A method and system for analyzing the flame of a hydrogen chloride synthesis furnace

Abstract

The present invention relates to the technical field of image processing, and in particular to a method and system for analyzing the flame of a hydrogen chloride synthesis furnace, including using a flame sensor to determine whether there is a flame in the hydrogen chloride synthesis furnace; collecting a flame video and obtaining a flame image by frame division; performing color detection on the obtained image under the RGB and HSI mixed color models to obtain a flame pixel point distribution matrix and obtain a preliminary image containing a suspected flame area; performing grayscale processing on the preliminary image containing the suspected flame area, using PSO search for global optimization, and substituting the global optimal solution back into the K-means clustering algorithm to perform clustering on the image; multiplying the matrix of the preliminary image containing the suspected flame area by the matrix of the flame image after clustering to obtain the final flame area; adjusting the gas concentration according to the RGB mean value and color temperature of the final flame area. The present invention solves the problem of poor control quality of manually observing the flame color.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a method and system for analyzing the flame of a hydrogen chloride synthesis furnace. Background Art

[0002] Chlor-alkali enterprises use synthesis furnaces to produce hydrogen chloride gas, which provides raw materials for the production of polyvinyl chloride. Chlorine and hydrogen are produced by electrolyzing industrial salt using the mercury method. Due to various reasons in the production process, the interference with the synthesis of hydrogen chloride is very serious, and it is difficult to ensure that hydrogen chloride maintains a reasonable ratio.

[0003] In the prior art, the method of manually observing the flame color of the synthesis furnace and regularly sampling and inspecting the hydrogen chloride content is used to ensure safe production. However, the quality of manual control is poor, and the lag time of off-line chemical analysis is long, making it difficult to ensure the product qualification, thus causing a certain economic loss. Summary of the Invention

[0004] Aiming at the deficiencies of the existing methods, the present invention solves the problem of poor control quality of manually observing the flame color.

[0005] The technical solution adopted by the present invention is that a method and system for analyzing the flame of a hydrogen chloride synthesis furnace includes the following steps:

[0006] Step 1: Use a flame sensor to determine whether there is a flame in the hydrogen chloride synthesis furnace;

[0007] Step 2: Collect a flame video and obtain a flame image by frame division;

[0008] Further, obtaining the flame image by frame division is to read the video at an interval of two frames.

[0009] Step 3: Perform color detection on the obtained image under the RGB and HSI mixed color models, obtain the flame pixel point distribution matrix, and obtain a preliminary image containing a suspected flame area;

[0010] Further, the formula of the RGB and HSI mixed color model is:

[0011] rule1: R > R mean

[0012] rule2: R ≥ G ≥ B

[0013] rules3: S ≥ (255 - R)S Threshold / R Threshold

[0014] In the formula, S Threshold is the saturation threshold, and the reference range is 55 - 65; R Threshold is the red component threshold, and the reference range is 115 - 135; Rmean is the mean value of the red component.

[0015] Step 4: Grayscale the image with the preliminary suspected flame area, perform global optimization using PSO search, and substitute the global optimal solution back into the K-means clustering algorithm to cluster the image;

[0016] Further, Step 4 specifically includes:

[0017] Step 41: Initialize the running parameters of the number of particles, and initialize the positions and velocities of the particle swarm;

[0018] Step 42: Calculate the fitness value of each particle;

[0019] Step 43: Update the individual optimal P best and the global optimal g best ;

[0020] Step 44: Update the learning factor, inertia weight, and the positions and velocities of the particles;

[0021] Step 45: Calculate the fitness variance;

[0022] Step 46: Set the fitness variance threshold as the iteration termination condition;

[0023] Step 47: Use the result calculated by the PSO algorithm as the K-means clustering center;

[0024] Step 48: Calculate the Euclidean distance for the distance from each center of all objects;

[0025] Step 48: Update the clustering center and terminate when the convergence condition is reached.

[0026] Step 5: Multiply the matrix of the image with the preliminary suspected flame area by the matrix of the flame image after clustering to obtain the final flame area;

[0027] Further, Step 5 specifically includes:

[0028] Let the matrix of the image with the preliminary suspected flame area be A, the matrix of the flame image after clustering be B, and the resulting matrix after multiplication be C. Then:

[0029] C(i,j) = A(i,j) * B(i,j)

[0030] where i represents the row number of the pixel point, j represents the column number of the pixel point; C(i,j) represents the value of the corresponding pixel point in the resulting matrix, and A(i,j) and B(i,j) respectively represent the values of the same pixel point in the two matrices;

[0031] Through threshold processing, the values in the matrix obtained after multiplication are converted into a binary image to obtain the contour and specific position of the flame.

[0032] Step 6: Adjust the gas concentration according to the RGB mean value and color temperature of the final flame area.

[0033] Further, Step 6 specifically includes:

[0034] Step 61: Compare the RGB mean value of the final flame area with the RGB values of the cyan-white flame color under normal concentration ratio;

[0035] Step 62: Compare the color temperature of the final flame with the color temperature of the cyan-white flame under normal concentration ratio;

[0036] Step 63: When the color temperature and RGB mean value of the flame are not satisfied, adjust the gas concentration and output an alarm signal.

[0037] Further, a flame analysis system based on a hydrogen chloride synthesis furnace includes: a Raspberry Pi, an explosion-proof flame camera, a flame sensor, and a spectral sensor. The explosion-proof flame camera, the flame sensor, and the spectral sensor are electrically connected to the Raspberry Pi;

[0038] The flame sensor is used to detect whether there is a flame in the hydrogen chloride synthesis furnace;

[0039] The explosion-proof flame camera is used to collect a flame video;

[0040] The spectral sensor is used to detect the flame color temperature;

[0041] The Raspberry Pi converts the flame video into an image, and the image is detected under a mixed color model of RGB and HSI color models to initially obtain an area containing a suspected flame; the image of the area containing the suspected flame obtained initially is grayscaled, PSO is used to search and find the global optimum, and the search value is substituted back into K-means for clustering; the area containing the suspected flame obtained initially is multiplied by the image matrix after clustering to obtain the final suspected flame area; the RGB mean value of the final suspected flame area is calculated; it is judged whether the gas concentration meets the conditions according to the color temperature and RGB values.

[0042] Advantages of the present invention:

[0043] 1. Using a mixed color model of RGB and HSI to obtain an image of an area containing a suspected flame, adopting a method combining PSO-K-means to cluster the image of the suspected flame area, and combining the two images to obtain an accurate flame area image, improving the automation degree and working efficiency, reducing the human and material costs, having a rapid early warning response, and improving the detection accuracy;

[0044] 2. Use the Raspberry Pi 4B as a controller to collect and send data. The Raspberry Pi is small in size and low in cost, saving costs for users.

[0045] 3. By arranging explosion-proof cameras at close range, the problem that testers in the existing solution are difficult to continuously observe and have limited vision is solved.

[0046] 4. Adopt a flame sensor to judge whether there is flame combustion in the hydrogen chloride synthesis furnace. If there is a flame, the camera will be automatically turned on to monitor the flame in real time. Combine with a spectral sensor to detect the flame-related color temperature corresponding to different concentrations during hydrogen chloride synthesis in real time, improving the test accuracy and traceability.

[0047] 5. Measure the distance and time with high precision throughout the process, and record the close-range process images, improving the automation level and work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is the flowchart of the method for analyzing the flame of the hydrogen chloride synthesis furnace according to the present invention;

[0049] Figure 2 is the flowchart of the PSO-Kmeans method of the present invention;

[0050] Figure 3 is the block diagram of the system for analyzing the flame of the hydrogen chloride synthesis furnace according to the present invention;

[0051] Figure 4 is the image segmentation diagram of the cyan-white flame burning under normal concentration ratio during hydrogen chloride synthesis according to the present invention;

[0052] Figure 5 is the image segmentation diagram of the yellow and red flames burning when chlorine is in excess during hydrogen chloride synthesis according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] The present invention will be further described below with reference to the drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, so it only shows the components related to the present invention.

[0054] As Figure 1 shown, a method for analyzing the flame of a hydrogen chloride synthesis furnace includes the following steps:

[0055] Step 1. Use a flame sensor to judge whether there is flame combustion in the hydrogen chloride synthesis furnace. If there is a flame, the explosion-proof flame monitoring camera will be automatically turned on to collect real-time video of the flame in the furnace.

[0056] Step 2. Extract the video to be tested in sequence by reading the video at intervals of a certain number of frames (every two frames in this embodiment).

[0057] Every two frames can reduce the burden on the computer to process the video stream and improve the program running efficiency. At the same time, reading frames at intervals can also reduce the amount of data and lower the storage and transmission costs. During the process of monitoring the flame, due to the large repetition rate of the flame combustion state and color, reading frames at intervals can reduce the demand for storage devices and lower the costs.

[0058] Step 3: Use the mixed color feature criterion of the RGB and HSI color models. According to this criterion, obtain the initial flame pixel points, record the number of flame pixel points in each small block in a matrix, which is the flame pixel point distribution matrix; the NumPy library in Python can be used to construct the matrix and store the statistical results in the matrix; extract the flame pixel points to obtain the flame pixel point distribution matrix to get a preliminary image containing the suspected flame area;

[0059] The RGB model is based on the additive color mixing principle of the three primary color lights of red, green, and blue, while the HSI model is based on the description of three parameters: hue, saturation, and brightness; however, the mixed model is a strategy of combining multiple models; fuse the RGB and HSI models to obtain better color description and higher image quality; by performing weighted averaging on the features of the RGB and HSI models, a new mixed model is obtained, which can not only retain the color information of the RGB model but also reflect the human visual perception characteristics of the HSI model.

[0060] The mixed model criterion is as follows:

[0061] rule1: R > R mean

[0062] rule2: R ≥ G ≥ B

[0063] rules3: S ≥ (255 - R)S Threshold / R Threshold

[0064] In the formula, S Threshold is the saturation threshold, and the reference range is 55 - 65; R Threshold is the red component threshold, and the reference range is 115 - 135; R mean is the average value of the red component.

[0065] It can be seen from the formula that this mixed criterion not only retains part of the RGB model but also integrates the saturation detection in the HIS color space.

[0066] Before image clustering, the color image should be converted into a grayscale image first. The reason is to reduce the complexity and dimension of the image. A color image usually consists of three channels: red, green, and blue, and each channel has 256 gray levels, which means a color image has millions of possible color combinations. This makes color image clustering very complex and requires more computing resources and time. In contrast, a grayscale image has only one channel, and each pixel has only 256 gray levels. This greatly reduces the computational amount of grayscale image clustering and allows for faster clustering analysis. The weighted average method is used for the grayscale processing of the flame image to be recognized;

[0067] The formula for the weighted average value is: Gray value = 0.299 * Red component + 0.587 * Green component + 0.114 * Blue component;

[0068] Among them, the three parameters 0.299, 0.587, and 0.114 respectively correspond to the weight values of the red, green, and blue components in the gray value.

[0069] As Figure 2 shown, in Step 4, the color image containing the suspected flame area obtained initially is converted into a grayscale image, and the particle swarm optimization (PSO) algorithm is used to search and perform global optimization, and the search value is substituted back into the K-means clustering algorithm to cluster the image;

[0070] When using K-means for clustering, the optimal solution obtained by the optimized particle swarm optimization (PSO) algorithm is used as the initial centroid, and then K-means clustering is performed.

[0071] PSO algorithm principle:

[0072] Suppose in a D-dimensional target search space, there is a community composed of N particles, and the i-th particle is represented as:

[0073] X i =(X i1 ,X i2 ,...,X iD ),i = 1,2,...,N (1)

[0074] The velocity of the i-th particle is denoted as:

[0075] V i =(V i1 ,V i2 ,...,V iD ),i = 1,2,...,N (2)

[0076] The optimal position searched by the i-th particle is called the individual extreme value, denoted as:

[0077] P best=(P i1 , P i2 ,..., P iD ), i = 1, 2,..., N (3)

[0078] The optimal position found by the entire particle swarm so far is the global extreme value, denoted as:

[0079] g best =(P g1 , P g2 ,... P gD ) (4)

[0080] When these two optimal values are found, the particles update their velocities and positions according to the following formula:

[0081]

[0082]

[0083] where C1 and C2 are learning factors, also known as acceleration constants; ω is the inertia weight; rand1 and rand1 are uniformly distributed random numbers in the range [0, 1].

[0084] The optimized inertia weight formula is as follows:

[0085]

[0086] where N(0, 1) is the normal distribution, is the inertia adjustment factor, ω max represents the maximum inertia weight; ω min represents the minimum inertia weight; iter max represents the maximum number of iterations; k represents the current number of iterations.

[0087] The values of the learning factors have a great influence on the "cognitive" ability and "social" ability of the particles. The changes made by the learning factors are:

[0088]

[0089]

[0090] where c 1s , c 2s represent the initial values of the learning factors c1 and c2 respectively, c 1e , c 2e represent the final values of the learning factors c1 and c2 respectively; t represents the current number of iterations; T represents the maximum number of iterations.

[0091] Finally, calculate the fitness variance σ of the particle swarm 2, and use this value to judge the convergence degree of the particle swarm, which is defined as follows:

[0092]

[0093] Among them, N represents the number of particles in the particle swarm, and h k represents the fitness of the k-th particle, represents the current average fitness of the particle swarm, and h is the normalization factor; the variance of the population fitness reflects the "convergence" degree of all particles in the particle swarm. The smaller σ 2 , the greater the convergence degree of the particle swarm.

[0094] Steps of the PSO-Kmeans algorithm:

[0095] The first step: Initialize the number of particles. First, set the relevant parameters for the operation of the PSO algorithm, and initialize the positions and velocities of the particle swarm; the initialization of the particle velocity can be randomly generated. For the initialization of the particle positions, each clustering sample can be randomly assigned to a certain cluster first to generate the initial clustering division result; on this basis, according to the mean values of the samples belonging to each cluster specified initially, calculate the clustering centers of each cluster, and use this center as the initial particle positions; continuously repeat the above particle initialization process until all particles are initialized, that is, complete the initialization of the particle swarm;

[0096] The second step: Calculate the fitness value of each particle;

[0097] The third step: Update the individual optimal P best and the global optimal g best ;

[0098] The third step: Update the learning factor using formulas (8) and (9), and update the inertia weight using formula (7);

[0099] The fourth step: Update the positions and velocities of the particles using formulas (5) and (6);

[0100] The fifth step: Calculate the fitness variance using formula (10);

[0101] The sixth step: Set the fitness variance threshold. When the fitness variance is less than the threshold, it means that the PSO algorithm has reached convergence. When convergence is achieved, switch to the K-means algorithm.

[0102] Steps of the K-means algorithm:

[0103] (1). Use the global optimal solution or approximate global optimal solution found by the PSO algorithm as the K-means clustering center;

[0104] (2). Calculate the distance from the point to the position of the clustering center using the Euclidean distance formula.

[0105] Euclidean distance formula

[0106] Given a data set:

[0107] X = {X m | m = 1, 2, 3, …, total}

[0108] where the samples in X are represented by d descriptive attributes A1, A2, …, A d , and the d descriptive attributes are all continuous attributes.

[0109] The data sample X l = (X l1 , X l2 , … X ld ),, X k = (X k1 , X k2 , … X kd ) where X l1 , X l2 , … X ld and X k1 , X k2 , … X kd are the specific values corresponding to the d descriptive attributes A1, A2, …, A l and X l . d

[0110] The Euclidean distance d(x l , x l ) between the samples X l and X k is

[0111]

[0112] where x l , x k is the data set.

[0113] (3) Update the cluster centers. If the maximum number of iterations is reached, it converges; otherwise, repeat (2).

[0114] Step Five: Multiply the preliminary image matrix containing the suspected flame region by the flame image matrix after clustering to complete image segmentation and obtain the final flame region;

[0115] The specific calculation formula is as follows:

[0116] Let the preliminary image matrix containing the suspected flame region be A, the flame image matrix after clustering be B, and the resulting matrix after multiplication be C. Then:

[0117] C(i, j) = A(i, j) * B(i, j)​

[0118] Among them, i represents the row number of the pixel, and j represents the column number of the pixel; C(i, j) represents the value of the corresponding pixel in the result matrix, and A(i, j) and B(i, j) respectively represent the values of the same pixel in the two matrices.

[0119] The result obtained by multiplication is the part where the matrix containing the suspected flame area obtained initially is similar to the flame image matrix after clustering. Finally, through threshold processing, the values in the matrix obtained by multiplication are converted into a binary image to obtain the contour and specific position of the flame.

[0120] Step Six: Calculate the RGB mean value of the final flame area, and determine whether the RGB mean value is within the RGB range under normal gas concentration ratio;

[0121] Whether the RGB mean value is within the RGB range under normal gas concentration ratio, the RGB mean value of the segmented flame area can be calculated by the following formula:

[0122] R mean value = (R1 + R2 +... + Rn) / n;

[0123] G mean value = (G1 + G2 +... + Gn) / n;

[0124] B mean value = (B1 + B2 +... + Bn) / n;

[0125] Among them, R1, G1, B1 to Rn, Gn, Bn are the red, green, and blue channel values of each pixel in the flame area respectively, and n is the number of pixels in the flame area.

[0126] The RGB intensity value range of the cyan-white flame color under normal concentration ratio is: the red component range is 50 - 90; the green component range is about 150 - 200; the blue component range is 200 - 255.

[0127] If the obtained RGB mean value is not within the RGB component range, adjust the gas concentration and re-detect; when the RGB component range is satisfied, then detect whether the color temperature is normal. If the color temperature is abnormal, give an alarm.

[0128] Connect the AS7341 spectral sensor to the controller (Raspberry Pi) for communication. Before starting to use, the sensor needs to be initialized first. After successful initialization, data can be read. Use the AS7341 spectral sensor to calculate the relevant color temperature of the final flame area and determine whether it is the normal flame color temperature; the relevant color temperature of the cyan-white flame is usually between 5000K and 7000K.

[0129] The output of the spectral sensor depends on the gain parameter AGAIN (gain) and the integration time TINT (integration time) of the device; the value of the integration time TINT in turn depends on the two registers ATIME and ASTEP; the calculation of the integration time TINT is as shown in the following formula (1):

[0130] TINT = (ATIME + 1) × (ASTEP + 1) × 2.78μ (1)

[0131] In the formula, ATIME is the time length of data acquisition, ASTEP is the time interval of data acquisition, and μ represents microseconds.

[0132] By configuring the two parameters AGAIN and TINT to make the device output as large as possible, the measurement accuracy is improved; if signal saturation is required, the GAIN value and the values of ATIME and ASTEP need to be changed to adjust the device output; therefore, during measurement, it is necessary to adjust the sensor settings according to the sensor output to keep the sensor output within the ideal range, which will improve the detection accuracy.

[0133] The calculation of the spectral related color temperature depends on the normalized value BasicCount, and the calculation of this value requires the sensor output and the configuration parameters during detection, as shown in the following formula (2):

[0134]

[0135] The color temperature can be calculated based on the correction data for mass production of the spectral sensor provided by AMS and the XYZ correction matrix based on the CIE1931 color-matching functions; X, Y, and Z are the standard XYZ chromaticity values; the XYZ correction matrix is shown in Table 1 below; the values of X, Y, and Z can be calculated according to the following formula (3):

[0136] X CIE1931 = [0.39814... -00.02347] · [F1 BasicCount ...NIR BasicCount -1

[0137] Y CIE1931 = [0.01396... -0.01993] · [F1 BasicCount ...NIR BasicCount -1 (3)

[0138] Z CIE1931 = [1.95010... -0.00938] · [F1 BasicCount ...NIR BasicCount -1 ​​​

[0139] Where: F1 BasicCount , NIR BasicCount is the basic value of F1-NIR in the XYZ correction matrix.

[0140] The relative coefficients x, y, z in the color space can be calculated using the following formula:

[0141]

[0142] And the corresponding correlated color temperature calculation formula (5):

[0143]

[0144] Table 1 XYZ correction matrix

[0145] F1 F2 F3 F4 F5 F6 F7 F8 Clear NIR 410 440 470 510 550 583 620 670 750 900 X 0.39814 1.29540 0.36956 0.10902 0.71942 1.78180 1.10110 -0.03991 -0.27597 -0.02347 Y 0.01396 0.16748 0.23538 1.42750 1.88670 1.14200 0.46497 -0.02702 -0.24468 -0.01993 Z 1.95010 6.45490 2.78010 0.18501 0.15325 0.09539 0.10563 0.08866 -0.61140 -0.00938

[0146] From this, the value of the correlated color temperature can be calculated. Use the obtained correlated color temperature to determine whether the flame area is between 5000K and 7000K. If not satisfied, an alarm signal will be issued. If satisfied, the determination ends.

[0147] Such as Figure 3 , a flame analysis system based on a hydrogen chloride synthesis furnace, including: Raspberry Pi, explosion-proof flame camera, flame sensor, spectral sensor, display; connect the flame sensor, explosion-proof flame camera, AS7341 spectral sensor and display to the Raspberry Pi 4B through GPIO pins or USB interfaces, etc.; the model of the Raspberry Pi is Raspberry Pi 4B; the model of the spectral sensor is AS7341.

[0148] The flame sensor is used to detect whether there is a flame in the hydrogen chloride synthesis furnace;

[0149] The explosion-proof flame camera is used to collect flame videos;

[0150] The spectral sensor is used to detect the flame color temperature;

[0151] The Raspberry Pi converts the flame video into an image, and the image is detected under a mixed color model of RGB and HSI color models to initially obtain an area containing a suspected flame;

[0152] The Raspberry Pi grayscales the image of the area containing the suspected flame obtained initially, uses PSO to search and find the global optimum, and substitutes the search value back into K-means for clustering; multiplies the area containing the suspected flame obtained initially by the image matrix after clustering to complete the segmentation to obtain the final suspected flame area;

[0153] The Raspberry Pi calculates the RGB mean value of the final suspected flame area and uses the RGB mean value to judge the flame color;

[0154] Judge whether the gas concentration meets the conditions according to the color temperature and color.

[0155] Such as Figure 4 It is the original image and the image segmentation diagram of the combustion flame being cyan-white under normal concentration ratio during hydrogen chloride synthesis when observed outside the hydrogen chloride synthesis furnace; such as Figure 5 It is the original image and the image segmentation diagram of the combustion flame turning yellow and red when chlorine is in excess during hydrogen chloride synthesis when observed outside the hydrogen chloride synthesis furnace.

[0156] Inspired by the ideal embodiments of the present invention described above, through the above description, relevant staff can completely make various changes and modifications without departing from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for analyzing the flame of a hydrogen chloride synthesis furnace, characterized in that, It includes the following steps: Step 1: Use a flame sensor to determine whether there is a flame in the hydrogen chloride synthesis furnace; Step 2: Collect a flame video and obtain flame images by frame extraction; Step 3: Perform color detection on the obtained images under the RGB and HSI mixed color models to obtain a flame pixel point distribution matrix and get a preliminary image containing a suspected flame area; Step 4: Grayscale the preliminary image containing the suspected flame area, use PSO search for global optimization, and substitute the global optimal solution back into the K-means clustering algorithm to cluster the image; Step 5: Multiply the preliminary image matrix containing the suspected flame area by the flame image matrix after clustering to obtain the final flame area; Step 6: Adjust the gas concentration according to the RGB mean value and color temperature of the final flame area.

2. The method for analyzing the flame of a hydrogen chloride synthesis furnace according to claim 1, characterized in that Obtaining flame images by frame extraction means reading the video at an interval of two frames.

3. The method for analyzing the flame of a hydrogen chloride synthesis furnace according to claim 1, wherein, The rules of the RGB and HSI mixed color model are: rule1: R > R mean rule2: R≥G≥B rules3: S≥(255 - R)S Threshold / R Threshold where S Threshold is the saturation threshold; R Threshold is the red component threshold; R mean is the mean value of the red component.

4. The method for analyzing the flame of a hydrogen chloride synthesis furnace according to claim 1, wherein Step 4 specifically includes: Step 41: Initialize the particle number operation parameters and initialize the positions and velocities of the particle swarm; Step 42: Calculate the fitness value of each particle; Step 43: Update the individual optimal P best and the global optimal g best ; Step 44: Update the learning factor, inertia weight, and the positions and velocities of the particles; Step 45: Calculate the fitness variance; Step 46: Set the fitness variance threshold as the iteration termination condition; Step 47: Use the global optimal solution or approximate global optimal solution found by the PSO algorithm as the K-means clustering center; Step 48: Calculate the Euclidean distance for the distance from each center of all objects; Step 48: Update the clustering center and terminate when the convergence condition is reached.

5. The method for analyzing the flame of a hydrogen chloride synthesis furnace according to claim 1, characterized in that, Step 5 specifically includes: Let the preliminary matrix containing the suspected flame area be A, the flame image matrix after clustering be B, and the resulting matrix after multiplication be C. Then: C(i,j) = A(i,j) * B(i,j) where i represents the row number of the pixel point, j represents the column number of the pixel point; C(i,j) represents the value of the corresponding pixel point in the resulting matrix, and A(i,j) and B(i,j) respectively represent the values of the same pixel point in the two matrices; Convert the values in the matrix obtained after multiplication into a binary image through threshold processing to obtain the contour and position of the flame.

6. The method for analyzing the flame of a hydrogen chloride synthesis furnace according to claim 1, wherein Step 6 specifically includes: Step 61: Compare the RGB mean value of the final flame area with the RGB values of the cyan-white flame color under normal concentration ratio; Step 62: Compare the color temperature of the final flame with the color temperature of the cyan-white flame under normal concentration ratio; Step 63: When the color temperature and RGB mean value of the flame are not satisfied, adjust the gas concentration and output an alarm signal.

7. A flame analysis system for a hydrogen chloride synthesis furnace, characterized in that, It includes: A Raspberry Pi, an explosion-proof flame camera, a flame sensor, and a spectral sensor. The explosion-proof flame camera, the flame sensor, and the spectral sensor are electrically connected to the Raspberry Pi; The flame sensor is used to detect whether there is a flame in the hydrogen chloride synthesis furnace; The explosion-proof flame camera is used to collect a flame video; The spectral sensor is used to detect the flame color temperature; The Raspberry Pi converts the flame video into an image and performs detection on the image under the mixed color model of the RGB and HSI color models to initially obtain an area containing a suspected flame; Gray-scale the initially obtained image containing the suspected flame area, use PSO to search for and find the global optimum, substitute the search value back into K-means for clustering; multiply the initially obtained image containing the suspected flame area by the clustered image matrix to obtain the final suspected flame area; Calculate the RGB mean value of the final suspected flame area; Judge whether the gas concentration meets the conditions according to the color temperature and RGB values.

Images