Image recognition method and system in the field of industrial safety based on AI big model

Abstract

The present invention discloses an image recognition method and system in the field of industrial safety based on an AI large model, and specifically relates to the technical field of industrial safety image recognition. The method and system collect a large number of images and video data samples related to factory smoke emission, classify the samples according to different environmental conditions, extract the color features and pixel fluctuation features of the sample images in each classified sample set for analysis, establish a color standard feature vector and a real-time time series pixel fluctuation standard feature vector, collect smoke images during factory production in real time, analyze the color features and pixel fluctuation features of the real-time images, establish a real-time color feature vector and a real-time time series pixel fluctuation feature vector, analyze the degree of deviation between the standard feature vector and the real-time feature vector, evaluate the degree of abnormal deviation of smoke emission during factory production, and issue an early warning for abnormal deviation of smoke emission during factory production.

Description

Technical Field

[0001] The present invention relates to the field of industrial safety image recognition technology, and more specifically, to an image recognition method and system in the field of industrial safety based on an AI big model. Background Art

[0002] In traditional industrial production, the early warning method for production safety is generally through on-site inspections or regular equipment maintenance. This method of early warning is not timely enough. If a production problem occurs between two inspections, personnel cannot detect it in time. In order to increase the timeliness of the early warning, sensor early warning technology is usually used in the production process, that is, production safety monitoring is carried out by detecting temperature, gas concentration, pressure, etc. However, in some factory production, the production environment may be carried out under extreme environmental conditions, resulting in problems such as sensor instability or failure under ultra-high temperature or ultra-high pressure production conditions.

[0003] How to use a solution that takes into account the timeliness of early warning and avoids the limited functionality of monitoring equipment under extreme production conditions is a technical problem that needs to be solved urgently.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an image recognition method and system in the field of industrial safety based on an AI big model to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Collect images and videos related to factory smoke emissions, find the ones with clear images as samples, classify the samples according to different environmental conditions, and establish a sample set feature vector hash table corresponding to different environmental conditions;

[0008] Process the sample smoke image according to the color feature processing flow, analyze the color features of the sample smoke image, and establish a color standard feature vector;

[0009] Process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the sample smoke image, and establish a time-series pixel fluctuation standard feature vector;

[0010] Collect real-time smoke images during factory operation for analysis, process the real-time smoke images according to the color feature processing flow, establish the real-time color feature vector of smoke, process the real-time smoke images according to the pixel fluctuation feature processing flow, and establish the real-time time-series pixel fluctuation feature vector of smoke;

[0011] Comprehensively analyze the standard feature vector and real-time feature vector, calculate the smoke feature deviation coefficient, evaluate the abnormal deviation degree of factory production smoke emissions, and issue early warning for abnormal factory production smoke emissions.

[0012] In a preferred embodiment, the classifying of samples according to different environmental conditions and establishing a sample set feature vector hash table corresponding to different environmental conditions specifically includes:

[0013] Pre-shoot and collect a large number of factory smoke exhaust related images and video materials, and find clear images and videos with long and continuous smoke exhaust time as samples;

[0014] Classify samples according to the light conditions, wind direction and wind level at the time of shooting, and group samples with the same light conditions and wind conditions into the same sample set;

[0015] Establishing an environment-color standard feature vector hash table according to the color standard feature vector corresponding to the environment condition and the sample set;

[0016] An environment-pixel fluctuation standard feature vector hash table is established according to the time-series pixel fluctuation standard feature vector corresponding to the environmental conditions and the sample set.

[0017] In a preferred embodiment, the processing of the sample smoke image according to the color feature processing flow, analyzing the color features of the sample smoke image, and establishing the color standard feature vector specifically includes:

[0018] Process the sample smoke image according to the color feature processing flow to obtain the color feature vector of the sample smoke image, perform feature aggregation on the color feature vectors of all the sample smoke images to obtain an aggregated feature vector, and mark it as a color standard feature vector;

[0019] Among them, the color feature processing flow is:

[0020] Extract the color features of the smoke image, calculate the number of pixels in each color grouping interval according to the predefined color grouping, and establish the color histogram H; H = [h1,h2,…h i ,…h n ];h i Represents the number of pixels in the i-th color grouping interval;

[0021] Select the RGB color space and construct the color moment through the mean, variance and skewness of the color channel of the smoke image. The expression is:

[0022]

[0023] In the formula, u i is the mean of the i-th color channel, is the variance of the ith color channel, γi is the skewness of the i-th color channel, F is the set of mean, variance and skewness of all color channels, f i are the mean, variance and skewness of the i-th color channel;

[0024] Construct a color feature vector based on the color histogram and color moment:

[0025] T C =[H,F];

[0026] Among them, T C is the color feature vector;

[0027] Perform feature aggregation on the color feature vectors of all sample smoke. The feature vector aggregation expression is:

[0028]

[0029] Where, T S is the color feature vector after feature aggregation, n is the number of selected samples, T Ci is the color feature vector of the i-th sample;

[0030] The color feature vector T after feature aggregation S Mark it as the color standard feature vector, and at the same time obtain the environmental condition information of the sample set and store it in the environment-color standard feature vector hash table.

[0031] In a preferred embodiment, the processing of the sample smoke image according to the pixel fluctuation feature processing flow, analyzing the pixel fluctuation feature of the sample smoke image, and establishing the time-series pixel fluctuation standard feature vector specifically includes:

[0032] Collect images of the sample video at fixed time intervals as sample smoke images for pixel fluctuation feature processing;

[0033] Set time points with continuous equal intervals, process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the extracted sample smoke image, calculate the pixel fluctuation feature vector of each corresponding time point, perform feature aggregation on the pixel fluctuation feature vectors of all sample smoke images at the time point, and then form a set of pixel fluctuation feature vectors after feature aggregation at each time point, which are marked as time series pixel fluctuation standard feature vectors;

[0034] Among them, the pixel fluctuation feature processing flow is:

[0035] Set fixed time intervals T, T = 1, 2, 3..., and analyze the changes in the number of smoke pixels and the distribution area of ​​the image at each time point;

[0036] For each time point of the smoke image, calculate the standard deviation of the number of pixels σ N , the calculation formula for the standard deviation of the number of pixels is:

[0037]

[0038] Where T is the current time point, N t is the number of pixels at the t-th time point, Represents the average number of pixels at the t-th time point;

[0039] Establish the minimum bounding rectangle based on the pixel boundary and calculate the standard deviation of the minimum bounding rectangle area change σ S , the calculation formula for the standard deviation of the minimum circumscribed rectangle area change is:

[0040]

[0041] Where T is the current time point, S t is the minimum circumscribed rectangle area at the t-th time point, Represents the average value of the minimum circumscribed rectangle area at the t-th time point;

[0042] The pixel fluctuation feature vector is calculated based on the standard deviation of the number of pixels and the standard deviation of the area of ​​the minimum circumscribed rectangle of the pixel boundary. The calculation formula of the pixel fluctuation feature vector is:

[0043]

[0044] In the formula, R C is the pixel fluctuation eigenvector;

[0045] The pixel fluctuation feature vectors of all samples at each time point are aggregated, and the aggregation expression is:

[0046]

[0047] In the formula, R S is the pixel fluctuation feature vector after feature aggregation, n is the number of selected samples, R Ci is the smoke pixel fluctuation feature vector of the i-th sample;

[0048] According to the aggregated pixel fluctuation feature vector at each time point, the time series pixel fluctuation feature vector R is established T =[R S1 ,R S2 ,...R Si ,...R Sn ]; R Si is the pixel fluctuation feature vector after feature aggregation at the i-th time point;

[0049] The time series pixel fluctuation feature vector R T Mark it as the time series pixel fluctuation standard feature vector, and at the same time obtain the environmental condition information of the sample set and store it in the environment-time series pixel fluctuation standard feature vector hash table.

[0050] In a preferred embodiment, the real-time smoke image collected during factory operation is analyzed, the real-time smoke image is processed according to the color feature processing flow to establish the real-time color feature vector of the smoke, and the real-time smoke image is processed according to the pixel fluctuation feature processing flow to establish the real-time sequential pixel fluctuation feature vector of the smoke specifically includes:

[0051] Obtain the real-time smoke emission images and video data of the factory that needs to be processed, collect images at fixed time intervals, and use them as real-time smoke images for pixel fluctuation feature processing;

[0052] Process the real-time smoke image according to the color feature processing flow and calculate the color feature vector T of the current real-time smoke image N ;

[0053] Process the real-time smoke image according to the pixel fluctuation feature processing flow and calculate the current real-time time series pixel fluctuation feature vector R N .

[0054] In a preferred embodiment, the comprehensive analysis of the standard feature vector and the real-time feature vector, calculation of the smoke feature deviation coefficient, evaluation of the abnormal deviation degree of the smoke emitted by the factory, and early warning of the abnormality of the smoke emitted by the factory specifically include:

[0055] Get the current light and wind data, find the corresponding sample standard feature vector in the hash table, compare the difference with the current real-time feature vector, and calculate the feature difference coefficient;

[0056] According to the current environmental conditions, the color standard feature vector is retrieved from the environment-color standard feature vector hash table, and the Euclidean distance d between the feature vector of the real-time smoke color image and the color standard feature vector is calculated. c , used to evaluate smoke color differences;

[0057] According to the current environmental conditions, the pixel fluctuation standard feature vector is retrieved from the environment-time series pixel fluctuation standard feature vector hash table, and the Euclidean distance d between the real-time smoke pixel fluctuation feature vector and the pixel fluctuation standard feature vector is calculated. p , used to evaluate smoke pixel fluctuation differences;

[0058] According to d c and d p Calculate the smoke characteristic deviation coefficient. The calculation formula for the smoke characteristic deviation coefficient is:

[0059]

[0060] In the formula, E d is the smoke characteristic deviation coefficient, β1 and β2 are d c and d p The corresponding weights;

[0061] Set the smoke characteristic deviation coefficient warning threshold, compare the smoke characteristic deviation coefficient with the smoke characteristic deviation coefficient threshold, and issue a warning for abnormal deviations in factory production smoke emissions:

[0062] The smoke characteristic deviation coefficient warning threshold is set. The smoke characteristic deviation coefficient threshold is set by professional and technical personnel in this field according to the size of the fog characteristic deviation coefficient and other actual conditions of factories of different types and specifications and their production environments, and will not be repeated here.

[0063] When the smoke characteristic deviation coefficient is greater than or equal to the smoke characteristic deviation coefficient threshold, a smoke characteristic abnormal deviation warning signal is generated;

[0064] When the smoke feature deviation coefficient is less than the smoke feature deviation coefficient threshold, a smoke feature abnormal deviation normal signal is generated.

[0065] In a preferred embodiment, an image recognition system in the field of industrial safety based on an AI big model includes:

[0066] An image acquisition and classification module, which is used for subsequent image analysis and processing by shooting and acquiring sample image data and real-time image data of the target and classifying them according to different environmental conditions;

[0067] A color feature processing module, wherein the color feature processing module extracts the color features of the smoke image and constructs a color feature vector according to a color histogram and a color moment;

[0068] A pixel fluctuation feature processing module, which establishes a time-series pixel fluctuation feature vector by analyzing the change in the number of smoke pixels and the change in the distribution area of ​​the image at each time point;

[0069] The abnormal difference assessment and warning module analyzes the difference between the sample feature vector and the real-time feature vector, and issues an abnormal difference warning when the smoke feature deviation is too large.

[0070] The technical effects and advantages of the image recognition method and system in the field of industrial safety based on the AI ​​big model of the present invention are as follows:

[0071] By collecting a large number of samples of images and video materials related to factory smoke emissions, the samples are classified according to different environmental conditions, the color features and pixel fluctuation features of the sample images in each classified sample set are extracted for analysis, and a color standard feature vector and a real-time time series pixel fluctuation standard feature vector are established. The smoke images of the factory during production are collected in real time, and the color features and pixel fluctuation features of the real-time images are analyzed. The real-time color feature vector and the real-time pixel fluctuation feature vector are established, and the degree of deviation between the standard feature vector and the real-time feature vector is analyzed. The degree of abnormal deviation of smoke emission during factory production is evaluated, and an early warning is issued for abnormal deviation of smoke emission during factory production;

[0072] Analyzing the characteristics of factory exhaust images based on image recognition can effectively solve the problems of timely warning of production problems and failure of monitoring equipment under extreme production conditions; BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 It is a schematic diagram of an image recognition method in the field of industrial safety based on an AI large model of the present invention;

[0074] Figure 2 The flow chart of the present invention classifies samples according to different environmental conditions and establishes a sample set feature vector hash table corresponding to different environmental conditions;

[0075] Figure 3 The flow chart of processing the sample smoke image according to the color feature processing flow of the present invention, analyzing the color features of the sample smoke image, and establishing the color standard feature vector;

[0076] Figure 4 The present invention processes the sample smoke image according to the pixel fluctuation feature processing flow, analyzes the pixel fluctuation feature of the sample smoke image, and establishes a time-series pixel fluctuation standard feature vector flow chart;

[0077] Figure 5 The present invention collects real-time smoke images during factory operation for analysis, processes the real-time smoke images according to a color feature processing flow, establishes a real-time color feature vector of smoke, processes the real-time smoke images according to a pixel fluctuation feature processing flow, and establishes a flow chart of a real-time sequential pixel fluctuation feature vector of smoke;

[0078] Figure 6 The flow chart of analyzing the standard characteristic vector and the real-time characteristic vector of the present invention, calculating the smoke characteristic deviation coefficient, evaluating the abnormal deviation degree of the smoke emitted by the factory production, and issuing an early warning for the abnormal smoke emitted by the factory production;

[0079] Figure 7 It is a structural schematic diagram of the image recognition system in the field of industrial safety based on the AI ​​big model of the present invention. DETAILED DESCRIPTION

[0080] 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.

[0081] Example 1

[0082] Reference Figure 1 As shown, the image recognition method in the field of industrial safety based on the AI ​​big model includes the following steps:

[0083] Collect images and videos related to factory smoke emissions, find the ones with clear images as samples, classify the samples according to different environmental conditions, and establish a sample set feature vector hash table corresponding to different environmental conditions;

[0084] Process the sample smoke image according to the color feature processing flow, analyze the color features of the sample smoke image, and establish a color standard feature vector;

[0085] Process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the sample smoke image, and establish a time-series pixel fluctuation standard feature vector;

[0086] Collect real-time smoke images during factory operation for analysis, process the real-time smoke images according to the color feature processing flow, establish the real-time color feature vector of smoke, process the real-time smoke images according to the pixel fluctuation feature processing flow, and establish the real-time time-series pixel fluctuation feature vector of smoke;

[0087] Comprehensively analyze the standard feature vector and real-time feature vector, calculate the smoke feature deviation coefficient, evaluate the abnormal deviation degree of factory production smoke emissions, and issue early warning for abnormal factory production smoke emissions.

[0088] Reference Figure 2 As shown, the samples are classified according to different environmental conditions, and the sample set feature vector hash table corresponding to different environmental conditions is established, which specifically includes:

[0089] Pre-shoot and collect a large number of factory smoke exhaust related images and video materials, and find clear images and videos with long and continuous smoke exhaust time as samples;

[0090] Classify samples according to the light conditions, wind direction and wind level at the time of shooting, and group samples with the same light conditions and wind conditions into the same sample set;

[0091] Establishing an environment-color standard feature vector hash table according to the color standard feature vector corresponding to the environment condition and the sample set;

[0092] An environment-pixel fluctuation standard feature vector hash table is established according to the time-series pixel fluctuation standard feature vector corresponding to the environmental conditions and the sample set.

[0093] Image recognition models require a large number of training samples, which can mainly improve the generalization ability and data balance of the model, improve the stability of the feature vector, and make the trained feature vector more stable and reliable;

[0094] Reference Figure 3 As shown, the sample smoke image is processed according to the color feature processing flow, the color features of the sample smoke image are analyzed, and the color standard feature vector is established, which specifically includes:

[0095] Process the sample smoke image according to the color feature processing flow to obtain the color feature vector of the sample smoke image, perform feature aggregation on the color feature vectors of all the sample smoke images to obtain an aggregated feature vector, and mark it as a color standard feature vector;

[0096] Among them, the color feature processing flow is:

[0097] Extract the color features of the smoke image, calculate the number of pixels in each color grouping interval according to the predefined color grouping, and establish the color histogram H; H = [h1,h2,…h i ,…h n ];h i Represents the number of pixels in the i-th color grouping interval;

[0098] Select the RGB color space and construct the color moment through the mean, variance and skewness of the color channel of the smoke image. The expression is:

[0099]

[0100] In the formula, u i is the mean of the i-th color channel, is the variance of the ith color channel, γ i is the skewness of the i-th color channel, F is the set of mean, variance and skewness of all color channels, f i are the mean, variance and skewness of the i-th color channel;

[0101] Construct a color feature vector based on the color histogram and color moment:

[0102] T C =[H,F];

[0103] Among them, T C is the color feature vector;

[0104] Perform feature aggregation on the color feature vectors of all sample smoke. The feature vector aggregation expression is:

[0105]

[0106] Where, T S is the color feature vector after feature aggregation, n is the number of selected samples, T Ci is the color feature vector of the i-th sample;

[0107] The color feature vector T after feature aggregation S Mark it as the color standard feature vector, and at the same time obtain the environmental condition information of the sample set and store it in the environment-color standard feature vector hash table.

[0108] The practical significance of analyzing the color characteristics of exhaust smoke images is that in actual production activities, such as in the field of metal smelting, when a large amount of yellow smoke is generated, it indicates that there may be problems with the dust removal system and it is unable to purify the flue gas generated by the converter. Other industrial fields also have their corresponding smoke color characteristics. Based on the differences in color characteristics, it is possible to preliminarily assess whether there are production problems.

[0109] Reference Figure 4 As shown, the sample smoke image is processed according to the pixel fluctuation feature processing flow, the pixel fluctuation features of the sample smoke image are analyzed, and the time series pixel fluctuation standard feature vector is established, which specifically includes:

[0110] Collect images of the sample video at fixed time intervals as sample smoke images for pixel fluctuation feature processing;

[0111] Set time points with continuous equal intervals, process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the extracted sample smoke image, calculate the pixel fluctuation feature vector of each corresponding time point, perform feature aggregation on the pixel fluctuation feature vectors of all sample smoke images at the time point, and then form a set of pixel fluctuation feature vectors after feature aggregation at each time point, which are marked as time series pixel fluctuation standard feature vectors;

[0112] Among them, the pixel fluctuation feature processing flow is:

[0113] Set fixed time intervals T, T = 1, 2, 3..., and analyze the changes in the number of smoke pixels and the distribution area of ​​the image at each time point;

[0114] For each time point of the smoke image, calculate the standard deviation of the number of pixels σ N , the calculation formula for the standard deviation of the number of pixels is:

[0115]

[0116] Where T is the current time point, N t is the number of pixels at the t-th time point, Represents the average number of pixels at the t-th time point;

[0117] Establish the minimum bounding rectangle based on the pixel boundary and calculate the standard deviation of the minimum bounding rectangle area change σ S , the calculation formula for the standard deviation of the minimum circumscribed rectangle area change is:

[0118]

[0119] Where T is the current time point, S t is the minimum circumscribed rectangle area at the t-th time point, Represents the average value of the minimum circumscribed rectangle area at the t-th time point;

[0120] The pixel fluctuation feature vector is calculated based on the standard deviation of the number of pixels and the standard deviation of the area of ​​the minimum circumscribed rectangle of the pixel boundary. The calculation formula of the pixel fluctuation feature vector is:

[0121]

[0122] In the formula, R C is the pixel fluctuation eigenvector;

[0123] The pixel fluctuation feature vectors of all samples at each time point are aggregated, and the aggregation expression is:

[0124]

[0125] In the formula, R S is the pixel fluctuation feature vector after feature aggregation, n is the number of selected samples, R Ci is the smoke pixel fluctuation feature vector of the i-th sample;

[0126] According to the aggregated pixel fluctuation feature vector at each time point, the time series pixel fluctuation feature vector R is established T =[R S1 ,R S2 ,…R Si ,…R Sn ]; R Si is the pixel fluctuation feature vector after feature aggregation at the i-th time point;

[0127] The time series pixel fluctuation feature vector R T Mark it as the time series pixel fluctuation standard feature vector, and at the same time obtain the environmental condition information of the sample set and store it in the environment-time series pixel fluctuation standard feature vector hash table.

[0128] By analyzing the pixel fluctuation characteristics of the image, we can intuitively monitor the smoke exhaust volume, smoke exhaust change and diffusion speed during factory production. When the pixel fluctuation of the smoke is obvious, it means that the concentration of particulate matter in the smoke is high, which may reflect problems with production raw materials or incomplete combustion.

[0129] Reference Figure 5 As shown, real-time smoke images during factory operation are collected for analysis, the real-time smoke images are processed according to the color feature processing flow to establish a real-time color feature vector of smoke, and the real-time smoke images are processed according to the pixel fluctuation feature processing flow to establish a real-time sequential pixel fluctuation feature vector of smoke. Specifically, the following steps are performed:

[0130] Obtain the real-time smoke emission images and video data of the factory that needs to be processed, collect images at fixed time intervals, and use them as real-time smoke images for pixel fluctuation feature processing;

[0131] Process the real-time smoke image according to the color feature processing flow and calculate the color feature vector T of the current real-time smoke image N ;

[0132] Process the real-time smoke image according to the pixel fluctuation feature processing flow and calculate the current real-time time series pixel fluctuation feature vector R N .

[0133] Reference Figure 6 As shown, the standard feature vector and the real-time feature vector are comprehensively analyzed, the smoke feature deviation coefficient is calculated, the abnormal deviation degree of the smoke emitted by the factory is evaluated, and the abnormality of the smoke emitted by the factory is warned. Specifically, the following are included:

[0134] Get the current light and wind data, find the corresponding sample standard feature vector in the hash table, compare the difference with the current real-time feature vector, and calculate the feature difference coefficient;

[0135] According to the current environmental conditions, the color standard feature vector is retrieved from the environment-color standard feature vector hash table, and the Euclidean distance d between the feature vector of the real-time smoke color image and the color standard feature vector is calculated. c , used to evaluate smoke color differences;

[0136] According to the current environmental conditions, the pixel fluctuation standard feature vector is retrieved from the environment-time series pixel fluctuation standard feature vector hash table, and the Euclidean distance d between the real-time smoke pixel fluctuation feature vector and the pixel fluctuation standard feature vector is calculated. p , used to evaluate smoke pixel fluctuation differences;

[0137] According to d c and d p Calculate the smoke characteristic deviation coefficient. The calculation formula for the smoke characteristic deviation coefficient is:

[0138]

[0139] In the formula, E d is the smoke characteristic deviation coefficient, β1 and β2 are d c and d p The corresponding weights;

[0140] Set the smoke characteristic deviation coefficient warning threshold, compare the smoke characteristic deviation coefficient with the smoke characteristic deviation coefficient threshold, and issue a warning for abnormal deviations in factory production smoke emissions:

[0141] The smoke characteristic deviation coefficient warning threshold is set. The smoke characteristic deviation coefficient threshold is set by professional and technical personnel in this field according to the size of the fog characteristic deviation coefficient and other actual conditions of factories of different types and specifications and their production environments, and will not be repeated here.

[0142] When the smoke characteristic deviation coefficient is greater than or equal to the smoke characteristic deviation coefficient threshold, a smoke characteristic abnormal deviation warning signal is generated;

[0143] When the smoke feature deviation coefficient is less than the smoke feature deviation coefficient threshold, a smoke feature abnormal deviation normal signal is generated.

[0144] When the smoke characteristic deviation coefficient is greater than or equal to the smoke characteristic deviation coefficient threshold, a smoke characteristic abnormal deviation warning signal is generated, indicating that the production smoke emissions at this time deviate greatly from normal conditions, indicating that the factory production has become abnormal or the risk of abnormality is high. Personnel should be notified to determine whether there are any production hazards and to check and maintain equipment in a timely manner.

[0145] Example 2

[0146] The difference between Example 2 of the present invention and Example 1 is that this example introduces an image recognition system in the field of industrial safety based on an AI big model.

[0147] Reference Figure 7 As shown, a schematic diagram of the structure of an image recognition system in the field of industrial safety based on an AI large model is shown. The image recognition system in the field of industrial safety based on an AI large model includes:

[0148] An image acquisition and classification module, which is used for subsequent image analysis and processing by shooting and acquiring sample image data and real-time image data of the target and classifying them according to different environmental conditions;

[0149] A color feature processing module, wherein the color feature processing module extracts the color features of the smoke image and constructs a color feature vector according to a color histogram and a color moment;

[0150] A pixel fluctuation feature processing module, which establishes a time-series pixel fluctuation feature vector by analyzing the change in the number of smoke pixels and the change in the distribution area of ​​the image at each time point;

[0151] The abnormal difference assessment and warning module analyzes the difference between the sample feature vector and the real-time feature vector, and issues an abnormal difference warning when the smoke feature deviation is too large.

[0152] 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.

[0153] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0154] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0155] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0156] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0157] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0158] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0159] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0160] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0161] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An image recognition method in the field of industrial safety based on an AI big model, characterized in that: The steps include: Collect images and videos related to factory smoke emissions, find the ones with clear images as samples, classify the samples according to different environmental conditions, and establish a sample set feature vector hash table corresponding to different environmental conditions; Process the sample smoke image according to the color feature processing flow, analyze the color features of the sample smoke image, and establish a color standard feature vector; Process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the sample smoke image, and establish a time-series pixel fluctuation standard feature vector; Collect real-time smoke images during factory operation for analysis, process the real-time smoke images according to the color feature processing flow, establish the real-time color feature vector of smoke, process the real-time smoke images according to the pixel fluctuation feature processing flow, and establish the real-time time-series pixel fluctuation feature vector of smoke; Comprehensively analyze the standard feature vector and real-time feature vector, calculate the smoke feature deviation coefficient, evaluate the abnormal deviation degree of factory production smoke emissions, and issue early warning for abnormal factory production smoke emissions.

2. The image recognition method in the field of industrial safety based on AI big model according to claim 1 is characterized in that: The classifying of samples according to different environmental conditions and establishing a sample set feature vector hash table corresponding to different environmental conditions specifically includes: Pre-shoot and collect a large number of factory smoke exhaust related images and video materials, and find clear images and videos with long and continuous smoke exhaust time as samples; Classify samples according to the light conditions, wind direction and wind level at the time of shooting, and group samples with the same light conditions and wind conditions into the same sample set; Establishing an environment-color standard feature vector hash table according to the color standard feature vector corresponding to the environment condition and the sample set; An environment-pixel fluctuation standard feature vector hash table is established according to the time-series pixel fluctuation standard feature vector corresponding to the environmental conditions and the sample set.

3. The image recognition method in the field of industrial safety based on AI big model according to claim 2 is characterized in that: The processing of the sample smoke image according to the color feature processing flow, analyzing the color features of the sample smoke image, and establishing the color standard feature vector specifically includes: Process the sample smoke image according to the color feature processing flow to obtain the color feature vector of the sample smoke image, perform feature aggregation on the color feature vectors of all the sample smoke images to obtain an aggregated feature vector, and mark it as a color standard feature vector; Among them, the color feature processing flow is: Extract the color features of the smoke image, calculate the number of pixels in each color grouping interval according to the predefined color grouping, and establish the color histogram H; H = [h1, h2, ... h i ,…h n ];h i Represents the number of pixels in the i-th color grouping interval; Select the RGB color space and construct the color moment through the mean, variance and skewness of the color channel of the smoke image. The expression is: In the formula, u i is the mean of the i-th color channel, is the variance of the ith color channel, γ i is the skewness of the i-th color channel, F is the set of mean, variance and skewness of all color channels, f i are the mean, variance and skewness of the i-th color channel; Construct a color feature vector based on the color histogram and color moment: T C =[H,F]; Among them, T C is the color feature vector; Perform feature aggregation on the color feature vectors of all sample smoke. The feature vector aggregation expression is: Where, T S is the color feature vector after feature aggregation, n is the number of selected samples, T Ci is the color feature vector of the i-th sample; The color feature vector T after feature aggregation S Mark it as the color standard feature vector, and at the same time obtain the environmental condition information of the sample set and store it in the environment-color standard feature vector hash table.

4. The image recognition method in the field of industrial safety based on AI big model according to claim 3 is characterized in that: The processing of the sample smoke image according to the pixel fluctuation feature processing flow, analyzing the pixel fluctuation feature of the sample smoke image, and establishing the time series pixel fluctuation standard feature vector specifically includes: Collect images of the sample video at fixed time intervals as sample smoke images for pixel fluctuation feature processing; Set time points with continuous equal intervals, process the sample smoke image according to the pixel fluctuation feature processing flow, analyze the pixel fluctuation features of the extracted sample smoke image, calculate the pixel fluctuation feature vector of each corresponding time point, perform feature aggregation on the pixel fluctuation feature vectors of all sample smoke images at the time point, and then form a set of pixel fluctuation feature vectors after feature aggregation at each time point, which are marked as time series pixel fluctuation standard feature vectors; Among them, the pixel fluctuation feature processing flow is: Set fixed time intervals T, T = 1, 2, 3..., and analyze the changes in the number of smoke pixels and the distribution area of ​​the image at each time point; For each time point of the smoke image, calculate the standard deviation of the number of pixels σ N ; Establish the minimum bounding rectangle based on the pixel boundary and calculate the standard deviation of the minimum bounding rectangle area change σ S ; The pixel fluctuation feature vector is calculated based on the standard deviation of the number of pixels and the standard deviation of the area of ​​the minimum circumscribed rectangle of the pixel boundary. The calculation formula of the pixel fluctuation feature vector is: In the formula, R C is the pixel fluctuation eigenvector; The pixel fluctuation feature vectors of all samples at each time point are aggregated, and the aggregation expression is: In the formula, R S is the pixel fluctuation feature vector after feature aggregation, n is the number of selected samples, R Ci is the pixel fluctuation feature vector of the smoke of the i-th sample; According to the aggregated pixel fluctuation feature vector at each time point, the time series pixel fluctuation feature vector R is established T =[R S1 , R S2 , … R Si , … R Sn ]; R Si is the pixel fluctuation feature vector after feature aggregation at the i-th time point; The temporal pixel fluctuation feature vector RT is marked as the temporal pixel fluctuation standard feature vector, and the environmental condition information of the sample set is obtained and stored in the environment-temporal pixel fluctuation standard feature vector hash table.

5. The image recognition method in the field of industrial safety based on AI big model according to claim 4 is characterized in that: The collecting and analyzing of real-time smoke images during factory operation, processing the real-time smoke images according to the color feature processing flow, establishing the real-time color feature vector of smoke, processing the real-time smoke images according to the pixel fluctuation feature processing flow, and establishing the real-time sequential pixel fluctuation feature vector of smoke specifically include: Obtain the real-time smoke emission images and video data of the factory that needs to be processed, collect images at fixed time intervals, and use them as real-time smoke images for pixel fluctuation feature processing; Process the real-time smoke image according to the color feature processing flow and calculate the color feature vector T of the current real-time smoke image N ; The real-time smoke image is processed according to the pixel fluctuation feature processing flow, and the current real-time time series pixel fluctuation feature vector RN is calculated.

6. The image recognition method in the field of industrial safety based on AI big model according to claim 5 is characterized in that: The comprehensive analysis of the standard feature vector and the real-time feature vector, calculation of the smoke feature deviation coefficient, evaluation of the abnormal deviation degree of the smoke emitted by the factory, and early warning of the abnormal smoke emitted by the factory specifically include: Get the current light and wind data, find the corresponding sample standard feature vector in the hash table, compare the difference with the current real-time feature vector, and calculate the feature difference coefficient; According to the current environmental conditions, the color standard feature vector is retrieved from the environment-color standard feature vector hash table, and the Euclidean distance d between the feature vector of the real-time smoke color image and the color standard feature vector is calculated. c , used to evaluate smoke color differences; According to the current environmental conditions, the pixel fluctuation standard feature vector is retrieved from the environment-time series pixel fluctuation standard feature vector hash table, and the Euclidean distance d between the real-time smoke pixel fluctuation feature vector and the pixel fluctuation standard feature vector is calculated. p , used to evaluate smoke pixel fluctuation differences; According to d c and d p Calculate the smoke characteristic deviation coefficient. The calculation formula for the smoke characteristic deviation coefficient is: In the formula, E d is the smoke characteristic deviation coefficient, β1 and β2 are d c and d p The corresponding weights; Set the smoke characteristic deviation coefficient warning threshold, compare the smoke characteristic deviation coefficient with the smoke characteristic deviation coefficient threshold, and issue a warning for abnormal deviations in factory production smoke emissions: When the smoke characteristic deviation coefficient is greater than or equal to the smoke characteristic deviation coefficient threshold, a smoke characteristic abnormal deviation warning signal is generated; When the smoke feature deviation coefficient is less than the smoke feature deviation coefficient threshold, a smoke feature abnormal deviation normal signal is generated.

7. An image recognition system in the field of industrial safety based on an AI large model, used to implement the image recognition method in the field of industrial safety based on an AI large model as described in any one of claims 1 to 6, characterized in that: include: An image acquisition and classification module, which is used for subsequent image analysis and processing by shooting and acquiring sample image data and real-time image data of the target and classifying them according to different environmental conditions; A color feature processing module, wherein the color feature processing module extracts the color features of the smoke image and constructs a color feature vector according to a color histogram and a color moment; A pixel fluctuation feature processing module, which establishes a time-series pixel fluctuation feature vector by analyzing the change in the number of smoke pixels and the change in the distribution area of ​​the image at each time point; The abnormal difference assessment and warning module analyzes the difference between the sample feature vector and the real-time feature vector, and issues an abnormal difference warning when the smoke feature deviation is too large.

Images