A method and system for visual detection of instantaneous explosion smoke

Abstract

The present invention belongs to the field of computer vision technology, and discloses a method and system for visually detecting instantaneous thick smoke, including collecting thick smoke image data, time series image data and ambient light data; processing the thick smoke image data to obtain a smoke image feature data set, and processing the time series image data to obtain a smoke dynamic change data set; fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set; training an instantaneous thick smoke prediction model, inputting the smoke comprehensive feature data set into the instantaneous thick smoke prediction model, and predicting a grade coefficient of instantaneous thick smoke; processing the predicted instantaneous thick smoke grade coefficient to obtain the instantaneous thick smoke danger degree; comparing the predicted instantaneous thick smoke danger degree with a preset instantaneous thick smoke danger degree threshold, and judging whether to generate fire warning information; effective fire warning and response measures can avoid potential safety accidents.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and more specifically, to a method and system for visually detecting instantaneous explosion smoke. Background Art

[0002] In some high-risk industries, such as chemical, mining, and military industries, sudden bursts of large amounts of thick smoke may occasionally occur due to equipment failures, improper operation, etc. Such abnormal thick smoke events usually indicate that there may be serious safety hazards, such as equipment damage, fire risks, etc. Therefore, timely detection and early warning of such events are crucial to effectively prevent major safety accidents.

[0003] Patent announcement number CN113505758B discloses a smoke warning system based on visual detection, which relates to the technical field of smoke warning. The present invention sets an image acquisition module, an image analysis module, a result judgment module, a parameter adjustment module and a warning determination module. When the image analysis module analyzes the captured image, the result judgment module determines whether to issue a warning based on the image texture complexity and grayscale value of the captured image, and when it is determined to issue a warning, the warning value of the smoke warning system is determined based on the image texture complexity and the standard image texture complexity, so as to ensure that a warning can be issued in time when smoke is generated, thereby improving the safety of the collection area. When the warning value is determined, the warning value is adjusted based on the comparison result of the image texture complexity with the texture complexity of the labeled image or the comparison result of the grayscale value with the grayscale value range of the standard image by the image analysis module, thereby improving the warning accuracy of the warning system and further ensuring the safety of the collection area.

[0004] Traditional smoke monitoring mainly relies on manual inspections and point-based sensor monitoring, which has the following main problems:

[0005] The lack of accurate processing of dense smoke image data and time series image data, and the failure to consider the impact of ambient light data on dense smoke image data and time series image data, lead to reduced recognition ability and accuracy of the instantaneous dense smoke prediction model, which in turn affects the accuracy of dense smoke warnings; inaccurate data processing and neglect of ambient light may cause the prediction model to be unable to correctly distinguish between normal conditions and instantaneous dense smoke phenomena, increase the risk of false alarms and missed alarms, and pose a challenge to fire prevention and response;

[0006] In the prior art, the hyperparameters of the model are insufficiently optimized, and the hyperparameters cannot be effectively adjusted to find the optimal performance instantaneous explosion smoke prediction model, resulting in limited prediction accuracy and generalization ability; it is impossible to accurately assess the hazard level of instantaneous explosion smoke, and the predicted instantaneous explosion smoke level coefficient cannot be processed and converted into the instantaneous explosion smoke hazard level. In terms of preventive measures and emergency preparedness, it may lead to a lack of effective prevention and timely response capabilities when facing actual fire situations; inaccurate hazard assessment and response measures may lead to improper handling of fire accidents, increase the risk of social public safety, and affect people’s lives and property safety.

[0007] In view of this, the present invention proposes a method and system for visual detection of instantaneous explosion smoke to solve the above problems. Summary of the invention

[0008] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for visually detecting instantaneous explosion smoke, comprising:

[0009] S1, collecting smoke image data, time series image data and ambient light data;

[0010] S2, processing the dense smoke image data to obtain a smoke image feature data set, and processing the time series image data to obtain a smoke dynamic change data set; fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set;

[0011] S3, training a transient explosion smoke prediction model, inputting the smoke comprehensive feature data set into the transient explosion smoke prediction model, and predicting the level coefficient of the transient explosion smoke;

[0012] S4, processing the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; comparing the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information;

[0013] S5. The smoke control terminal starts the smoke fire fighting command according to the generated fire warning information, and promptly handles and controls the instantaneous explosion smoke.

[0014] Furthermore, the thick smoke image data includes color distribution data, texture feature data and morphological feature data; the time series image data includes continuity image data, smoke change feature data between images and timestamp data; the ambient light data includes light intensity data, light type data and light change data.

[0015] Furthermore, the method of processing the thick smoke image data to obtain the smoke image feature data set includes:

[0016] Processing the dense smoke image data includes processing the color distribution data, texture feature data and morphological feature data;

[0017] Processing of color distribution data: selecting thick smoke image data, each pixel in the thick smoke image data is composed of three color channels: red, green and blue. The three color channels are mixed in different proportions, and the values ​​in the three color channels are integer values ​​between 0 and 255; obtaining the values ​​of the three color channels from the thick smoke image data and combining them together to form an RGB color space, and converting the RGB color space into a LAB color space;

[0018] The process of converting the RGB color space smoke image into the LAB color space smoke image is as follows: the nonlinear response of the three color channels is eliminated by gamma correction to make them linear; the value of each channel after correction is: ;in, is the gamma value, usually 2.2;

[0019] Use the standardized conversion matrix to convert the corrected RGB value to XYZ value; the conversion formula is: ;in, , and is the RGB value after gamma correction; are the coefficients in the transformation matrix;

[0020] Standardize the XYZ values ​​and preset the reference white point coordinates as ; The normalized value of XYZ is: ; Apply the function to the normalized values , ;in, is a known value, usually 0.008856; is the normalized X, Y or Z component; For When, through To simulate the visual nonlinear induction in the bright area of ​​the color channel; For When, through To deal with color perception discontinuities in dark areas of color channels;

[0021] The values ​​in LAB color space are calculated using the normalized and nonlinearly processed XYZ values; the calculation formula is: ;in, , and is the standardized XYZ value; is the brightness in the LAB color space; and is the color information in the LAB color space;

[0022] Processing of texture feature data:

[0023] The three channels L, A and B are denoted as , and ; For each channel, perform the following steps for each pixel in the image:

[0024] S31. For each pixel in the image, define it as the central pixel , which is in the channel The brightness value in , the position coordinates are ;

[0025] S32, around the center pixel ,choose Neighborhood pixels are equally spaced, and the neighborhood pixels are located at a distance of The radius is On the circumference, the neighborhood pixels are ;in, ;and and is the horizontal and vertical coordinates of the neighborhood pixel points;

[0026] For each neighborhood pixel, Compare brightness, the comparison formula is: ;in, is the area pixel and the center pixel Brightness comparison; If the brightness of the neighboring pixel is greater than or equal to the brightness of the central pixel, the output is 1; If the brightness of the neighboring pixel is less than that of the central pixel, the output is 0; For each field pixel;

[0027] S33, for the center pixel For each neighboring pixel point, the result obtained by applying the comparison formula is integrated into a binary number of one bit, and the calculation value;

[0028] The calculation formula is: ;in, The center pixel and a digital encoding of the texture information of the neighboring pixels; is the total number of neighborhood pixels selected in the neighborhood; The center pixel The distance to the neighboring pixels; is the number of points in the neighborhood; To convert the comparison result of each neighborhood pixel into a binary number;

[0029] For each channel Value, calculate the feature histogram: ;in, The histogram is Columns, To traverse all pixels in the image; and are the horizontal and vertical coordinates of the pixel in the image respectively; and are the height and width of the image respectively; is a function used to determine the coordinates Pixels Is the value equal to a specific value? ;

[0030] The feature histograms of L, A and B channels are integrated in series to form a texture feature dataset of the smoke image;

[0031] Processing of morphological feature data: grayscale and denoise preprocessing of the smoke image, binarization of the processed smoke image to separate the smoke object and background image area in the smoke image; skeleton representation of the smoke object in the smoke image is obtained through the skeleton extraction algorithm, and shape features of the smoke object are extracted based on the skeleton image;

[0032] The processed color distribution data, texture feature data and morphological feature data are integrated to form a smoke image feature data set.

[0033] Furthermore, the method of processing the time series image data to obtain the smoke dynamic change data set includes:

[0034] Timed shooting via surveillance camera The smoke video images in the area generate continuous image data; each image has a timestamp to record the shooting time; the collected smoke images are grayed and denoised by Gaussian filtering;

[0035] Select image frames at two time points t1 and t2; compare the two consecutive frame images t1 and t2 pixel by pixel, and calculate the difference between the two , the calculated difference value set of each pixel forms a new image, called the difference image;

[0036] A pixel gray value threshold is preset, and if the gray value of a pixel is greater than the preset pixel gray value threshold, the pixel is marked as a changed area;

[0037] If the gray value of a pixel is less than or equal to the preset pixel gray value threshold, the pixel is marked as a non-changing area;

[0038] The dynamic change area of ​​smoke is identified by threshold comparison; based on the difference value between frames, the characteristic data of smoke change between images are extracted; the characteristic data of smoke change between images include the increase or decrease of smoke area, smoke diffusion speed and direction;

[0039] Processing of smoke area increase and decrease: The smoke area in the preset difference image is: ;in, For at the moment The area of ​​the smoke zone; is the coordinate of the pixel in the difference image; is the area identified as smoke; It is the mathematical counting unit value represented by each pixel when calculating the area of ​​the smoke area;

[0040] The area change between two consecutive frames is: ;in, for The area of ​​the smoke zone at the time;

[0041] Processing of smoke diffusion speed: Preset two consecutive moments and The coordinates of the center of mass of the smoke are and ;

[0042] The smoke diffusion distance is: ; The smoke diffusion speed is: ;in, is the smoke diffusion speed; is the smoke diffusion distance;

[0043] Processing of smoke diffusion direction: The direction of smoke movement can be determined by calculating the direction angle of the displacement vector; the direction angle is: ;in, is the inverse tangent function; is the horizontal coordinate of the displacement vector; is the ordinate of the displacement vector;

[0044] The horizontal coordinate of the displacement vector is: ; The vertical coordinate of the displacement vector is: ;

[0045] The processed continuous image data, smoke change feature data between images and timestamp data are combined into a smoke dynamic change dataset.

[0046] Furthermore, the method of fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set includes:

[0047] The preset smoke image dataset is , the smoke dynamic change dataset is , the ambient light data is ,and , and Aligned in the time dimension; Predicted the smoke impact index through a multivariate linear regression model;

[0048] The multiple linear regression model is: ;in, is the predicted smoke impact index; It is a smoke image dataset; It is a dataset of smoke dynamic changes; is the ambient light data; , , and is the regression coefficient; are variables and random errors that the model cannot explain;

[0049] The objective function is constructed to measure the difference between the smoke impact index predicted by the multivariate linear regression model and the actual smoke impact index;

[0050] The objective function is: ;in, is the objective function; is the total number of data points; For the Smoke impact index for each data point; For the Smoke image dataset with data points; For the Smoke dynamic change dataset with data points; For the Ambient light data for data points;

[0051] By the objective function In , , and Find the partial derivative of the regression coefficient and set it to 0; ;

[0052] Solve the above equations with partial derivatives equal to 0 by operating the matrix to find , , and The estimated value of ; the operation matrix is: ;in, is a matrix containing all regression coefficients , , and A vector of is an operation matrix containing all , and The value of is a vector containing all , and The value of

[0053] The regression coefficient obtained by , , and The estimated value of each independent variable , and The degree of influence on the dependent variable smoke impact index, thus completing the construction of the smoke comprehensive characteristic data set;

[0054] The constructed smoke comprehensive feature dataset is normalized by standard deviation and converted into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimensional influence of the data and obtain the normalized smoke comprehensive feature dataset. The smoke comprehensive feature dataset is clustered using a density clustering algorithm to identify outliers in the smoke comprehensive feature dataset, and the identified outliers are removed to finally obtain the processed smoke comprehensive feature dataset.

[0055] Furthermore, the training method of the instantaneous explosion smoke prediction model includes:

[0056] The data set is divided into a training set, a validation set and a test set; a transient explosion smoke prediction model is constructed, and the transient explosion smoke prediction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; the input layer is a historical smoke comprehensive feature data set, and the output layer is a level coefficient of the transient explosion smoke; softmax is used as an activation function; the transient explosion smoke prediction model is a convolutional neural network model;

[0057] The mean square error is used as the loss function to measure the difference between the model's predicted value and the actual value; the mean square error loss function is: ;in, is the number of data sets; Data points in the dataset The actual value of Data points in the dataset The actual value of

[0058] The training set data is used to train the model, and the loss function is minimized through the Adam optimizer. The validation set is used to evaluate the performance of the health status prediction model, and the accuracy index is calculated to measure the performance of the model. The hyperparameters of the model are tuned to improve the model performance. The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the instantaneous explosion smoke prediction model.

[0059] Furthermore, the method of tuning the hyperparameters of the model to improve the model performance includes:

[0060] Define the objective optimal function is the accuracy of the model on the validation set, are model parameters; select an initial solution , as the starting point; set the initial temperature to , cooling rate ; Set stop temperature ,arrive Then stop the algorithm and define the maximum number of iterations as ;

[0061] In each iteration , perform the following steps:

[0062] S71, based on the current solution , randomly generate a new solution ;

[0063] S72. Calculate the accuracy of the new solution ;

[0064] S73, if , then the new solution is better and we accept it; if , then with probability Accept new interpretations; among them, is the current temperature;

[0065] S74, update current temperature ,use Formula, lower the temperature after each iteration and narrow the search range; is the temperature reduction factor;

[0066] S75, when the temperature Or reach the preset number of iterations The iteration ends when

[0067] From all accepted solutions, select the parameter combination with the highest accuracy And its corresponding accuracy As the final result.

[0068] Furthermore, the method of processing the predicted instantaneous explosion smoke level coefficient to obtain the predicted instantaneous explosion smoke danger level includes:

[0069] The grade coefficient of instantaneous explosion smoke is evaluated by the grade coefficient calculation formula, which is: ;in, is the level coefficient of instantaneous explosion smoke; is the smoke diffusion speed; is the amount of dust in the smoke; The temperature of the instantaneous explosion smoke itself;

[0070] The speed at which smoke spreads It is described by the Maxwell-Boltzmann distribution, which is: ;in, Smoke diffusion speed The molecular probability density of is the molecular mass; is the Boltzmann constant; The temperature of the instantaneous explosion smoke itself;

[0071] The predicted instantaneous explosion smoke level coefficients are divided into K clusters of different levels through the k-means clustering algorithm, and different instantaneous explosion smoke danger levels are set based on clusters with different level coefficients.

[0072] Furthermore, the method of comparing the predicted instantaneous explosion smoke danger level with a preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information includes:

[0073] If the predicted instantaneous explosion smoke danger level is less than or equal to the preset instantaneous explosion smoke danger level threshold, it is determined that no fire warning information is generated;

[0074] If the predicted instantaneous explosion smoke danger level is greater than the preset instantaneous explosion smoke danger level threshold, it is determined to generate fire warning information.

[0075] A visual detection system for instantaneous explosion and dense smoke, comprising:

[0076] A data acquisition module is used to collect smoke image data, time series image data and ambient light data;

[0077] The data processing module is used to process the dense smoke image data to obtain a smoke image feature data set, and to process the time series image data to obtain a smoke dynamic change data set; and to fuse the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set;

[0078] A model training module is used to train the instantaneous explosion smoke prediction model, input the smoke comprehensive feature data set into the instantaneous explosion smoke prediction model, and predict the instantaneous explosion smoke level coefficient;

[0079] The smoke warning module is used to process the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; compare the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information;

[0080] The smoke control module and the smoke control terminal initiate the smoke fire fighting command according to the generated fire warning information, and promptly handle and control the instantaneous explosion smoke.

[0081] The technical effects and advantages of the instantaneous explosion smoke visual detection method and system of the present invention are as follows:

[0082] By accurately processing the dense smoke image data and time series image data and considering the influence of the ambient light data on the dense smoke image data and time series image data, the ambient light data is integrated into the smoke image feature data set and the smoke dynamic change data set to form a comprehensive smoke feature data set, thereby improving the overall accuracy of smoke detection and prediction. Considering the ambient light as a feature dimension can make the instantaneous dense smoke prediction model more robust, and can still maintain good recognition and prediction performance in the face of changing ambient light conditions. The recognition ability and accuracy of the instantaneous dense smoke prediction model are improved, thereby improving the accuracy of dense smoke warning.

[0083] By simulating the heating and slow cooling of the material during the annealing process to solve the optimization problem, it is possible to explore in a larger search space, which helps to avoid falling into the local optimal solution and thus find the global optimal solution; through careful parameter adjustment, the best hyperparameter combination is found, so that the model achieves a higher accuracy on the validation set, thereby achieving better performance in the model learning task; the level coefficient calculation formula can be used to quantitatively evaluate the degree of danger of instantaneous smoke explosions; using the Maxwell-Boltzmann distribution to describe the smoke diffusion speed can more accurately reflect the actual situation of smoke diffusion, thereby improving the calculation accuracy of the instantaneous smoke explosion level coefficient; through the k-means clustering algorithm, the instantaneous smoke explosion level coefficient can be accurately predicted and effectively classified, which can enhance the prevention and response capabilities of instantaneous smoke explosions, promote the scientific and refined safety management and disaster prevention, and facilitate the timely discovery and warning of potential dangerous situations, thereby reducing the risk of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A schematic diagram of a process flow of a method for visually detecting instantaneous explosion smoke according to the present invention;

[0085] Figure 2 The present invention is a schematic structural diagram of a visual detection system for instantaneous explosion and dense smoke. DETAILED DESCRIPTION

[0086] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example 1

[0087] See also Figure 1 and Figure 2 As shown, this embodiment provides a method for visually detecting instantaneous explosion smoke, including:

[0088] S1, collecting smoke image data, time series image data and ambient light data;

[0089] S2, processing the dense smoke image data to obtain a smoke image feature data set, and processing the time series image data to obtain a smoke dynamic change data set; fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set;

[0090] S3, training a transient explosion smoke prediction model, inputting the smoke comprehensive feature data set into the transient explosion smoke prediction model, and predicting the level coefficient of the transient explosion smoke;

[0091] S4, processing the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; comparing the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information;

[0092] S5. The smoke control terminal starts the smoke fire fighting command according to the generated fire warning information, and promptly handles and controls the instantaneous explosion smoke.

[0093] Furthermore, the thick smoke image data includes color distribution data, texture feature data and morphological feature data; the time series image data includes continuity image data, smoke change feature data between images and timestamp data; the ambient light data includes light intensity data, light type data and light change data.

[0094] The method for processing the thick smoke image data to obtain the smoke image feature data set includes:

[0095] Processing the dense smoke image data includes processing the color distribution data, texture feature data and morphological feature data;

[0096] Processing of color distribution data: selecting thick smoke image data, each pixel in the thick smoke image data is composed of three color channels: red, green and blue. The three color channels are mixed in different proportions, and the values ​​in the three color channels are integer values ​​between 0 and 255; obtaining the values ​​of the three color channels from the thick smoke image data and combining them together to form an RGB color space, and converting the RGB color space into a LAB color space;

[0097] The process of converting the RGB color space smoke image into the LAB color space smoke image is as follows: the nonlinear response of the three color channels is eliminated by gamma correction to make them linear; the value of each channel after correction is: ;in, is the gamma value, usually 2.2;

[0098] Use the standardized conversion matrix to convert the corrected RGB value to XYZ value; the conversion formula is: ;in, , and is the RGB value after gamma correction; are the coefficients in the transformation matrix;

[0099] Standardize the XYZ values ​​and preset the reference white point coordinates as ; The normalized value of XYZ is: ; Apply the function to the normalized values , ;in, is a known value, usually 0.008856; is the normalized X, Y or Z component; For When, through To simulate the visual nonlinear induction in the bright area of ​​the color channel; For When, through To deal with color perception discontinuities in dark areas of color channels;

[0100] The values ​​in LAB color space are calculated using the normalized and nonlinearly processed XYZ values; the calculation formula is: ;in, , and is the standardized XYZ value; is the brightness in the LAB color space; and is the color information in the LAB color space;

[0101] Processing of texture feature data:

[0102] The three channels L, A and B are denoted as , and ; For each channel, perform the following steps for each pixel in the image:

[0103] S31. For each pixel in the image, define it as the central pixel , which is in the channel The brightness value in , the position coordinates are ;

[0104] S32, around the center pixel ,choose Neighborhood pixels are equally spaced, and the neighborhood pixels are located at a distance of The radius is On the circumference, the neighborhood pixels are ;in, ;and and is the horizontal and vertical coordinates of the neighborhood pixel points;

[0105] For each neighborhood pixel, Compare brightness, the comparison formula is: ;in, is the area pixel and the center pixel Brightness comparison; If the brightness of the neighboring pixel is greater than or equal to the brightness of the central pixel, the output is 1; If the brightness of the neighboring pixel is less than that of the central pixel, the output is 0; For each field pixel;

[0106] S33, for the center pixel For each neighboring pixel point, the result obtained by applying the comparison formula is integrated into a binary number of one bit, and the calculation value;

[0107] The calculation formula is: ;in, The center pixel and a digital encoding of the texture information of the neighboring pixels; is the total number of neighborhood pixels selected in the neighborhood; The center pixel The distance to the neighboring pixels; is the number of points in the neighborhood; To convert the comparison result of each neighborhood pixel into a binary number;

[0108] For each channel Value, calculate the feature histogram: ;in, The histogram is Columns, To traverse all pixels in the image; and are the horizontal and vertical coordinates of the pixel in the image respectively; and are the height and width of the image respectively; is a function used to determine the coordinates Pixels Is the value equal to a specific value? ;

[0109] The feature histograms of L, A and B channels are integrated in series to form a texture feature dataset of the smoke image;

[0110] Processing of morphological feature data: grayscale and denoise preprocessing of the smoke image, binarization of the processed smoke image to separate the smoke object and background image area in the smoke image; skeleton representation of the smoke object in the smoke image is obtained through the skeleton extraction algorithm, and shape features of the smoke object are extracted based on the skeleton image;

[0111] The processed color distribution data, texture feature data and morphological feature data are integrated to form a smoke image feature data set.

[0112] The method for processing the time series image data to obtain the smoke dynamic change data set includes:

[0113] Timed shooting via surveillance camera The smoke video images in the area generate continuous image data; each image has a timestamp to record the shooting time; the collected smoke images are grayed and denoised by Gaussian filtering;

[0114] Select image frames at two time points t1 and t2; compare the two consecutive frame images t1 and t2 pixel by pixel, and calculate the difference between the two , the calculated difference value set of each pixel forms a new image, called the difference image;

[0115] A pixel gray value threshold is preset, and if the gray value of a pixel is greater than the preset pixel gray value threshold, the pixel is marked as a changed area;

[0116] If the gray value of a pixel is less than or equal to the preset pixel gray value threshold, the pixel is marked as a non-changing area;

[0117] The dynamic change area of ​​smoke is identified by threshold comparison; based on the difference value between frames, the characteristic data of smoke change between images are extracted; the characteristic data of smoke change between images include the increase or decrease of smoke area, smoke diffusion speed and direction;

[0118] Processing of smoke area increase and decrease: The smoke area in the preset difference image is: ;in, For at the moment The area of ​​the smoke zone; is the coordinate of the pixel in the difference image; is the area identified as smoke; It is the mathematical counting unit value represented by each pixel when calculating the area of ​​the smoke area;

[0119] The area change between two consecutive frames is: ;in, for The area of ​​the smoke zone at the time;

[0120] Processing of smoke diffusion speed: Preset two consecutive moments and The coordinates of the center of mass of the smoke are and ;

[0121] The smoke diffusion distance is: ; The smoke diffusion speed is: ;in, is the smoke diffusion speed; is the smoke diffusion distance;

[0122] Processing of smoke diffusion direction: The direction of smoke movement can be determined by calculating the direction angle of the displacement vector; the direction angle is: ;in, is the inverse tangent function; is the horizontal coordinate of the displacement vector; is the ordinate of the displacement vector;

[0123] The horizontal coordinate of the displacement vector is: ; The vertical coordinate of the displacement vector is: ;

[0124] For example, select the image frames of t1 and t2, and compare the two frames pixel by pixel to calculate the difference between them; generate a new difference image according to the difference, in which the pixels with high gray values ​​represent the areas with large changes;

[0125] Assume that the preset pixel gray value threshold is 200; in the difference image, pixels with gray values ​​greater than 200 are marked as smoke change areas, while pixels with gray values ​​less than or equal to 200 are marked as non-change areas;

[0126] Assume that the area of ​​the smoke region at time t1 is 80 square meters, and the area of ​​the smoke region at time t2 is 120 square meters; this means that the smoke area has increased by 40 square meters; the coordinates of the smoke center of mass are (100, 150) at time t1 and become (110, 160) at time t2; based on the change in the position of the smoke center of mass, we can calculate the smoke diffusion distance as: meters; if the time difference between t1 and t2 is 2 minutes, then the smoke diffusion speed is m / 2min;

[0127] The horizontal coordinate of the displacement vector changes to 110-100=10 meters, and the vertical coordinate of the displacement vector changes to 160-150=10 meters; the direction angle of the smoke movement direction is , which means the smoke is spreading in a northeasterly direction.

[0128] The processed continuous image data, smoke change feature data between images and timestamp data are combined into a smoke dynamic change dataset.

[0129] The method of fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set includes:

[0130] The preset smoke image dataset is , the smoke dynamic change dataset is , the ambient light data is ,and , and Aligned in the time dimension; Predicted the smoke impact index through a multivariate linear regression model;

[0131] The multiple linear regression model is: ;in, is the predicted smoke impact index; It is a smoke image dataset; It is a dataset of smoke dynamic changes; is the ambient light data; , , and is the regression coefficient; are variables and random errors that the model cannot explain;

[0132] The objective function is constructed to measure the difference between the smoke impact index predicted by the multivariate linear regression model and the actual smoke impact index;

[0133] The objective function is: ;in, is the objective function; is the total number of data points; For the Smoke impact index for each data point; For the Smoke image dataset with data points; For the Smoke dynamic change dataset with data points; For the Ambient light data for data points;

[0134] By the objective function In , , and Find the partial derivative of the regression coefficient and set it to 0; ;

[0135] Solve the above equations with partial derivatives equal to 0 by operating the matrix to find , , and The estimated value of ; the operation matrix is: ;in, is a matrix containing all regression coefficients , , and A vector of is an operation matrix containing all , and The value of is a vector containing all , and The value of

[0136] The regression coefficient obtained by , , and The estimated value of each independent variable , and The degree of influence on the dependent variable smoke impact index, thus completing the construction of the smoke comprehensive characteristic data set;

[0137] The constructed smoke comprehensive feature dataset is normalized by standard deviation and converted into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimensional influence of the data and obtain the normalized smoke comprehensive feature dataset. The smoke comprehensive feature dataset is clustered using a density clustering algorithm to identify outliers in the smoke comprehensive feature dataset, and the identified outliers are removed to finally obtain the processed smoke comprehensive feature dataset.

[0138] The training method of the instantaneous explosion smoke prediction model includes:

[0139] The data set is divided into a training set, a validation set and a test set; a transient explosion smoke prediction model is constructed, and the transient explosion smoke prediction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; the input layer is a historical smoke comprehensive feature data set, and the output layer is a level coefficient of the transient explosion smoke; softmax is used as an activation function; the transient explosion smoke prediction model is a convolutional neural network model;

[0140] The mean square error is used as the loss function to measure the difference between the model's predicted value and the actual value; the mean square error loss function is: ;in, is the number of data sets; Data points in the dataset The actual value of Data points in the dataset The actual value of

[0141] The training set data is used to train the model, and the loss function is minimized through the Adam optimizer. The validation set is used to evaluate the performance of the health status prediction model, and the accuracy index is calculated to measure the performance of the model. The hyperparameters of the model are tuned to improve the model performance. The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the instantaneous explosion smoke prediction model.

[0142] The method of tuning the hyperparameters of the model to improve the model performance includes:

[0143] Define the objective optimal function is the accuracy of the model on the validation set, are model parameters; select an initial solution , as the starting point; set the initial temperature to , cooling rate ; Set stop temperature ,arrive Then stop the algorithm and define the maximum number of iterations as ;

[0144] In each iteration , perform the following steps:

[0145] S71, based on the current solution , randomly generate a new solution ;

[0146] S72. Calculate the accuracy of the new solution ;

[0147] S73, if , then the new solution is better and we accept it; if , then with probability Accept new interpretations; among them, is the current temperature;

[0148] S74, update current temperature ,use Formula, lower the temperature after each iteration and narrow the search range; is the temperature reduction factor;

[0149] S75, when the temperature Or reach the preset number of iterations The iteration ends when

[0150] From all accepted solutions, select the parameter combination with the highest accuracy And its corresponding accuracy As the final result.

[0151] For example, suppose there are two parameters to be optimized, denoted as z1 and z2, and the initial conditions are set as follows: =(z1,z2)=(0.5,0.5) as the starting point.

[0152] Initial temperature =1000℃, cooling rate is 0.95; stop temperature is 1℃, and the maximum number of iterations is set to 1000;

[0153] Assume the first iteration: Based on the current solution, randomly generate a new solution = (0.55, 0.45); Calculate the accuracy of the new solution: Assume that the accuracy of the new solution is 75%, and the accuracy of the current solution is 70%;

[0154] The decision to accept the new solution is that 75%>70%, so the new solution is better. Accept the new solution, that is, now = (0.55, 0.45); Update current temperature =1000×0.95=950; continue to the next iteration until the temperature drops to 1 or the number of iterations reaches 1000;

[0155] Assume that a solution is found before reaching the stopping temperature 1 or the maximum number of iterations 1000 = (0.65, 0.35) Its accuracy rate reaches the highest, assuming it is 80%. After searching and optimizing, the parameter combination with the highest accuracy rate is finally selected. = (0.65, 0.35) as the final result.

[0156] The method for processing the predicted instantaneous explosion smoke level coefficient to obtain the predicted instantaneous explosion smoke danger level includes:

[0157] The grade coefficient of instantaneous explosion smoke is evaluated by the grade coefficient calculation formula, which is: ;in, is the level coefficient of instantaneous explosion smoke; is the smoke diffusion speed; is the amount of dust in the smoke; The temperature of the instantaneous explosion smoke itself;

[0158] The speed at which smoke spreads It is described by the Maxwell-Boltzmann distribution, which is: ;in, Smoke diffusion speed The molecular probability density of is the molecular mass; is the Boltzmann constant; The temperature of the instantaneous explosion smoke itself;

[0159] For example, suppose a sudden explosion smoke event needs to be evaluated for its level coefficient, and the given data is as follows:

[0160] Because the smoke diffusion speed has been calculated in the processing of smoke diffusion speed for m / 2min; preset dust quantity in smoke is 0.5 g / m3; the temperature of the instantaneous explosion smoke itself is 1000 Kelvin; then according to the calculation formula of the grade coefficient, the grade coefficient of the instantaneous explosion smoke is about 3.47;

[0161] Smoke diffusion speed The higher the molecular probability density, the larger the distribution range of molecular motion; at the same time, the temperature of the instantaneous explosion smoke increases, which increases the average kinetic energy of the molecules in the smoke, and the collisions between molecules are more frequent, which promotes the diffusion of smoke faster, that is, the level coefficient of the instantaneous explosion smoke is higher;

[0162] The predicted instantaneous explosion smoke level coefficients are divided into K clusters of different levels through the k-means clustering algorithm, and different instantaneous explosion smoke danger levels are set based on clusters with different level coefficients.

[0163] The method of comparing the predicted instantaneous explosion smoke danger level with a preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information includes:

[0164] If the predicted instantaneous explosion smoke danger level is less than or equal to the preset instantaneous explosion smoke danger level threshold, it is determined that no fire warning information is generated;

[0165] If the predicted instantaneous explosion smoke danger level is greater than the preset instantaneous explosion smoke danger level threshold, it is determined to generate fire warning information;

[0166] The preset instantaneous explosion smoke danger level threshold is set by the staff, and different instantaneous explosion smoke danger levels are collected through the instantaneous explosion smoke intelligent management terminal, and the average value of multiple instantaneous explosion smoke danger levels is taken as the preset instantaneous explosion smoke danger level threshold; the preset performance threshold is set in the same way;

[0167] In this embodiment, by accurately processing the dense smoke image data and the time series image data, and considering the influence of the ambient light data on the dense smoke image data and the time series image data, the ambient light data is integrated into the smoke image feature data set and the smoke dynamic change data set to form a comprehensive smoke feature data set, thereby improving the overall accuracy of smoke detection and prediction; considering the ambient light as a feature dimension can make the instantaneous dense smoke prediction model more robust, and can still maintain good recognition and prediction performance when facing variable ambient light conditions, the recognition ability and accuracy of the instantaneous dense smoke prediction model are improved, and the accuracy of dense smoke warning is improved;

[0168] By simulating the heating and slow cooling of the material during the annealing process to solve the optimization problem, it is possible to explore in a larger search space, which helps to avoid falling into the local optimal solution and thus find the global optimal solution; through careful parameter adjustment, the best hyperparameter combination is found, so that the model achieves a higher accuracy on the validation set, thereby achieving better performance in the model learning task; the level coefficient calculation formula can be used to quantitatively evaluate the degree of danger of instantaneous smoke explosions; using the Maxwell-Boltzmann distribution to describe the smoke diffusion speed can more accurately reflect the actual situation of smoke diffusion, thereby improving the calculation accuracy of the instantaneous smoke explosion level coefficient; through the k-means clustering algorithm, the instantaneous smoke explosion level coefficient can be accurately predicted and effectively classified, which can enhance the prevention and response capabilities of instantaneous smoke explosions, promote the scientific and refined safety management and disaster prevention, and facilitate the timely discovery and warning of potential dangerous situations, thereby reducing the risk of accidents. Example 2

[0169] See also Figure 1 As shown, this embodiment provides a visual detection system for instantaneous explosion and dense smoke, including:

[0170] A data acquisition module is used to collect smoke image data, time series image data and ambient light data;

[0171] The data processing module is used to process the dense smoke image data to obtain a smoke image feature data set, and to process the time series image data to obtain a smoke dynamic change data set; and to fuse the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set;

[0172] A model training module is used to train the instantaneous explosion smoke prediction model, input the smoke comprehensive feature data set into the instantaneous explosion smoke prediction model, and predict the instantaneous explosion smoke level coefficient;

[0173] The smoke warning module is used to process the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; compare the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information;

[0174] The smoke control module and the smoke control terminal initiate the smoke fire fighting command according to the generated fire warning information, and promptly handle and control the instantaneous explosion smoke.

[0175] Since the electronic device introduced in this embodiment is an electronic device used to implement a method and system for visually detecting instantaneous thick smoke in the embodiment of this application, based on the method and system for visually detecting instantaneous thick smoke in the embodiment of this application, a person skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as a person skilled in the art implements the electronic device used in a method and system for visually detecting instantaneous thick smoke in the embodiment of this application, it belongs to the scope of protection of this application.

[0176] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0177] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A method for visual detection of instantaneous explosion smoke, characterized in that: include: S1, collecting smoke image data, time series image data and ambient light data; S2, processing the dense smoke image data to obtain a smoke image feature data set, and processing the time series image data to obtain a smoke dynamic change data set; fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set; S3, training a transient explosion smoke prediction model, inputting the smoke comprehensive feature data set into the transient explosion smoke prediction model, and predicting the level coefficient of the transient explosion smoke; S4, processing the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; comparing the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information; S5. The smoke control terminal starts the smoke fire command according to the generated fire warning information, and promptly handles and controls the generated instantaneous explosion smoke; The thick smoke image data includes color distribution data, texture feature data and morphological feature data; the time series image data includes continuous image data, smoke change feature data between images and timestamp data; the ambient light data includes light intensity data, light type data and light change data; The method for processing the thick smoke image data to obtain the smoke image feature data set includes: Processing the dense smoke image data includes processing the color distribution data, texture feature data and morphological feature data; Processing of color distribution data: selecting thick smoke image data, each pixel in the thick smoke image data is composed of three color channels: red, green and blue. The three color channels are mixed in different proportions, and the values ​​in the three color channels are integer values ​​between 0 and 255; obtaining the values ​​of the three color channels from the thick smoke image data and combining them together to form an RGB color space, and converting the RGB color space into a LAB color space; The process of converting the RGB color space smoke image into the LAB color space smoke image is as follows: the nonlinear response of the three color channels is eliminated by gamma correction to make them linear; the value of each channel after correction is: ;in, is the gamma value, usually 2.2; Use the standardized conversion matrix to convert the corrected RGB value to XYZ value; the conversion formula is: ;in, , and is the RGB value after gamma correction; are the coefficients in the transformation matrix; Standardize the XYZ values ​​and preset the reference white point coordinates as ; The normalized value of XYZ is: ; Apply the function to the normalized values , ;in, is a known value, usually 0.008856; is the normalized X, Y or Z component; For When, through To simulate the visual nonlinear induction in the bright area of ​​the color channel; For When, through To deal with color perception discontinuities in dark areas of color channels; The values ​​in the LAB color space are calculated using the normalized and nonlinearly processed XYZ values; the calculation formula is: ;in, , and is the standardized XYZ value; is the brightness in the LAB color space; and is the color information in the LAB color space; Processing of texture feature data: The three channels L, A and B are denoted as , and ; For each channel, perform the following steps for each pixel in the image: S31. For each pixel in the image, define it as the central pixel , which is in the channel The brightness value in , the position coordinates are ; S32, around the center pixel ,choose Neighborhood pixels are equally spaced, and the neighborhood pixels are located at a distance of The radius is On the circumference, the neighborhood pixels are ;in, ;and and is the horizontal and vertical coordinates of the neighborhood pixel points; For each neighborhood pixel, Compare brightness, the comparison formula is: ;in, is the area pixel and the center pixel Brightness comparison; If the brightness of the neighboring pixel is greater than or equal to the brightness of the central pixel, the output is 1; If the brightness of the neighboring pixel is less than that of the central pixel, the output is 0; For each field pixel; S33, for the center pixel For each neighboring pixel point, the result obtained by applying the comparison formula is integrated into a binary number of one bit, and the calculation value; The calculation formula is: ;in, The center pixel and a digital encoding of the texture information of the neighboring pixels; is the total number of neighborhood pixels selected in the neighborhood; The center pixel The distance to the neighboring pixels; is the number of points in the neighborhood; To convert the comparison result of each neighborhood pixel into a binary number; For each channel Value, calculate the feature histogram: ;in, The histogram is Columns, To traverse all pixels in the image; and are the horizontal and vertical coordinates of the pixel in the image respectively; and are the height and width of the image respectively; is a function used to determine the coordinates Pixels Is the value equal to a specific value? ; The feature histograms of L, A and B channels are integrated in series to form a texture feature dataset of the smoke image; Processing of morphological feature data: grayscale and denoise preprocessing of the smoke image, binarization of the processed smoke image to separate the smoke object and background image area in the smoke image; skeleton representation of the smoke object in the smoke image is obtained through the skeleton extraction algorithm, and shape features of the smoke object are extracted based on the skeleton image; Integrate the processed color distribution data, texture feature data and morphological feature data to form a smoke image feature data set; The method for processing the time series image data to obtain the smoke dynamic change data set includes: Timed shooting via surveillance camera The smoke video images in the area generate continuous image data; each image has a timestamp to record the shooting time; the collected smoke images are grayed and denoised by Gaussian filtering; Select image frames at two time points t1 and t2; compare the two consecutive frame images t1 and t2 pixel by pixel, and calculate the difference between the two The calculated difference value set of each pixel forms a new image, called the difference image; A pixel gray value threshold is preset, and if the gray value of a pixel is greater than the preset pixel gray value threshold, the pixel is marked as a changed area; If the gray value of a pixel is less than or equal to the preset pixel gray value threshold, the pixel is marked as a non-changing area; The dynamic change area of ​​smoke is identified by threshold comparison; based on the difference value between frames, the characteristic data of smoke change between images are extracted; the characteristic data of smoke change between images include the increase or decrease of smoke area, smoke diffusion speed and direction; Processing of smoke area increase and decrease: The smoke area in the preset difference image is: ;in, For at the moment The area of ​​the smoke zone; is the coordinate of the pixel in the difference image; is the area identified as smoke; It is the mathematical counting unit value represented by each pixel when calculating the area of ​​the smoke area; The area change between two consecutive frames is: ;in, for The area of ​​the smoke zone at the time; Processing of smoke diffusion speed: Preset two consecutive moments and The coordinates of the smoke center of mass at and ; The smoke diffusion distance is: ; The smoke diffusion speed is: ;in, is the smoke diffusion speed; is the smoke diffusion distance; Processing of smoke diffusion direction: The direction of smoke movement can be determined by calculating the direction angle of the displacement vector; the direction angle is: ;in, is the inverse tangent function; is the horizontal coordinate of the displacement vector; is the ordinate of the displacement vector; The horizontal coordinate of the displacement vector is: ; The vertical coordinate of the displacement vector is: ; The processed continuous image data, the smoke change feature data between images and the timestamp data are combined into a smoke dynamic change data set; The method of fusing the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set includes: The preset smoke image dataset is , the smoke dynamic change dataset is , the ambient light data is ,and , and Aligned in the time dimension; Predicted the smoke impact index through a multivariate linear regression model; The multiple linear regression model is: ;in, is the predicted smoke impact index; It is a smoke image dataset; It is a dataset of dynamic changes of smoke; is the ambient light data; , , and is the regression coefficient; are variables and random errors that the model cannot explain; The objective function is constructed to measure the difference between the smoke impact index predicted by the multivariate linear regression model and the actual smoke impact index; The objective function is: ;in, is the objective function; is the total number of data points; For the Smoke impact index for each data point; For the A smoke image dataset with data points; For the Smoke dynamic change dataset with data points; For the Ambient light data for data points; By the objective function In , , and Find the partial derivative of the regression coefficient and set it to 0; ; Solve the above equations with partial derivatives equal to 0 by operating the matrix to find , , and The estimated value of ; the operation matrix is: ;in, is a matrix containing all regression coefficients , , and A vector of is an operation matrix containing all , and The value of is a vector containing all , and The value of The regression coefficient obtained by , , and The estimated value of each independent variable , and The degree of influence on the dependent variable smoke impact index, thus completing the construction of the smoke comprehensive characteristic data set; The constructed smoke comprehensive feature dataset is normalized by standard deviation and converted into a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimensional influence of the data and obtain the normalized smoke comprehensive feature dataset. The smoke comprehensive feature dataset is clustered using a density clustering algorithm to identify outliers in the smoke comprehensive feature dataset, and the identified outliers are removed to finally obtain the processed smoke comprehensive feature dataset.

2. A method for visually detecting instantaneous explosion smoke according to claim 1, characterized in that: The training method of the instantaneous explosion smoke prediction model includes: The data set is divided into a training set, a validation set and a test set; a transient explosion smoke prediction model is constructed, and the transient explosion smoke prediction model includes an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer; the input layer is a historical smoke comprehensive feature data set, and the output layer is a level coefficient of the transient explosion smoke; softmax is used as an activation function; the transient explosion smoke prediction model is a convolutional neural network model; The mean square error is used as the loss function to measure the difference between the model's predicted value and the actual value; the mean square error loss function is: ;in, is the number of data sets; Data points in the dataset The actual value of Data points in the dataset The actual value of The training set data is used to train the model, and the loss function is minimized through the Adam optimizer. The validation set is used to evaluate the performance of the health status prediction model, and the accuracy index is calculated to measure the performance of the model. The hyperparameters of the model are tuned to improve the model performance. The performance of the model in the prediction task is evaluated through the test set. When the performance of the model in the prediction task reaches the preset performance threshold, the test is stopped to obtain the instantaneous explosion smoke prediction model.

3. A method for visually detecting instantaneous explosion smoke according to claim 2, characterized in that: The method of tuning the hyperparameters of the model to improve the model performance includes: Define the objective optimal function is the accuracy of the model on the validation set, are model parameters; select an initial solution , as the starting point; set the initial temperature to , cooling rate ; Set stop temperature ,arrive Then stop the algorithm and define the maximum number of iterations as ; In each iteration , perform the following steps: S71, based on the current solution , randomly generate a new solution ; S72. Calculate the accuracy of the new solution ; S73, if , then the new solution is better and we accept it; if , then with probability Accept new interpretations; among them, is the current temperature; S74, update current temperature ,use Formula, lower the temperature after each iteration and narrow the search range; is the temperature reduction factor; S75, when the temperature Or reach the preset number of iterations The iteration ends when From all accepted solutions, select the parameter combination with the highest accuracy And its corresponding accuracy As the final result.

4. A method for visually detecting instantaneous explosion smoke according to claim 3, characterized in that: The method for processing the predicted instantaneous explosion smoke level coefficient to obtain the predicted instantaneous explosion smoke danger level includes: The grade coefficient of instantaneous explosion smoke is evaluated by the grade coefficient calculation formula, which is: ;in, is the level coefficient of instantaneous explosion smoke; is the smoke diffusion speed; is the amount of dust in the smoke; The temperature of the instantaneous explosion smoke itself; The speed at which smoke spreads It is described by the Maxwell-Boltzmann distribution, which is: ;in, Smoke diffusion speed The molecular probability density of is the molecular mass; is the Boltzmann constant; The temperature of the instantaneous explosion smoke itself; The predicted instantaneous explosion smoke level coefficients are divided into K clusters of different levels through the k-means clustering algorithm, and different instantaneous explosion smoke danger levels are set based on clusters with different level coefficients.

5. A method for visually detecting instantaneous explosion smoke according to claim 4, characterized in that: The method of comparing the predicted instantaneous explosion smoke danger level with a preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information includes: If the predicted instantaneous explosion smoke danger level is less than or equal to the preset instantaneous explosion smoke danger level threshold, it is determined that no fire warning information is generated; If the predicted instantaneous explosion smoke danger level is greater than the preset instantaneous explosion smoke danger level threshold, it is determined to generate fire warning information.

6. A visual detection system for instantaneous explosion and dense smoke, used to implement a visual detection method for instantaneous explosion and dense smoke according to any one of claims 1 to 5, characterized in that: include: Data acquisition module, used to collect smoke image data, time series image data and ambient light data; The data processing module is used to process the dense smoke image data to obtain a smoke image feature data set, and to process the time series image data to obtain a smoke dynamic change data set; and to fuse the ambient light data into the smoke image feature data set and the smoke dynamic change data set to form a smoke comprehensive feature data set; A model training module is used to train the instantaneous explosion smoke prediction model, input the smoke comprehensive feature data set into the instantaneous explosion smoke prediction model, and predict the instantaneous explosion smoke level coefficient; The smoke warning module is used to process the predicted instantaneous explosion smoke level coefficient to obtain the instantaneous explosion smoke danger level; compare the predicted instantaneous explosion smoke danger level with the preset instantaneous explosion smoke danger level threshold to determine whether to generate fire warning information; The smoke control module and the smoke control terminal initiate the smoke fire fighting command according to the generated fire warning information, and promptly handle and control the instantaneous explosion smoke.

Images