Intelligent operation risk management and control method and system based on image recognition

Abstract

The invention relates to the technical field of operation risk management and control, in particular to an intelligent operation risk management and control method and system based on image recognition, and the method comprises the steps: constructing different types of risk recognition models through collected different recognition content risk feature data, carrying out the precise matching of the different types of risk recognition models with operation scene text feature data, and generating an operation scene risk recognition model; performing data preprocessing on the real-time condition image data of the operation scene to generate real-time condition image feature data of the operation scene, performing scientific analysis on the real-time condition of the operation scene according to the operation scene risk identification model and the real-time condition image feature data of the operation scene to generate real-time risk analysis data of the operation scene, and performing risk analysis on the real-time condition of the operation scene according to the real-time risk analysis data of the operation scene. If so, ending the operation risk management and control task; otherwise, generating real-time risk text feature data of the operation scene and pushing the real-time risk text feature data to the operation risk management and control platform to give an early warning, thereby realizing accurate identification and intelligent management and control of the operation risk.

Description

Technical Field

[0001] The present invention relates to the technical field of operation risk management and control, and specifically to an operation risk intelligent management and control method and system based on image recognition. Background Art

[0002] According to the data in the "Accident Report and Annual Accident Analysis Report in December 2020" issued by the National Energy Administration. In 2020, there were 35 power-related personal injury accidents and 44 deaths in the country. Among them, according to statistics on the causes of the accidents, unsafe human behavior caused 22 accidents (accounting for 63%) and 31 deaths (accounting for 70%). This exposes the vacuum in the management of high-risk operations and the failure to effectively implement risk prevention and control. Accidents caused by unsafe human behavior account for the highest proportion. One of the main reasons is that the existing operation risk management technology cannot comprehensively control the safety risks of operators when performing tasks, resulting in low safety when performing tasks.

[0003] The Chinese invention patent application with publication number CN113869757A introduces a real-time monitoring system for high-risk work sites, including a real-time monitoring management platform for work sites, a mobile APP platform, and intelligent safety equipment. The intelligent safety equipment is a smart helmet with functional modules such as work video upload, work photo upload, work live broadcast, group voice call and voice intercom. The work video needs to be manually analyzed by the real-time monitoring management platform. When the network allows, the manual analysis of this method has a certain delay, and it is impossible to identify the risks on site in time and alarm; when there is no network, the intelligent safety equipment cannot accept the platform's manual analysis results and alarm, and only has a recording function, without risk identification and alarm function. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In order to solve the deficiencies in the background technology, the present invention provides a method and system for intelligent management and control of operation risks based on image recognition, so as to realize accurate identification and intelligent management and control of operation risks.

[0006] (II) Technical solution

[0007] An intelligent operation risk management method based on image recognition includes the following steps:

[0008] S1. Collect risk feature data of different identification contents;

[0009] S2. Constructing different types of risk identification models based on the risk feature data of different identification contents;

[0010] S3, collecting text feature data of the operation scene, performing risk identification model search processing on the text feature data of the operation scene and the different types of risk identification models, and generating an operation scene risk identification model;

[0011] S4, collecting real-time status image data of the operation scene;

[0012] S5, performing data preprocessing on the real-time status image data of the operation scene to generate real-time status image feature data of the operation scene;

[0013] S6. Perform real-time risk analysis of the operation scene according to the operation scene risk identification model and the real-time status image feature data of the operation scene to generate real-time risk analysis data of the operation scene. If it is normal, end the current operation risk management task; if it is abnormal, generate real-time risk text feature data of the operation scene;

[0014] S7. Push the real-time risk text feature data of the operation scenario to the operation risk management and control platform and issue an early warning.

[0015] The present invention constructs different types of risk identification models by collecting risk feature data of different identification contents, and accurately matches them with text feature data of operation scenes to generate risk identification models of operation scenes, and then preprocesses the real-time status image data of operation scenes to generate image feature data of real-time status of operation scenes. The real-time status of the operation scenes is scientifically analyzed according to the risk identification model of the operation scenes and the real-time status image feature data of the operation scenes to generate real-time risk analysis data of the operation scenes. If the data is normal, the operation risk management task is terminated; otherwise, the real-time risk text feature data of the operation scenes is generated and pushed to the operation risk management platform to issue an early warning, so as to realize accurate identification and intelligent management of operation risks.

[0016] Preferably, the specific steps for collecting different identification content risk feature data are as follows:

[0017] S11, collect risk feature data corresponding to different identification contents in the operation scene online through the operation risk management and control platform, and generate risk feature data sets A of different identification contents = {a 1 ,a 2 ,…,a i ,…,a k}, where a i represents the risk feature data set corresponding to the i-th type of recognition content, k represents the total number of recognition content categories, and the recognition content includes but is not limited to face recognition, violation recognition, safety protection equipment wearing recognition, and safety sign placement recognition. The risk feature data represents the image feature data and the corresponding risk text feature data when abnormalities occur in various types of recognition content in the operation scene.

[0018] Preferably, the specific steps of constructing different types of risk identification models based on the risk feature data of different identification contents are as follows:

[0019] S21, construct models based on the risk feature data sets corresponding to each type of identification content in the risk feature data sets of different identification contents, and generate different types of risk identification model sets in, Represents the risk identification model corresponding to the i-th type of identification content; the model construction includes the following steps:

[0020] S211, randomly selecting a risk feature data set corresponding to any type of identification content from the different identification content risk feature data sets as a sample data set;

[0021] S212, setting a training data ratio and a test data ratio, and dividing the sample data set according to the training data ratio and the test data ratio to obtain a training sample data set and a test sample data set respectively;

[0022] S213, constructing an initial convolutional neural network model, setting the number of input nodes of the initial convolutional neural network model to be a, the number of hidden nodes to be b, the number of output nodes to be c, the initial weight to be ω, and the initial bias to be ξ;

[0023] S214, setting a training error threshold and a maximum number of training times, inputting the training sample data in the training sample data set into the initial convolutional neural network model for training, if the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, then stopping the training to obtain a trained convolutional neural network model; otherwise, continuing the training until the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times;

[0024] S215, setting an accuracy threshold, inputting the test sample data in the test sample data set into the trained convolutional neural network model for testing, and if the test accuracy is greater than the accuracy threshold, obtaining a risk identification model corresponding to such identification content; otherwise, using a grid optimization algorithm to adjust the weights and biases of the trained convolutional neural network model to obtain a risk identification model corresponding to such identification content;

[0025] S216. Repeat steps S211 to S215 until all risk feature data sets corresponding to the class identification contents in the risk feature data sets of different identification contents are traversed to generate risk identification model sets of different types.

[0026] Different types of risk identification models are constructed by collecting risk feature data corresponding to different identification contents in the work scenes, providing a reliable tool for identifying whether abnormalities occur in the work scenes, realizing rapid identification of real-time risks in the work scenes, and improving the safety of workers when performing tasks.

[0027] Preferably, the specific steps of collecting the text feature data of the operation scene, performing risk identification model search processing on the text feature data of the operation scene and the risk identification models of different types, and generating the risk identification model of the operation scene are as follows:

[0028] S31. Collecting real-time operation scene text feature data of operators online through the operation risk management and control platform to generate operation scene text feature data C;

[0029] S32, performing risk identification model search processing on the operation scene text feature data C and the different types of risk identification models in the different types of risk identification model sets through the salp optimization algorithm, and generating an operation scene risk identification model set D = {d 1 ,d 2 ,…,d i ,…,d l}, where d i represents the i-th type of risk identification model required for the real-time operation scenario of the operator, and l represents the total number of types of risk identification models required for the real-time operation scenario of the operator;

[0030] S321. Construct a risk identification model to search for Salp populations, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max And the dimension of the search space of different types of risk identification models is P;

[0031] The different types of risk identification model sets are used as different types of risk identification model search spaces, N different types of risk identification models are randomly generated in the different types of risk identification model search spaces, and each different type of risk identification model corresponds to a risk identification model search salp individual in the risk identification model search salp population;

[0032] S322, calculating the fitness values ​​of each risk identification model search salp individual in the risk identification model search salp population and the operation scene text feature data, arranging each risk identification model search salp individual in the risk identification model search salp population from large to small according to the fitness value, and selecting the risk identification model search salp individual with the highest fitness value as the current optimal individual;

[0033] S323, each risk identification model search salp individual in the risk identification model search salp population updates its position in the different types of risk identification model search spaces by forming a salp chain, and the risk identification model search salp individual at the first position of the salp chain is used as a leader individual, and the remaining risk identification model search salps individuals are used as follower individuals;

[0034] S3231. The leader individual in the search space of the different types of risk identification models will be affected by the current optimal individual position and update its position; the position update formula is as follows:

[0035]

[0036] in, represents the position of the leader individual after the position update in the search space of different types of risk identification models in the jth dimension, X best represents the position of the current optimal individual in the search space of different types of risk identification models in the jth dimension, λ j and η j They represent the upper and lower bounds of the search space of the risk identification models of different types, rand 1 and rand 2 Both represent random numbers that are uniformly distributed between [0,1], α represents the convergence factor, and

[0037] S3232, each follower individual moves in the search space of the different types of risk identification models in a chain form, and follows the previous risk identification model search salp individual to update its position; the position update formula is as follows:

[0038]

[0039] in, represents the position of the i-th follower individual after the position update in the j-th dimension different types of risk identification model search space, represents the current position of the i-th follower individual in the search space of different types of risk identification models in the j-th dimension, represents the position of the previous risk identification model search salp individual followed by the i-th follower individual in the j-th dimension different types of risk identification model search space;

[0040] S324, calculating the fitness value of each risk identification model searching salp individual in the risk identification model searching salp population after the position is updated, if the fitness value of the risk identification model searching salp individual after the position is updated is greater than the original fitness value, then the position after the risk identification model searching salp individual is updated is used to replace the original position; otherwise, the original position of the risk identification model searching salp individual is retained;

[0041] Rearranging the risk identification model searching salp individuals in the risk identification model searching salp population from large to small according to fitness values, and selecting the risk identification model searching salp individual with the highest fitness value as the new current optimal individual;

[0042] S325: Determine whether the current number of iterations t is less than the maximum number of iterations t max , if the current number of iterations t is less than the maximum number of iterations t max , then the current iteration number t is increased by 1, and the process returns to S323; otherwise, a fitness threshold is set, and each risk identification model search salp individual whose fitness value is greater than the fitness threshold is screened out, and the different types of risk identification models corresponding to the screened risk identification model search salp individuals are combined to generate an operation scenario risk identification model set.

[0043] The risk identification model search and processing is performed on the text feature data of the operation scene and the different types of risk identification models through the salp sea squirt optimization algorithm, so as to quickly and accurately search for the risk identification model required for the current operation scene, avoid unnecessary resource consumption, and at the same time accelerate the convergence speed of the search process to ensure the accuracy of the search results.

[0044] Preferably, the specific steps of collecting real-time status image data of the operation scene are as follows:

[0045] S41. After the operator arrives at the work site, the operator uses the intelligent management and control terminal to shoot the real-time status image data of the work scene online to obtain the real-time status image data E of the work scene.

[0046] Preferably, the specific steps of preprocessing the real-time status image data of the operation scene to generate the real-time status image feature data of the operation scene are as follows:

[0047] S51, performing data noise reduction processing on the real-time status image data E of the operation scene by using an adaptive median filtering method to generate real-time status image feature data of the operation scene

[0048] The real-time status image data of the operation scene is subjected to data noise reduction processing by using an adaptive median filtering method, thereby effectively filtering out the noise in the real-time status image data of the operation scene and improving the accuracy of acquiring the image data.

[0049] Preferably, the operation scene real-time risk analysis processing is performed according to the operation scene risk identification model and the operation scene real-time status image feature data to generate the operation scene real-time risk analysis data. If it is normal, the operation risk management task is terminated; if it is abnormal, the specific steps of generating the operation scene real-time risk text feature data are as follows:

[0050] S61. According to the model priority order preset in the operation risk management and control platform, the operation scene risk identification models with high model priority in the operation scene risk identification model set are selected in turn to perform operation scene real-time risk analysis on the operation scene real-time status image feature data. If, after traversing all the operation scene risk identification models in the operation scene risk identification model set, no risk is identified in the operation scene real-time status image feature data, then the output of the operation scene real-time risk analysis data is normal, and the operation scene real-time risk analysis data is pushed to the operation risk management and control platform through the Internet of Things communication network, and the operation risk management and control task is terminated; otherwise, the output of the operation scene real-time risk analysis data is abnormal, and the recognition results output by the operation scene risk identification model that identifies the risk in the operation scene real-time status image feature data are summarized to generate an operation scene real-time risk text feature data set F={f 1 ,f 2 ,…,f i ,…,f p}, where f i It represents the text feature data corresponding to the i-th risk appearing in the real-time status of the operation scene, and p represents the total number of real-time risk text feature data of the operation scene.

[0051] According to the priority, the appropriate operation scenario risk identification model is selected in turn to accurately analyze whether there are any abnormalities in the current operation scenario, so as to achieve accurate identification of operation risks.

[0052] Preferably, the specific steps of pushing the real-time risk text feature data of the operation scene to the operation risk management and control platform and issuing an early warning are as follows:

[0053] S71. Push the real-time risk text feature data set F of the operation scene to the operation risk management and control platform through the Internet of Things communication network, and issue an early warning according to the alarm method pre-stored in the operation risk management and control platform, wherein the alarm method includes vibration of the intelligent management and control terminal, buzzer alarm, management background pop-up window and sound alarm.

[0054] The present invention also includes an intelligent operation risk management and control system based on image recognition, including a risk feature data collection module for different identification contents, a risk identification model construction module for different types, an operation scene risk identification model search module, an operation scene real-time status image data collection module, an operation scene real-time status image data preprocessing module, an operation scene real-time risk analysis module, and an operation risk early warning module;

[0055] The different identification content risk feature data collection module collects risk feature data corresponding to different identification contents in the operation scene online through the operation risk management and control platform to generate different identification content risk feature data;

[0056] The different types of risk identification model construction modules respectively construct models based on the risk feature data of each different identification content to generate different types of risk identification models;

[0057] The operation scene risk identification model search module collects the real-time operation scene text feature data of the operators online through the operation risk management and control platform to generate the operation scene text feature data, and performs risk identification model search processing on the operation scene text feature data and the different types of risk identification models through the salp optimization algorithm to generate the operation scene risk identification model;

[0058] After the operator arrives at the work site, the work scene real-time status image data acquisition module shoots the real-time status image data of the work scene online through the intelligent management and control terminal to obtain the real-time status image data of the work scene;

[0059] The operation scene real-time status image data preprocessing module performs data noise reduction processing on the operation scene real-time status image data by using an adaptive median filtering method to generate operation scene real-time status image feature data;

[0060] The operation scene real-time risk analysis module selects the operation scene risk identification model with high priority in turn according to the model priority order preset in the operation risk management and control platform to perform operation scene real-time risk analysis on the operation scene real-time status image feature data, and generates operation scene real-time risk analysis data. If it is normal, the operation risk management and control task is terminated; if it is abnormal, the operation scene real-time risk text feature data is generated;

[0061] The operation risk warning module pushes the real-time risk text feature data set of the operation scene to the operation risk management and control platform through the Internet of Things communication network, and issues a warning according to the alarm method pre-stored in the operation risk management and control platform.

[0062] (III) Beneficial effects

[0063] 1. The present invention constructs different types of risk identification models through the collected risk feature data of different identification contents, and accurately matches them with the text feature data of the operation scene to generate the risk identification model of the operation scene, and then preprocesses the real-time status image data of the operation scene to generate the real-time status image feature data of the operation scene, and scientifically analyzes the real-time status of the operation scene according to the risk identification model of the operation scene and the real-time status image feature data of the operation scene to generate the real-time risk analysis data of the operation scene. If it is normal, the operation risk management task is terminated; otherwise, the real-time risk text feature data of the operation scene is generated and pushed to the operation risk management platform to issue an early warning, so as to realize the accurate identification and intelligent management of operation risks;

[0064] 2. Construct different types of risk identification models through the risk feature data corresponding to different identification contents in the collected operation scenes, which provides a reliable tool for identifying whether there are abnormalities in the operation scenes, realizes the rapid identification of real-time risks in the operation scenes, and improves the safety of operators when performing tasks;

[0065] 3. Perform risk identification model search processing on the text feature data of the operation scene and the different types of risk identification models through the salp optimization algorithm, quickly and accurately search for the risk identification model required for the current operation scene, avoid unnecessary resource consumption, and accelerate the convergence speed of the search process to ensure the accuracy of the search results;

[0066] 4. The real-time status image data of the operation scene is subjected to data noise reduction processing through the adaptive median filtering method, so as to effectively filter out the noise in the real-time status image data of the operation scene and improve the accuracy of acquiring the image data. At the same time, the appropriate operation scene risk identification model is selected in sequence according to the priority to accurately analyze whether there is any abnormality in the current operation scene, so as to realize the accurate identification of operation risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0068] Figure 1 A flowchart of an intelligent operation risk management method based on image recognition provided by the present invention;

[0069] Figure 2 A module schematic diagram of an intelligent operation risk management and control system based on image recognition provided by the present invention. DETAILED DESCRIPTION

[0070] 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.

[0071] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0072] Embodiment 1 is as follows:

[0073] See also Figure 1 , an intelligent operation risk management method based on image recognition, comprising the following steps:

[0074] S1. Collect risk feature data of different identification contents;

[0075] S11, collect risk feature data corresponding to different identification contents in the operation scene online through the operation risk management and control platform, and generate risk feature data sets A of different identification contents = {a 1 ,a 2 ,…,a i ,…,a k}, where a i represents the risk feature data set corresponding to the i-th type of recognition content, k represents the total number of recognition content categories, and the recognition content includes but is not limited to face recognition, violation recognition, safety protection equipment wearing recognition, and safety sign placement recognition. The risk feature data represents the image feature data and the corresponding risk text feature data when abnormalities occur in various types of recognition content in the operation scene.

[0076] S2. Constructing different types of risk identification models based on the risk feature data of different identification contents;

[0077] S21, construct models based on the risk feature data sets corresponding to each type of identification content in the risk feature data sets of different identification contents, and generate different types of risk identification model sets in, Represents the risk identification model corresponding to the i-th type of identification content; the model construction includes the following steps:

[0078] S211, randomly selecting a risk feature data set corresponding to any type of identification content from the different identification content risk feature data sets as a sample data set;

[0079] S212, setting a training data ratio and a test data ratio, and dividing the sample data set according to the training data ratio and the test data ratio to obtain a training sample data set and a test sample data set respectively;

[0080] S213, constructing an initial convolutional neural network model, setting the number of input nodes of the initial convolutional neural network model to be a, the number of hidden nodes to be b, the number of output nodes to be c, the initial weight to be ω, and the initial bias to be ξ;

[0081] S214, setting a training error threshold and a maximum number of training times, inputting the training sample data in the training sample data set into the initial convolutional neural network model for training, if the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, then stopping the training to obtain a trained convolutional neural network model; otherwise, continuing the training until the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times;

[0082] S215, setting an accuracy threshold, inputting the test sample data in the test sample data set into the trained convolutional neural network model for testing, and if the test accuracy is greater than the accuracy threshold, obtaining a risk identification model corresponding to such identification content; otherwise, using a grid optimization algorithm to adjust the weights and biases of the trained convolutional neural network model to obtain a risk identification model corresponding to such identification content;

[0083] S216. Repeat steps S211 to S215 until all risk feature data sets corresponding to the class identification contents in the risk feature data sets of different identification contents are traversed to generate risk identification model sets of different types.

[0084] S3, collecting text feature data of the operation scene, performing risk identification model search processing on the text feature data of the operation scene and the different types of risk identification models, and generating an operation scene risk identification model;

[0085] S31. Collecting real-time operation scene text feature data of operators online through the operation risk management and control platform to generate operation scene text feature data C;

[0086] S32, performing risk identification model search processing on the operation scene text feature data C and the different types of risk identification models in the different types of risk identification model sets through the salp optimization algorithm, and generating an operation scene risk identification model set D = {d 1 ,d 2 ,…,d i ,…,d l}, where d i represents the i-th type of risk identification model required for the real-time operation scenario of the operator, and l represents the total number of types of risk identification models required for the real-time operation scenario of the operator;

[0087] S321. Construct a risk identification model to search for Salp populations, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max And the dimension of the search space of different types of risk identification models is P;

[0088] The different types of risk identification model sets are used as different types of risk identification model search spaces, N different types of risk identification models are randomly generated in the different types of risk identification model search spaces, and each different type of risk identification model corresponds to a risk identification model search salp individual in the risk identification model search salp population;

[0089] S322, calculating the fitness values ​​of each risk identification model search salp individual in the risk identification model search salp population and the operation scene text feature data, arranging each risk identification model search salp individual in the risk identification model search salp population from large to small according to the fitness value, and selecting the risk identification model search salp individual with the highest fitness value as the current optimal individual;

[0090] S323, each risk identification model search salp individual in the risk identification model search salp population updates its position in the different types of risk identification model search spaces by forming a salp chain, and the risk identification model search salp individual at the first position of the salp chain is used as a leader individual, and the remaining risk identification model search salps individuals are used as follower individuals;

[0091] S3231. The leader individual in the search space of the different types of risk identification models will be affected by the current optimal individual position and update its position; the position update formula is as follows:

[0092]

[0093] in, represents the position of the leader individual after the position update in the search space of different types of risk identification models in the jth dimension, X best represents the position of the current optimal individual in the search space of different types of risk identification models in the jth dimension, λ j and η j They represent the upper and lower bounds of the search space of the risk identification models of different types, rand 1 and rand 2Both represent random numbers that are uniformly distributed between [0,1], α represents the convergence factor, and

[0094] S3232, each follower individual moves in the search space of the different types of risk identification models in a chain form, and follows the previous risk identification model search salp individual to update its position; the position update formula is as follows:

[0095]

[0096] in, represents the position of the i-th follower individual after the position update in the j-th dimension different types of risk identification model search space, represents the current position of the i-th follower individual in the search space of different types of risk identification models in the j-th dimension, represents the position of the previous risk identification model search salp individual followed by the i-th follower individual in the j-th dimension different types of risk identification model search space;

[0097] S324, calculating the fitness value of each risk identification model searching salp individual in the risk identification model searching salp population after the position is updated, if the fitness value of the risk identification model searching salp individual after the position is updated is greater than the original fitness value, then the position after the risk identification model searching salp individual is updated is used to replace the original position; otherwise, the original position of the risk identification model searching salp individual is retained;

[0098] Rearranging the risk identification model searching salp individuals in the risk identification model searching salp population from large to small according to fitness values, and selecting the risk identification model searching salp individual with the highest fitness value as the new current optimal individual;

[0099] S325: Determine whether the current number of iterations t is less than the maximum number of iterations t max , if the current number of iterations t is less than the maximum number of iterations t max , then the current iteration number t is increased by 1, and the process returns to S323; otherwise, a fitness threshold is set, and each risk identification model search salp individual whose fitness value is greater than the fitness threshold is screened out, and the different types of risk identification models corresponding to the screened risk identification model search salp individuals are combined to generate an operation scenario risk identification model set.

[0100] S4, collecting real-time status image data of the operation scene;

[0101] S41. After the operator arrives at the work site, the operator uses the intelligent management and control terminal to shoot the real-time status image data of the work scene online to obtain the real-time status image data E of the work scene.

[0102] S5, performing data preprocessing on the real-time status image data of the operation scene to generate real-time status image feature data of the operation scene;

[0103] S51, performing data noise reduction processing on the real-time status image data E of the operation scene by using an adaptive median filtering method to generate real-time status image feature data of the operation scene

[0104] S6. Perform real-time risk analysis of the operation scene according to the operation scene risk identification model and the real-time status image feature data of the operation scene to generate real-time risk analysis data of the operation scene. If it is normal, end the current operation risk management task; if it is abnormal, generate real-time risk text feature data of the operation scene;

[0105] S61. According to the model priority order preset in the operation risk management and control platform, the operation scene risk identification models with high model priority in the operation scene risk identification model set are selected in turn to perform operation scene real-time risk analysis on the operation scene real-time status image feature data. If, after traversing all the operation scene risk identification models in the operation scene risk identification model set, no risk is identified in the operation scene real-time status image feature data, then the output of the operation scene real-time risk analysis data is normal, and the operation scene real-time risk analysis data is pushed to the operation risk management and control platform through the Internet of Things communication network, and the operation risk management and control task is terminated; otherwise, the output of the operation scene real-time risk analysis data is abnormal, and the recognition results output by the operation scene risk identification model that identifies the risk in the operation scene real-time status image feature data are summarized to generate an operation scene real-time risk text feature data set F={f 1 ,f 2 ,…,f i ,…,f p}, where f i It represents the text feature data corresponding to the i-th risk appearing in the real-time status of the operation scene, and p represents the total number of real-time risk text feature data of the operation scene.

[0106] S7, pushing the real-time risk text feature data of the operation scenario to the operation risk management and control platform, and issuing an early warning;

[0107] S71. Push the real-time risk text feature data set F of the operation scene to the operation risk management and control platform through the Internet of Things communication network, and issue an early warning according to the alarm method pre-stored in the operation risk management and control platform, wherein the alarm method includes vibration of the intelligent management and control terminal, buzzer alarm, management background pop-up window and sound alarm.

[0108] Embodiment 2 is as follows:

[0109] See also Figure 2 , an intelligent operation risk management and control system based on image recognition, including risk feature data collection modules for different identification contents, risk identification model construction modules for different types, operation scene risk identification model search modules, operation scene real-time status image data collection modules, operation scene real-time status image data preprocessing modules, operation scene real-time risk analysis modules, and operation risk warning modules;

[0110] The different identification content risk feature data collection module collects risk feature data corresponding to different identification contents in the operation scene online through the operation risk management and control platform to generate different identification content risk feature data;

[0111] The different types of risk identification model construction modules respectively construct models based on the risk feature data of each different identification content to generate different types of risk identification models;

[0112] The operation scene risk identification model search module collects the real-time operation scene text feature data of the operators online through the operation risk management and control platform to generate the operation scene text feature data, and performs risk identification model search processing on the operation scene text feature data and the different types of risk identification models through the salp optimization algorithm to generate the operation scene risk identification model;

[0113] After the operator arrives at the work site, the work scene real-time status image data acquisition module shoots the real-time status image data of the work scene online through the intelligent management and control terminal to obtain the real-time status image data of the work scene;

[0114] The operation scene real-time status image data preprocessing module performs data noise reduction processing on the operation scene real-time status image data by using an adaptive median filtering method to generate operation scene real-time status image feature data;

[0115] The operation scene real-time risk analysis module selects the operation scene risk identification model with high priority in turn according to the model priority order preset in the operation risk management and control platform to perform operation scene real-time risk analysis on the operation scene real-time status image feature data, and generates operation scene real-time risk analysis data. If it is normal, the operation risk management and control task is terminated; if it is abnormal, the operation scene real-time risk text feature data is generated;

[0116] The operation risk warning module pushes the real-time risk text feature data set of the operation scene to the operation risk management and control platform through the Internet of Things communication network, and issues a warning according to the alarm method pre-stored in the operation risk management and control platform.

[0117] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0118] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. An intelligent operation risk management method based on image recognition, characterized in that: The steps include: S1. Collect risk feature data of different identification contents; S2. Constructing different types of risk identification models based on the risk feature data of different identification contents; S3, collecting text feature data of the operation scene, performing risk identification model search processing on the text feature data of the operation scene and the different types of risk identification models, and generating an operation scene risk identification model; S4, collecting real-time status image data of the operation scene; S5, performing data preprocessing on the real-time status image data of the operation scene to generate real-time status image feature data of the operation scene; S6. Perform real-time risk analysis of the operation scene according to the operation scene risk identification model and the real-time status image feature data of the operation scene to generate real-time risk analysis data of the operation scene. If the data is normal, end the current operation risk management task; If it is abnormal, real-time risk text feature data of the operation scenario is generated; S7. Push the real-time risk text feature data of the operation scenario to the operation risk management and control platform and issue an early warning.

2. According to claim 1, a method for intelligent management and control of operation risks based on image recognition is characterized in that: The S1 comprises the following steps: S11. Collect risk feature data corresponding to different identification contents in the operation scene online through the operation risk management and control platform to generate risk feature data sets of different identification contents A = {a1, a2, ..., a i ,…,a k }, where a i represents the risk feature data set corresponding to the i-th type of recognition content, k represents the total number of recognition content categories, and the recognition content includes but is not limited to face recognition, violation recognition, safety protection equipment wearing recognition, and safety sign placement recognition. The risk feature data represents the image feature data and the corresponding risk text feature data when abnormalities occur in various types of recognition content in the operation scene.

3. The method for intelligent operation risk management based on image recognition according to claim 2 is characterized in that: The S2 comprises the following steps: S21, construct models based on the risk feature data sets corresponding to the various types of identification content in A, and generate different types of risk identification model sets in, Represents the risk identification model corresponding to the i-th type of identification content; the model construction includes the following steps: S211, randomly select the The risk feature dataset corresponding to any type of identification content is used as the sample dataset; S212, setting a training data ratio and a test data ratio, and dividing the sample data set according to the training data ratio and the test data ratio to obtain a training sample data set and a test sample data set respectively; S213, constructing an initial convolutional neural network model, setting the number of input nodes of the initial convolutional neural network model to be a, the number of hidden nodes to be b, the number of output nodes to be c, the initial weight to be ω, and the initial bias to be ξ; S214, setting a training error threshold and a maximum number of training times, inputting the training sample data in the training sample data set into the initial convolutional neural network model for training, if the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times, then stopping the training to obtain a trained convolutional neural network model; otherwise, continuing the training until the training error in the training process is less than the training error threshold or the number of training times is greater than the maximum number of training times; S215, setting an accuracy threshold, inputting the test sample data in the test sample data set into the trained convolutional neural network model for testing, and if the test accuracy is greater than the accuracy threshold, obtaining a risk identification model corresponding to such identification content; otherwise, using a grid optimization algorithm to adjust the weights and biases of the trained convolutional neural network model to obtain a risk identification model corresponding to such identification content; S216, repeat the steps from S211 to S215 until the After collecting the risk feature data sets corresponding to all the class identification contents, different types of risk identification model sets are generated.

4. The method for intelligent operation risk management based on image recognition according to claim 3 is characterized in that: The S3 comprises the following steps: S31. Collecting real-time operation scene text feature data of operators online through the operation risk management and control platform to generate operation scene text feature data C; S32, optimizing the C and the The different types of risk identification models in the risk identification model search process are used to generate the operation scenario risk identification model set D = {d1, d2, …, d i ,…,d l }, where d i represents the i-th type of risk identification model required for the real-time operation scenario of the operator, and l represents the total number of types of risk identification models required for the real-time operation scenario of the operator.

5. The method for intelligent operation risk management based on image recognition according to claim 4 is characterized in that: The S32 comprises the following steps: S321. Construct a risk identification model to search for Salp populations, set the population size to N, the current number of iterations to t, and the maximum number of iterations to t. max And the dimension of the search space of different types of risk identification models is P; The different types of risk identification model sets are used as different types of risk identification model search spaces, N different types of risk identification models are randomly generated in the different types of risk identification model search spaces, and each different type of risk identification model corresponds to a risk identification model search salp individual; S322, calculating the fitness value of each risk identification model for searching the salp individuals and the text feature data of the operation scene; S323, each risk identification model search salp individual updates its position in the different types of risk identification model search space by forming a salp chain, and the risk identification model search salp individual at the first position of the salp chain is used as a leader individual, and the remaining risk identification model search salps individuals are used as follower individuals; S3231, the leader individual in the search space of the risk identification model of different types will be affected by the current optimal individual position and update its position; S3232, each follower individual moves in the search space of the different types of risk identification models in a chain form, and follows the previous risk identification model search salp individual to update its position; S324, calculating the fitness value of each risk identification model after searching for the salp individual for position update, if the fitness value of the risk identification model after searching for the salp individual for position update is greater than the original fitness value, then the position after the risk identification model searches for the salp individual for position update is used to replace the original position; otherwise, the original position of the risk identification model searching for the salp individual is retained; Arrange the risk identification model search salp individuals from large to small according to their fitness values, and select the risk identification model search salp individual with the highest fitness value as the new current optimal individual; S325, determine whether t is less than t max , if t is less than t max , then t is increased by 1, and the process returns to S323; otherwise, a fitness threshold is set, and each risk identification model search salp individual whose fitness value is greater than the fitness threshold is screened out, and the different types of risk identification models corresponding to the screened risk identification model search salp individuals are combined to generate an operation scenario risk identification model set.

6. The method for intelligent operation risk management based on image recognition according to claim 5 is characterized in that: The S4 comprises the following steps: S41. After the operator arrives at the work site, the operator uses the intelligent management and control terminal to shoot the real-time status image data of the work scene online to obtain the real-time status image data E of the work scene.

7. The method for intelligent operation risk management based on image recognition according to claim 6 is characterized in that: The S5 comprises the following steps: S51, performing data noise reduction processing on the E by using an adaptive median filtering method to generate real-time image feature data of the operation scene 8. The method for intelligent operation risk management based on image recognition according to claim 7 is characterized in that: The S6 comprises the following steps: S61, according to the model priority order preset in the operation risk management and control platform, select the operation scenario risk identification model with the highest model priority in D in turn to identify the operation scenario risk of the operation scenario risk identification model. Perform real-time risk analysis of the operation scene. If all the operation scene risk identification models in D are traversed and no operation scene risk is identified, If there is a risk in the operation, the real-time risk analysis data of the operation scene is output as normal, and the real-time risk analysis data of the operation scene is pushed to the operation risk management and control platform through the Internet of Things communication network, and the operation risk management and control task is ended; otherwise, the real-time risk analysis data of the operation scene is output as abnormal, and the The recognition results of the risk identification model output by the operation scene with risks are summarized to generate the real-time risk text feature dataset F = {f1,f2,…,f i ,…,f p }, where f i It represents the text feature data corresponding to the i-th risk appearing in the real-time status of the operation scene, and p represents the total number of real-time risk text feature data of the operation scene.

9. The method for intelligent operation risk management based on image recognition according to claim 8, characterized in that: The S7 comprises the following steps: S71. Push the F to the operation risk management and control platform through the Internet of Things communication network, and issue an early warning according to the alarm method pre-stored in the operation risk management and control platform, wherein the alarm method includes vibration of the intelligent management and control terminal, buzzer alarm, management background pop-up window and sound alarm.

10. A system for implementing the method for intelligent operation risk management based on image recognition as described in any one of claims 1 to 9.

Images