obtain a first reference horizontal line y=P1, a first standard deviation S1 of the first grayscale histogram is calculated, a second evaluation horizontal line L1 is drawn according to the first reference horizontal line and the first standard deviation S1, and a grayscale histogram T1 in the first grayscale histogram intersecting with the second evaluation horizontal line L1 is obtained, wherein the second evaluation horizontal line L1 is y=P1+2×S1;

[0034] Set the gray value greater than X u A second grayscale histogram is linearly fitted to obtain a second reference horizontal line y=P2, a second standard deviation S2 of the second grayscale histogram is calculated, a third evaluation horizontal line L2 is drawn according to the second reference horizontal line and the second standard deviation S2, and a grayscale histogram T in the second grayscale histogram intersecting with the third evaluation horizontal line L2 is obtained. u , wherein the third evaluation level L2 is y=P2+2×S2;

[0035] Segment the denoised flame crack image (T1, T u ) is used as the target to evaluate the crack image.

[0036] Optionally, the flame morphology includes an average radius of flame cells and a total number of flame cells of the measured spherical flame;

[0037] The step of determining the flame shape of the measured spherical flame based on the actual flame crack length and the actual flame radius includes:

[0038] Based on the actual flame crack length , the actual flame radius , determine the average radius of the flame cell of the measured spherical flame ,in, ;

[0039] Based on the average radius of the flame cell , the actual flame radius , determine the total number of flame cells of the measured spherical flame ,in, .

[0040] According to a second aspect of the present invention, there is provided a device for determining a spherical flame shape, comprising:

[0041] A collection unit, configured to collect a flame image of the spherical flame under test in response to a flame shape determination signal of the spherical flame under test;

[0042] A first determining unit is used to determine an actual flame radius of the measured spherical flame based on the flame picture, and determine, based on the actual flame radius, a local area image in which no cellular structure overlaps with the measured spherical flame in the flame picture;

[0043] A prediction unit, used for inputting the local area image into a preset flame crack prediction model to perform crack prediction, and obtaining a flame crack image in the local area image;

[0044] The second determination unit is used to determine the actual flame crack length of the measured spherical flame based on the flame crack image, and determine the flame shape of the measured spherical flame based on the actual flame crack length and the actual flame radius.

[0045] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above method for determining the spherical flame morphology.

[0046] According to a fourth aspect of the present invention, there is provided a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method for determining the spherical flame morphology when executing the program.

[0047] According to a method, device, storage medium and computer equipment for determining a spherical flame morphology provided by the present invention, compared with the current method of determining flame morphology based on an artificial threshold method, the present invention collects a flame image of the measured spherical flame in response to a flame morphology determination signal of the measured spherical flame; and based on the flame image, determines the actual flame radius of the measured spherical flame, and based on the actual flame radius, determines in the flame image a local area image where there is no cellular structure overlap of the measured spherical flame; then inputs the local area image into a preset flame crack prediction model for crack prediction, and obtains a flame crack image in the local area image; finally, based on the flame crack image, determines the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determines the flame morphology of the measured spherical flame. Therefore, by determining the local area image without cellular structure overlap in the flame picture corresponding to the measured spherical flame, and using the model to predict the flame crack image in the local area image, and finally determining the flame morphology of the measured spherical flame according to the flame crack image, thereby using the model to predict the flame crack image, and finally automatically determining the flame morphology according to the flame crack image, it is possible to avoid the low determination efficiency and determination errors caused by the artificial threshold method for determining the flame morphology, that is, the present invention can improve the determination efficiency and determination accuracy of the flame morphology. At the same time, the present invention uses the local area image without cellular structure overlap to predict the morphology of the entire spherical flame, which can avoid the morphology prediction error caused by the cellular structure overlap at the edge of the spherical flame, so that the present invention can further improve the determination accuracy of the spherical flame morphology. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0049] Figure 1 A flow chart of a method for determining a spherical flame shape provided by an embodiment of the present invention is shown;

[0050] Figure 2 A flow chart of another method for determining a spherical flame shape provided by an embodiment of the present invention is shown;

[0051] Figure 3 A schematic diagram of a grayscale histogram provided by an embodiment of the present invention is shown;

[0052] Figure 4 A schematic structural diagram of a device for determining a spherical flame shape provided by an embodiment of the present invention is shown;

[0053] Figure 5 A schematic structural diagram of another device for determining a spherical flame shape provided by an embodiment of the present invention is shown;

[0054] Figure 6 A schematic diagram of the physical structure of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0056] At present, the method of determining flame morphology based on the manual threshold method is highly subjective, resulting in low accuracy in determining the flame morphology. At the same time, the manual assignment method requires staff to consult relevant information in advance, resulting in low efficiency in determining the flame morphology.

[0057] In order to solve the above problems, the embodiment of the present invention provides a method for determining the spherical flame shape, such as Figure 1 As shown, the method includes:

[0058] 101. In response to a flame shape determination signal of a measured spherical flame, a flame image of the measured spherical flame is collected.

[0059] Among them, the spherical flame to be measured can be a spherical flame generated by the combustion of an internal combustion engine, or a flame generated by the explosion of a premixed gas. The embodiment of the present invention does not specifically limit the generation method of the spherical flame to be measured; the flame morphology includes the average radius of the flame cells and the total number of flame cells of the spherical flame to be measured; the spherical flame to be measured can be a laminar flame, a turbulent flame, etc.

[0060] For an embodiment of the present invention, for example, when an internal combustion engine burns to produce a spherical flame, or a constant volume combustion bomb produces a spherical flame, the spherical flame is determined as the spherical flame to be measured, and an image acquisition device such as a high-speed camera is used to photograph the spherical flame produced by the internal combustion engine combustion, or the hydrogen flame propagation process in the constant volume combustion bomb, to obtain a flame picture of the spherical flame to be measured, and then a local area image of the spherical flame to be measured without overlapping cellular structures is determined in the flame picture, and the cellular crack structure in the local area image is determined, and then converted to the entire spherical flame according to the arc surface formula, and the cellular structure size and the number of cells of the entire spherical flame are calculated. The embodiment of the present invention solves the problems of strong subjectivity, poor flexibility and low recognition accuracy in the existing artificial threshold method for quantitatively characterizing the morphology of spherical flames, and expands the existing quantitative characterization range of spherical flame morphology from laminar flames to turbulent flames that are closer to reality, thereby improving the determination efficiency and accuracy of the spherical flame morphology.

[0061] 102. Based on the flame image, determine the actual flame radius of the measured spherical flame, and based on the actual flame radius, determine in the flame image a local area image where the measured spherical flame does not have overlapping cellular structures.

[0062] Among them, during the propagation of the spherical flame, when the combustion conditions (such as fuel concentration, pressure, temperature, etc.) reach a certain threshold, the flame surface may gradually split into many small "cell"-like structures, which are called cellular flames. These "cells" usually have a convex surface, which leads to an increase in the flame area, thereby accelerating the overall propagation speed of the flame. As the flame continues to propagate and expand, especially when the flame radius gradually increases, these cellular structures may overlap in space, that is, different cellular flame units are intertwined and overlapped in three-dimensional space.

[0063] For the embodiment of the present invention, since the overlap of cellular structures will affect the accuracy of determining the morphology of the spherical flame, the embodiment of the present invention first needs to determine the local area image where there is no overlap of cellular structures in the spherical flame, for example, using a self-written program in MATLAB (Matrix Laboratory, numerical calculation and visualization software) to process the flame image, after obtaining the actual flame radius, cut out the area with a radius of half the original circle radius (actual flame radius) as the local area where there is no overlap of cellular structures, for example, in the flame image, create a binary mask image according to the fitted flame contour curve (circle), where the pixels inside the circle are marked as true (or white). Use this mask to operate the original image (flame image), extract the area with a radius of half the original circle radius, that is, the local area image where there is no overlap of cellular structures. In order to avoid the influence of overlapping cellular structures at the edge of the flame, the flame shape of the entire spherical flame being measured is then determined based on the flame cracks in the local area image, thereby determining the shape of the entire spherical flame through the local area image where there is no overlapping cellular structure. This can avoid the error in the determination of the flame shape caused by the overlapping cellular structures on the surface of the spherical flame, so that the embodiment of the present invention can improve the accuracy of determining the spherical flame shape.

[0064] 103. Input the local area image into a preset flame crack prediction model to perform crack prediction, and obtain a flame crack image in the local area image.

[0065] Among them, the preset flame crack prediction model can specifically be a model built based on a neural network. For an embodiment of the present invention, the preset flame crack prediction model includes multiple feature extraction layers. First, the local area image is input into the preset flame crack prediction model. For example, each layer of the preset flame crack prediction model includes two 3×3 convolution layers, each convolution layer uses a ReLU (Rectified Linear Unit, an activation function widely used in neural networks) activation function, and there is a maximum pooling layer with a step size of 2 between layers. After each pooling, the resolution of the image will be halved, the number of channels between layers will be doubled, and the number of channels between each convolution layer will not change. The network parameters are initialized first, and the network parameters are initialized to obey the Gaussian distribution. The coding path replaces the pooling layer in the conventional convolutional neural network with a 2×2 upsampling layer. In the fully convolutional neural network, upsampling is also called deconvolution. The size of the feature map output by each deconvolution will be doubled, and the number of channels will be halved. Then it is spliced ​​and fused with a feature map of a corresponding coding path, and finally classified using a 1×1 convolution layer. Finally, the detection result image, i.e., the flame crack image, is output. Then, the flame shape of the entire spherical flame is determined based on the flame crack image of the area where the cellular structure does not overlap in the spherical flame, which can avoid the influence of the overlapping cellular structure of the flame edge on the flame shape determination. Therefore, the embodiment of the present invention can improve the accuracy of the determination of the flame shape.

[0066] 104. Based on the flame crack image, determine the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determine the flame shape of the measured spherical flame.

[0067] According to the embodiment of the present invention, the flame crack image is skeletonized to obtain the image crack length in the flame crack image, and then the actual local flame crack length in the local area where there is no cellular structure overlap in the measured spherical flame can be calculated according to the pixel size of the flame crack image and the image scaling ratio of the image acquisition device, and the actual flame crack length of the entire measured spherical flame can be calculated according to the actual local flame crack length, and finally the flame morphology of the measured spherical flame can be determined according to the actual flame crack length and the actual flame radius, thereby accurately identifying the cellular crack structure in the area where there is no cellular structure overlap in the entire flame through the preset flame crack prediction model, and then converting to the entire sphere according to the arc surface formula, the cellular structure size and the number of cells of the entire spherical flame are calculated. The embodiment of the present invention solves the problems of strong subjectivity, poor flexibility and low recognition accuracy of the existing artificial threshold method for quantitatively characterizing the spherical flame morphology, and at the same time, the embodiment of the present invention avoids the error caused by the overlap of cellular structures, thereby improving the accuracy of determining the spherical flame morphology.

[0068] According to a method for determining a spherical flame morphology provided by the present invention, compared with the current method of determining the flame morphology based on an artificial threshold method, the present invention collects a flame image of the measured spherical flame in response to a flame morphology determination signal of the measured spherical flame; and based on the flame image, determines the actual flame radius of the measured spherical flame, and based on the actual flame radius, determines in the flame image a local area image where there is no cellular structure overlap of the measured spherical flame; then inputs the local area image into a preset flame crack prediction model for crack prediction, and obtains a flame crack image in the local area image; finally, based on the flame crack image, determines the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determines the flame morphology of the measured spherical flame. Therefore, by determining the local area image without cellular structure overlap in the flame picture corresponding to the measured spherical flame, and using the model to predict the flame crack image in the local area image, and finally determining the flame morphology of the measured spherical flame according to the flame crack image, thereby using the model to predict the flame crack image, and finally automatically determining the flame morphology according to the flame crack image, it is possible to avoid the low determination efficiency and determination errors caused by the artificial threshold method for determining the flame morphology, that is, the present invention can improve the determination efficiency and determination accuracy of the flame morphology. At the same time, the present invention uses the local area image without cellular structure overlap to predict the morphology of the entire spherical flame, which can avoid the morphology prediction error caused by the cellular structure overlap at the edge of the spherical flame, so that the present invention can further improve the determination accuracy of the spherical flame morphology.

[0069] Further, in order to better illustrate the process of determining the spherical flame shape, as a refinement and extension of the above embodiment, the embodiment of the present invention provides another method for determining the spherical flame shape, such as Figure 2 As shown, the method includes:

[0070] 201. In response to a flame shape determination signal of a measured spherical flame, collect a flame image of the measured spherical flame.

[0071] Specifically, when a spherical flame is generated by combustion of an internal combustion engine or explosion of a premixed gas, an image acquisition device is used to acquire a picture of the spherical flame during its propagation, namely, a flame picture.

[0072] 202. Based on the flame image, determine the actual flame radius of the measured spherical flame.

[0073] For the embodiment of the present invention, in order to determine the local area image without cellular structure overlap in the entire spherical flame to be measured, it is first necessary to determine the actual flame radius of the spherical flame to be measured. Based on this, step 202 specifically includes: removing the background image that does not contain the flame in the flame image to obtain a background-removed flame image; performing grayscale processing and binarization processing on the background-removed flame image to obtain a binary flame image corresponding to the background-removed flame image; fitting the edge pixel points of the binary flame image to obtain an edge contour curve of the spherical flame to be measured in the binary flame image; and determining the actual flame radius of the spherical flame to be measured based on the edge contour curve and the scaling ratio of the image acquisition device corresponding to the flame image.

[0074] Specifically, after collecting the flame image of the spherical flame to be measured, the background image is first removed from the flame image, and the image after the background image is removed is grayed to obtain a grayscale image. Then, a suitable threshold is selected to perform binarization on the grayscale image to obtain a binary flame image with only black and white pixels. Further, the boundary of the binary flame image is extracted and an algorithm is used to find the outermost boundary point pixels of the image, and the general equation of a circle is used, such as (in, , , are constants, is the horizontal coordinate of any point in the circle, is the ordinate of any point in the circle), and the outer boundary point pixels are fitted to obtain the edge contour curve of the flame in the binary flame image. Then, the edge contour curve is scaled according to the scaling ratio of the image acquisition device to obtain the actual flame radius of the measured spherical flame.

[0075] 203. Based on the actual flame radius and the preset radius interception coefficient, determine the interception area radius corresponding to the measured spherical flame.

[0076] Specifically, the preset radius interception coefficient is set according to actual needs. Through the preset radius interception coefficient, a local area where there is no overlap of cellular structures can be intercepted in the measured spherical flame. For example, if the preset radius interception coefficient is 0.5 and the actual flame radius is 10 cm, the radius of the interception area is determined to be 5 cm. Then, an arc area with a radius of 5 cm is intercepted in the measured spherical flame, and the area is determined as a local area where there is no overlap of cellular structures.

[0077] 204. Determine a spherical direction in the measured spherical flame where there is no overlapping of cellular structures, and based on the spherical direction and the radius of the intercepted area, determine an arc flame area in the measured spherical flame where there is no overlapping of cellular structures, and determine a local area image corresponding to the arc flame area in the flame image.

[0078] For the embodiment of the present invention, after determining the interception radius of the local area where there is no overlap of cellular structures, it is also necessary to determine the spherical direction in the measured spherical flame where there is no overlap of cellular structures, that is, the interception direction, and then determine the arc flame area where there is no overlap of cellular structures in the measured spherical flame based on the interception area radius and the interception direction, so as to determine the local area image corresponding to the arc flame area in the flame image.

[0079] 205. Input the local area image into a preset flame crack prediction model to perform crack prediction, and obtain a flame crack image in the local area image.

[0080] For the embodiment of the present invention, in order to improve the prediction accuracy of the preset flame crack prediction model, it is first necessary to construct and train the preset flame crack prediction model. Based on this, the method includes: constructing at least one preset initial flame crack prediction model; obtaining a sample picture set, wherein the sample picture set includes multiple sample flame pictures with annotation information, the annotation information includes an annotation box and an annotation type of an object in the annotation box, and the annotation type includes flame crack; based on the number of models of the preset initial flame crack prediction model, the sample picture set is divided into multiple groups of training sets and multiple groups of test sets, each group of training sets is used to train the corresponding preset initial flame crack prediction model, and each group of test sets is used to test the corresponding trained preset initial flame crack prediction model, and the preset initial flame crack prediction model that meets the test conditions is used as the preset flame crack prediction model.

[0081] The test condition may be that the prediction accuracy of the model is greater than a preset threshold (wherein the preset threshold is set according to actual needs).

[0082] Specifically, the model structures of each preset initial flame crack prediction model can be the same or different. For example, the model structure can be a neural network structure. First, a sample image set with a flame crack annotation box is obtained. The sample image set can be a high-speed camera that captures the hydrogen flame propagation process in a constant volume combustion bomb to obtain flame surface images under different equivalence ratios, turbulence intensities and initial pressures. For example, in this embodiment, a preset initial flame crack prediction model 1 and a preset initial flame crack prediction model 2 are respectively constructed, and the training set can also be randomly divided into training sub-data 1 and training sub-data 2, and the test set can be randomly divided into test sub-data 1 and test sub-data 2. Further, the preset initial flame crack prediction model 1 can be trained using the training sub-data 1, and the preset initial flame crack prediction model 2 can be trained using the training sub-data 2. Thus, two preset initial flame crack prediction models 1 and preset initial flame crack prediction models 2 after training optimization can be obtained. After that, the preset initial flame crack prediction model 1 after training optimization is tested using the test sub-data 1, and the preset initial flame crack prediction model 2 after training optimization is tested using the test sub-data 2. Finally, the model with the highest prediction accuracy is selected from the two models as the preset flame crack prediction model. In the model training process, the small batch gradient descent method can be used to continuously perform iterative training, calculate the output value of the loss function, and then use back propagation derivation to update the network parameters. The model training batch size can be set to 5, the training rounds to 100, the learning rate to 0.00001, the number of training samples to 100, and the model parameters to be saved after the training to obtain the trained preset initial flame crack prediction model. Thus, by training and constructing the preset flame crack prediction model, the subsequent preset flame crack prediction model can make the prediction result more accurate when performing flame crack prediction.

[0083] Furthermore, after training and constructing the preset flame crack prediction model, the local area image without cellular structure overlap in the flame picture is input into the preset flame crack prediction model, and the preset flame crack prediction model can output the flame crack image in the local area image.

[0084] 206. Based on the flame crack image, determine the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determine the flame shape of the measured spherical flame.

[0085] For the embodiment of the present invention, after obtaining the flame crack image, it is necessary to determine the actual flame crack length of the measured spherical flame. Based on this, step 206 specifically includes: determining the pixel value of the reference pixel point in the preset neighborhood corresponding to each pixel point in the flame crack image, and sorting the reference pixel points in the preset neighborhood based on the pixel value of the reference pixel point in the preset neighborhood to obtain each sorted pixel point in the preset neighborhood; determining the median pixel point in each sorted pixel point in the preset neighborhood, and using the median pixel point to replace each corresponding pixel point in the flame crack image to obtain a denoised flame crack image; determining the denoised flame crack image; and determining the median pixel point in the denoised flame crack image. A grayscale histogram H(x, y) corresponding to the denoised flame crack image is obtained, and based on the grayscale histogram H(x, y), a target evaluation crack image in the denoised flame crack image is determined, and a binary crack image corresponding to the target evaluation crack image is determined; a preset skeleton algorithm is used to skeletonize the binary crack image to obtain a flame skeleton image with a single pixel width; based on the flame skeleton image, a reference flame crack length of the arc-shaped flame area corresponding to the local area image is determined; based on the flame radius of the arc-shaped flame area, the actual flame radius of the measured spherical flame, and the reference flame crack length, the actual flame crack length of the measured spherical flame is determined.

[0086] Among them, the preset neighborhood is set according to actual needs. Specifically, taking pixel point 1 in the flame crack image as an example, the pixels in the preset neighborhood corresponding to pixel point 1 are sorted in the order of pixel value from small to large, and the median pixel is determined in each sorted pixel point, and the median pixel is finally used to replace pixel point 1. For example, the sorted pixels in the preset neighborhood corresponding to pixel point 1 are: reference pixel point 1, reference pixel point 2, reference pixel point 3, reference pixel point 4, reference pixel point 5, so that the median pixel can be determined as reference pixel point 3, and finally the reference pixel point 3 is used to replace pixel point 1. Thus, according to the above method, each pixel in the flame crack image can be replaced, so as to obtain a denoised flame crack image. By denoising the flame crack image, the noise in the image can be effectively removed, thereby improving the image quality, and further improving the accuracy of determining the flame morphology.

[0087] Furthermore, there will be abnormally high signal areas in the denoised flame crack image. In order to eliminate the influence of the abnormally high signal areas on the flame morphology determination, it is necessary to determine a target evaluation crack image without abnormally high signals in the denoised flame crack image. Based on this, the method includes: obtaining the peak value P0 of the number of pixels corresponding to each grayscale histogram in the grayscale histogram H (x, y), and calculating the first evaluation value corresponding to the grayscale histogram according to the peak value P0 of the number of pixels and a preset coefficient a, wherein the first evaluation value = a × P0; drawing a first evaluation horizontal line L0 corresponding to the first evaluation value, and obtaining two grayscale histograms X0 that intersect with the first evaluation horizontal line L0 and are closest to each other. l and X u , where X l Less than X u , the first evaluation level L0 is y=a×P0; the gray value is less than X l The first grayscale histogram is linearly fitted to obtain a first reference horizontal line y=P1, the first standard deviation S1 of the first grayscale histogram is calculated, the second evaluation horizontal line L1 is drawn according to the first reference horizontal line and the first standard deviation S1, and the grayscale histogram T1 intersecting with the second evaluation horizontal line L1 in the first grayscale histogram is obtained, wherein the second evaluation horizontal line L1 is y=P1+2×S1; the grayscale values ​​greater than X u A second grayscale histogram is linearly fitted to obtain a second reference horizontal line y=P2, a second standard deviation S2 of the second grayscale histogram is calculated, a third evaluation horizontal line L2 is drawn according to the second reference horizontal line and the second standard deviation S2, and a grayscale histogram T in the second grayscale histogram intersecting with the third evaluation horizontal line L2 is obtained. u , wherein the third evaluation horizontal line L2 is y=P2+2×S2; segmenting the denoised flame crack image (T1, T u ) is used as the target to evaluate the crack image.

[0088] Specifically, firstly, the corresponding grayscale histogram H (x, y) is drawn according to the denoised flame crack image. The grayscale histogram includes the grayscale histogram corresponding to each pixel in the denoised flame crack image. Any grayscale histogram is used to represent the grayscale value of each pixel and the number of pixels corresponding to each grayscale value. x represents the grayscale value of each pixel in the denoised flame crack image, and y represents the number of pixels corresponding to the corresponding grayscale value. The normal grayscale interval corresponding to the grayscale histogram is counted by mathematical statistics, and the target evaluation crack image matching the normal grayscale interval is extracted from the denoised flame crack image as the basis for determining the flame morphology, so as to eliminate the influence of noisy areas and abnormally high signal areas on the determination of flame morphology.

[0089] Specifically, each grayscale histogram in the grayscale histogram is normally distributed, the highest peak value P0 in the grayscale histogram is calculated, and then a first evaluation horizontal line L0 of y=preset coefficient a×P0 is drawn. For example, the preset coefficient a is 1 / 3, and the first evaluation horizontal line is y=(1 / 3)×P0. Obtain two grayscale histograms X where L0 intersects with each grayscale histogram in the histogram and the number of pixels is closest to (1 / 3)×P0. l and X u ,like Figure 3 As shown, the one on the left is X l , the one on the right is X u . Then, take x in (0, X l ) range, use the least squares method to fit a first reference horizontal line y=P1, and calculate the value of x in (0, X l ) range, then draw a second evaluation horizontal line L1 y=P1+2×S1, and calculate the intersection point T1 of L1 and the left side of the histogram. Similarly, take x in (X u , X max ) range, use the least squares method to fit a second reference horizontal line y=P2, and calculate the value of x in (X u , X max ) The standard deviation S2 of the number of pixels corresponding to all grayscale histograms within the range, then draw a third evaluation horizontal line L2 of y=P2+2×S2, and calculate the intersection point T of L2 and the right side of the histogram u Finally, the gray value of the flame crack image after segmentation and denoising is (T1, T u ) is used as the target evaluation crack image. Thus, the target evaluation crack image is segmented in the form of a grayscale histogram, which can eliminate the interference of abnormally high signals in the flame crack image, thereby improving the accuracy of determining the flame morphology.

[0090] Furthermore, after the target evaluation crack image is extracted from the flame crack image, the target evaluation crack image is gray-scaled and binarized to obtain a binary crack image. After that, the binary crack image can be simplified into a flame skeleton image with a single pixel width using a bwmorph (Mathematical Morphology) function such as in Matlab (MatrixLaboratory, commercial mathematical software), and the bwareaopen (Binary image area open, an image processing function) function is used to remove interference points with an area less than 10 pixels. The bwarea (an image processing function) function and the 'thin' (a skeleton length calculation function) option of bwmorph are used to calculate the length of the skeleton, that is, the image flame crack length L of the arc flame area corresponding to the local area image. t Then, according to the pixel size of the image and the scaling ratio of the image acquisition device, the flame crack length L t , determine the reference flame crack length corresponding to the arc flame , then based on the flame radius of the arc flame area , the actual flame radius of the measured spherical flame , calculate the surface area of ​​the arc region , and the surface area corresponding to the spherical flame , and calculate the crack length conversion coefficient according to the following formula :

[0091]

[0092] Furthermore, based on the crack length conversion factor and reference flame crack length , calculate the actual flame crack length of the measured spherical flame according to the following formula: :

[0093]

[0094] Furthermore, after calculating the actual flame crack length of the entire spherical flame based on the reference flame crack length in the local area where there is no cellular structure overlap, it is necessary to determine the flame morphology of the measured spherical flame based on the actual flame crack length and the actual flame radius. Based on this, the method includes: determining the flame morphology of the measured spherical flame based on the actual flame crack length , the actual flame radius , determine the average radius of the flame cell of the measured spherical flame ,in, ; Based on the average radius of the flame cell , the actual flame radius , determine the total number of flame cells of the measured spherical flame ,in, .

[0095] Specifically, in order to more intuitively reflect the degree of spherical flame cellularization, this paper proposes the average radius of flame cells This parameter is used to quantitatively describe the development process of spherical flame cells. If all flame cells are simplified to ideal circles with the same radius, then the average radius of the flame cells is , the circumference of the flame cell With area The following relationship exists:

[0096]

[0097] The area of ​​the flame sphere is equal to the sum of the areas of the flame cells, so and There is the following relationship, where is the total number of flame cells:

[0098]

[0099] Since the wrinkle crack is the "cell wall" that divides each "flame cell", and each section of the cell wall is shared by two "flame cells", the sum of the perimeters of all "flame cells" is equal to twice the total length of the crack, that is, the following relationship exists:

[0100]

[0101] According to the above formula, the average radius of the flame cell can be evaluated based on the flame radius and total crack length obtained in the previous processing. and the total number of flame cells , the specific calculation formula is as follows:

[0102]

[0103]

[0104] Thus, the average radius of the flame cells and the total number of flame cells of the measured spherical flame can be calculated in the above manner. The present invention can quantitatively characterize the morphology of spherical flames under a wide range of working conditions, quickly and accurately obtain a series of flame morphology characterization results such as flame crack length, cellular structure size, and cellular number, and provide a quantitative basis for the morphological evolution analysis of spherical flames, thereby effectively guiding the combustion organization of internal combustion engines and the risk assessment of gas explosion. The advantage of the present invention is that it uses a model to automatically identify flame cracks. This method is more flexible than the artificial threshold method, has less subjective influence, and has a wider range of applications. It can expand the range of spherical flame morphology characterization from simple and regular laminar flames to complex turbulent flames. In addition, the present invention selects the flame arc surface area where there is no overlapping cellular structure for processing, and then uses the arc surface formula to convert it to the entire spherical flame. This step eliminates the error of overlapping cellular structures at the edge of the flame, and the accuracy of quantitative characterization is higher.

[0105] Furthermore, in another embodiment of the present invention, the following method is also included: in response to a flame morphology determination signal of the measured spherical flame, a current flame image of the measured spherical flame at the current moment is collected in real time; based on the current flame image, an actual flame radius of the measured spherical flame at the current moment is determined, and based on the actual flame radius, a local area image of the measured spherical flame without overlapping cellular structures is determined in the flame image; the local area image is input into a preset flame crack prediction model for crack prediction to obtain a flame crack image in the local area image; based on the flame crack image, an actual flame crack length of the measured spherical flame at the current moment is determined, and based on the actual flame crack length and the actual flame radius, the flame morphology of the measured spherical flame at the current moment is determined; based on the flame morphology, a combustion state of the internal combustion engine combustion organization that produces the measured spherical flame is determined; based on the combustion state, process parameters of the internal combustion engine combustion organization are adjusted, wherein the process parameters include combustion technology, ignition advance angle, ignition energy, etc. Therefore, the present invention can further provide a basis for the optimization process of the combustion organization of the internal combustion engine by accurately determining the flame morphology.

[0106] According to another method for determining the morphology of a spherical flame provided by the present invention, compared with the current method of determining the morphology of a flame based on an artificial threshold method, the present invention collects a flame image of the spherical flame to be measured in response to a flame morphology determination signal of the spherical flame to be measured; and based on the flame image, determines the actual flame radius of the spherical flame to be measured, and based on the actual flame radius, determines in the flame image a local area image where there is no overlapping cellular structure of the spherical flame to be measured; then inputs the local area image into a preset flame crack prediction model for crack prediction, and obtains a flame crack image in the local area image; finally, based on the flame crack image, determines the actual flame crack length of the spherical flame to be measured, and based on the actual flame crack length and the actual flame radius, determines the flame morphology of the spherical flame to be measured. Therefore, by determining the local area image without cellular structure overlap in the flame picture corresponding to the measured spherical flame, and using the model to predict the flame crack image in the local area image, and finally determining the flame morphology of the measured spherical flame according to the flame crack image, thereby using the model to predict the flame crack image, and finally automatically determining the flame morphology according to the flame crack image, it is possible to avoid the low determination efficiency and determination errors caused by the artificial threshold method for determining the flame morphology, that is, the present invention can improve the determination efficiency and determination accuracy of the flame morphology. At the same time, the present invention uses the local area image without cellular structure overlap to predict the morphology of the entire spherical flame, which can avoid the morphology prediction error caused by the cellular structure overlap at the edge of the spherical flame, so that the present invention can further improve the determination accuracy of the spherical flame morphology.

[0107] Further, as Figure 1 In a specific implementation, an embodiment of the present invention provides a device for determining a spherical flame shape, such as Figure 4 As shown, the device includes: a collection unit 31, a first determination unit 32, a prediction unit 33 and a second determination unit 34.

[0108] The acquisition unit 31 may be used to acquire a flame image of the spherical flame to be measured in response to a flame shape determination signal of the spherical flame to be measured.

[0109] The first determination unit 32 may be configured to determine an actual flame radius of the measured spherical flame based on the flame image, and determine, based on the actual flame radius, a local region image in the flame image where no cellular structure overlaps with the measured spherical flame.

[0110] The prediction unit 33 may be used to input the local area image into a preset flame crack prediction model to perform crack prediction, and obtain a flame crack image in the local area image.

[0111] The second determination unit 34 may be configured to determine an actual flame crack length of the measured spherical flame based on the flame crack image, and determine a flame shape of the measured spherical flame based on the actual flame crack length and the actual flame radius.

[0112] In a specific application scenario, in order to determine the actual flame radius of the measured spherical flame, such as Figure 5 As shown, the first determination unit 32 includes a removal module 321 , a processing module 322 , a fitting module 323 , and a first determination module 324 .

[0113] The removal module 321 may be used to remove the background image not containing flames from the flame image to obtain a background-removed flame image.

[0114] The processing module 322 may be used to perform grayscale processing and binarization processing on the background-removed flame image to obtain a binarized flame image corresponding to the background-removed flame image.

[0115] The fitting module 323 may be used to fit the edge pixel points of the binary flame image to obtain the edge contour curve of the measured spherical flame in the binary flame image.

[0116] The first determination module 324 may be configured to determine an actual flame radius of the measured spherical flame based on the edge contour curve and a scaling ratio of an image acquisition device corresponding to the flame image.

[0117] In a specific application scenario, in order to determine a local area image where there is no overlapping cellular structure of the measured spherical flame, the first determination module 324 can also be used to determine the radius of the intercepted area corresponding to the measured spherical flame based on the actual flame radius and the preset radius interception coefficient; determine a spherical direction where there is no overlapping cellular structure in the measured spherical flame, and based on the spherical direction and the intercepted area radius, determine an arc-shaped flame area where there is no overlapping cellular structure in the measured spherical flame, and determine a local area image corresponding to the arc-shaped flame area in the flame image.

[0118] In a specific application scenario, in order to construct a preset flame crack prediction model, the device further includes: a construction unit 35.

[0119] The construction unit 35 can be used to construct at least one preset initial flame crack prediction model; obtain a sample picture set, wherein the sample picture set includes multiple sample flame pictures with annotation information, the annotation information includes an annotation box and an annotation type of an object in the annotation box, and the annotation type includes flame crack; based on the number of models of the preset initial flame crack prediction model, divide the sample picture set into multiple groups of training sets and multiple groups of test sets, use each group of training sets to train a corresponding preset initial flame crack prediction model, and use each group of test sets to test the corresponding trained preset initial flame crack prediction model, and use the preset initial flame crack prediction model that meets the test conditions as the preset flame crack prediction model.

[0120] In a specific application scenario, in order to determine the actual flame crack length of the measured spherical flame, the second determination unit 34 includes a second determination module 341 , a replacement module 342 , and a skeletonization module 343 .

[0121] The second determination module 341 can be used to determine the pixel value of the reference pixel point in the preset neighborhood corresponding to each pixel point in the flame crack image, and sort the reference pixel points in the preset neighborhood based on the pixel value of the reference pixel point in the preset neighborhood to obtain the sorted pixel points in the preset neighborhood.

[0122] The replacement module 342 may be used to determine a median pixel point among the sorted pixel points in the preset neighborhood, and use the median pixel point to replace the corresponding pixel points in the flame crack image to obtain a denoised flame crack image.

[0123] The second determination module 341 can be specifically used to determine the grayscale histogram H (x, y) corresponding to the denoised flame crack image, and based on the grayscale histogram H (x, y), determine the target evaluation crack image in the denoised flame crack image, and determine the binary crack image corresponding to the target evaluation crack image.

[0124] The skeletonization module 343 may be used to perform skeletonization processing on the binary crack image using a preset skeleton algorithm to obtain a flame skeleton image with a single pixel width.

[0125] The second determination module 341 may be specifically configured to determine a reference flame crack length of the arc-shaped flame region corresponding to the local region image based on the flame skeleton image.

[0126] The second determination module 341 may be specifically configured to determine the actual flame crack length of the measured spherical flame based on the flame radius of the arc-shaped flame region, the actual flame radius of the measured spherical flame, and the reference flame crack length.

[0127] In a specific application scenario, in order to determine the target evaluation crack image in the denoised flame crack image, the second determination module 341 can be specifically used to obtain the peak value P0 of the number of pixels corresponding to each grayscale histogram in the grayscale histogram H (x, y), and calculate the first evaluation value corresponding to the grayscale histogram according to the peak value P0 of the number of pixels and the preset coefficient a, wherein the first evaluation value = a × P0; draw a first evaluation horizontal line L0 corresponding to the first evaluation value, and obtain two grayscale histograms X0 that intersect with the first evaluation horizontal line L0 and are closest to each other. l and X u , where X l Less than X u , the first evaluation level L0 is y=a×P0; the gray value is less than X l The first grayscale histogram is linearly fitted to obtain a first reference horizontal line y=P1, the first standard deviation S1 of the first grayscale histogram is calculated, the second evaluation horizontal line L1 is drawn according to the first reference horizontal line and the first standard deviation S1, and the grayscale histogram T1 intersecting with the second evaluation horizontal line L1 in the first grayscale histogram is obtained, wherein the second evaluation horizontal line L1 is y=P1+2×S1; the grayscale values ​​greater than X u A second grayscale histogram is linearly fitted to obtain a second reference horizontal line y=P2, a second standard deviation S2 of the second grayscale histogram is calculated, a third evaluation horizontal line L2 is drawn according to the second reference horizontal line and the second standard deviation S2, and a grayscale histogram T in the second grayscale histogram intersecting with the third evaluation horizontal line L2 is obtained. u , wherein the third evaluation horizontal line L2 is y=P2+2×S2; segmenting the denoised flame crack image (T1, T u ) is used as the target to evaluate the crack image.

[0128] In a specific application scenario, the flame shape includes the average radius of the flame cells and the total number of the flame cells of the measured spherical flame. In order to determine the flame shape of the measured spherical flame, the second determining unit 34 can be specifically used to determine the flame shape based on the actual flame crack length. , the actual flame radius , determine the average radius of the flame cell of the measured spherical flame ,in, ; Based on the average radius of the flame cell , the actual flame radius , determine the total number of flame cells of the measured spherical flame ,in, .

[0129] It should be noted that for other corresponding descriptions of the functional modules involved in the device for determining the spherical flame form provided in the embodiment of the present invention, reference can be made to Figure 1 The corresponding description of the method shown will not be repeated here.

[0130] Based on the above Figure 1 The method shown, accordingly, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the following steps are implemented: in response to a flame morphology determination signal of a measured spherical flame, a flame image of the measured spherical flame is collected; based on the flame image, an actual flame radius of the measured spherical flame is determined, and based on the actual flame radius, a local area image of the measured spherical flame without cellular structure overlap is determined in the flame image; the local area image is input into a preset flame crack prediction model for crack prediction, and a flame crack image in the local area image is obtained; based on the flame crack image, an actual flame crack length of the measured spherical flame is determined, and based on the actual flame crack length and the actual flame radius, the flame morphology of the measured spherical flame is determined.

[0131] Based on the above Figure 1 The method shown and Figure 4 The embodiment of the device shown in the figure, the embodiment of the present invention also provides a physical structure diagram of a computer device, such as Figure 6 As shown, the computer device includes: a processor 41, a memory 42, and a computer program stored in the memory 42 and executable on the processor, wherein the memory 42 and the processor 41 are both arranged on a bus 43, and the processor 41 implements the following steps when executing the program: in response to a flame morphology determination signal of a measured spherical flame, a flame image of the measured spherical flame is collected; based on the flame image, an actual flame radius of the measured spherical flame is determined, and based on the actual flame radius, a local area image of the measured spherical flame without cellular structure overlap is determined in the flame image; the local area image is input into a preset flame crack prediction model for crack prediction, and a flame crack image in the local area image is obtained; based on the flame crack image, an actual flame crack length of the measured spherical flame is determined, and based on the actual flame crack length and the actual flame radius, the flame morphology of the measured spherical flame is determined.

[0132] Through the technical solution of the present invention, the present invention collects a flame image of the measured spherical flame in response to a flame morphology determination signal of the measured spherical flame; and based on the flame image, determines the actual flame radius of the measured spherical flame, and based on the actual flame radius, determines in the flame image a local area image where there is no overlapping cellular structure of the measured spherical flame; then inputs the local area image into a preset flame crack prediction model for crack prediction, and obtains a flame crack image in the local area image; finally, based on the flame crack image, determines the actual flame crack length of the measured spherical flame, and based on the actual flame crack length and the actual flame radius, determines the flame morphology of the measured spherical flame. Therefore, by determining the local area image without cellular structure overlap in the flame picture corresponding to the measured spherical flame, and using the model to predict the flame crack image in the local area image, and finally determining the flame morphology of the measured spherical flame according to the flame crack image, thereby using the model to predict the flame crack image, and finally automatically determining the flame morphology according to the flame crack image, it is possible to avoid the low determination efficiency and determination errors caused by the artificial threshold method for determining the flame morphology, that is, the present invention can improve the determination efficiency and determination accuracy of the flame morphology. At the same time, the present invention uses the local area image without cellular structure overlap to predict the morphology of the entire spherical flame, which can avoid the morphology prediction error caused by the cellular structure overlap at the edge of the spherical flame, so that the present invention can further improve the determination accuracy of the spherical flame morphology.

[0133] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0134] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining a spherical flame shape, characterized in that: include: In response to a flame shape determination signal of the measured spherical flame, collecting a flame picture of the measured spherical flame; A background image not containing flames is removed from the flame image to obtain a background-removed flame image, grayscale processing and binarization processing are performed on the background-removed flame image to obtain a binary flame image corresponding to the background-removed flame image, and edge pixel points of the binary flame image are fitted to obtain an edge contour curve of the measured spherical flame in the binary flame image; Determine the actual flame radius of the measured spherical flame based on the edge contour curve and the scaling ratio of the image acquisition device corresponding to the flame picture, and determine, based on the actual flame radius, a local area image of the measured spherical flame in which no cellular structure overlaps in the flame picture; The local area image is input into a preset flame crack prediction model for crack prediction, and a flame crack image in the local area image is obtained, wherein the method for obtaining the preset flame crack prediction model includes: constructing at least one preset initial flame crack prediction model, and obtaining a sample picture set, wherein the sample picture set includes a plurality of sample flame pictures with annotation information, wherein the annotation information includes an annotation box and an annotation type of an object in the annotation box, wherein the annotation type includes flame crack, and based on the number of models of the preset initial flame crack prediction model, the sample picture set is divided into a plurality of training sets and a plurality of test sets, wherein each training set is used to train a corresponding preset initial flame crack prediction model, and each test set is used to test the corresponding trained preset initial flame crack prediction model, and the preset initial flame crack prediction model that meets the test conditions is used as the preset flame crack prediction model; Based on the flame crack image, the actual flame crack length of the measured spherical flame is determined, and based on the actual flame crack length and the actual flame radius, the flame morphology of the measured spherical flame is determined.

2. The method for determining the spherical flame shape according to claim 1, characterized in that: The determining, based on the actual flame radius, in the flame image, a local area image where the measured spherical flame does not have overlapping cellular structures comprises: Determining the radius of the intercepted area corresponding to the measured spherical flame based on the actual flame radius and the preset radius interception coefficient; A spherical direction in which no cellular structure overlaps is determined in the measured spherical flame, and an arc-shaped flame region in which no cellular structure overlaps is determined in the measured spherical flame based on the spherical direction and the radius of the intercepted region, and a local area image corresponding to the arc-shaped flame region is determined in the flame image.

3. The method for determining the spherical flame shape according to claim 1, characterized in that: The determining, based on the flame crack image, an actual flame crack length of the measured spherical flame comprises: Determine the pixel value of the reference pixel point in the preset neighborhood corresponding to each pixel point in the flame crack image, and sort the reference pixel points in the preset neighborhood based on the pixel value of the reference pixel point in the preset neighborhood to obtain each sorted pixel point in the preset neighborhood; Determine a median pixel point among the sorted pixel points in the preset neighborhood, and use the median pixel point to replace each corresponding pixel point in the flame crack image to obtain a denoised flame crack image; Determine a grayscale histogram H(x, y) corresponding to the denoised flame crack image, and based on the grayscale histogram H(x, y), determine a target evaluation crack image in the denoised flame crack image, and determine a binary crack image corresponding to the target evaluation crack image; The binary crack image is skeletonized by using a preset skeleton algorithm to obtain a flame skeleton image with a single pixel width; Based on the flame skeleton image, determining a reference flame crack length of the arc-shaped flame area corresponding to the local area image; The actual flame crack length of the measured spherical flame is determined based on the flame radius of the arc-shaped flame area, the actual flame radius of the measured spherical flame, and the reference flame crack length.

4. The method for determining the spherical flame shape according to claim 3, characterized in that: The grayscale histogram H(x, y) includes the grayscale histogram corresponding to each pixel in the denoised flame crack image, and any grayscale histogram is used to represent the grayscale value of each pixel and the number of pixels corresponding to each grayscale value; The step of determining a target evaluation crack image in the denoised flame crack image based on the grayscale histogram H(x, y) includes: Obtaining a peak value P0 of the number of pixels corresponding to each grayscale histogram in the grayscale histogram H(x, y), and calculating a first evaluation value corresponding to the grayscale histogram according to the peak value P0 of the number of pixels and a preset coefficient a, wherein the first evaluation value=a×P0; Draw a first evaluation horizontal line L0 corresponding to the first evaluation value, and obtain two grayscale histograms X that intersect with the first evaluation horizontal line L0 and are closest to each other. l and X u , where X l Less than X u , the first evaluation level L0 is y=a×P0; Set the gray value less than X l A first grayscale histogram is linearly fitted to obtain a first reference horizontal line y=P1, a first standard deviation S1 of the first grayscale histogram is calculated, a second evaluation horizontal line L1 is drawn according to the first reference horizontal line and the first standard deviation S1, and a grayscale histogram T in the first grayscale histogram intersecting with the second evaluation horizontal line L1 is obtained. l , wherein the second evaluation level L1 is y=P1+2×S1; Set the gray value greater than X u A second grayscale histogram is linearly fitted to obtain a second reference horizontal line y=P2, a second standard deviation S2 of the second grayscale histogram is calculated, a third evaluation horizontal line L2 is drawn according to the second reference horizontal line and the second standard deviation S2, and a grayscale histogram T in the second grayscale histogram intersecting with the third evaluation horizontal line L2 is obtained. u , wherein the third evaluation level L2 is y=P2+2×S2; Segment the denoised flame crack image (T l , T u ) is used as the target to evaluate the crack image.

5. The method for determining the spherical flame shape according to claim 1, characterized in that: The flame morphology includes the average radius of flame cells and the total number of flame cells of the measured spherical flame; The step of determining the flame shape of the measured spherical flame based on the actual flame crack length and the actual flame radius includes: Based on the actual flame crack length , the actual flame radius , determine the average radius of the flame cell of the measured spherical flame ,in, ; Based on the average radius of the flame cell , the actual flame radius , determine the total number of flame cells of the measured spherical flame ,in, .

6. A device for determining a spherical flame shape, characterized in that: include: A collection unit, configured to collect a flame image of the spherical flame under test in response to a flame shape determination signal of the spherical flame under test; The first determining unit is used to remove the background image that does not contain flames from the flame picture to obtain a background-removed flame picture, perform grayscale processing and binarization processing on the background-removed flame picture to obtain a binary flame picture corresponding to the background-removed flame picture, and perform fitting on edge pixel points of the binary flame picture to obtain an edge contour curve of the measured spherical flame in the binary flame picture; Determine the actual flame radius of the measured spherical flame based on the edge contour curve and the scaling ratio of the image acquisition device corresponding to the flame picture, and determine, based on the actual flame radius, a local area image of the measured spherical flame in which no cellular structure overlaps in the flame picture; A prediction unit, used for inputting the local area image into a preset flame crack prediction model for crack prediction, and obtaining a flame crack image in the local area image, wherein the method for obtaining the preset flame crack prediction model comprises: constructing at least one preset initial flame crack prediction model, and obtaining a sample picture set, wherein the sample picture set comprises a plurality of sample flame pictures with annotation information, wherein the annotation information comprises an annotation box and an annotation type of an object in the annotation box, wherein the annotation type comprises flame crack, and based on the number of models of the preset initial flame crack prediction model, the sample picture set is divided into a plurality of training sets and a plurality of test sets, wherein each training set is used to train a corresponding preset initial flame crack prediction model, wherein each test set is used to test the corresponding trained preset initial flame crack prediction model, and the preset initial flame crack prediction model that meets the test conditions is used as the preset flame crack prediction model; The second determination unit is used to determine the actual flame crack length of the measured spherical flame based on the flame crack image, and determine the flame shape of the measured spherical flame based on the actual flame crack length and the actual flame radius.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.