A method for AI image recognition of hidden dangers in on-site operations

Abstract

The present invention belongs to the field of hidden dangers in on-site operations and relates to AI image recognition technology. It is used to solve the problem that the prior art cannot intelligently identify hidden dangers in on-site operations by shielding different types of on-site operations. Specifically, it is a method for AI image recognition of hidden dangers in on-site operations, comprising the following steps: segmenting an image of an on-site operation area to obtain monitoring units, identifying each monitoring unit through AI image recognition and classifying them into a personnel set, an equipment set and a space set; within the monitoring time period, alarming the hidden dangers in the monitoring area through real-time monitoring of the total number of personnel RY, the personnel flow value RL, the average recognition accuracy rate P of the equipment and the average density M of the space; and periodically updating the hidden danger warning threshold YXmax in combination with the number of hidden danger alarms generated. The present invention can combine multi-source data to identify hidden dangers for different on-site operations, thereby ensuring the versatility and accuracy of on-site operation hidden danger identification.

Description

Technical Field

[0001] The present invention belongs to the field of hidden dangers in on-site operations, relates to AI image recognition technology, and specifically is a method for AI image recognition of hidden dangers in on-site operations. Background Art

[0002] The hidden dangers existing in on-site operations are issues that need to be focused on in production safety management. Factors that may threaten the safety and health of workers at the work site include but are not limited to safety management, civilized construction, high-altitude operations, construction electricity, mechanical equipment, etc.; enterprises can effectively reduce safety hazards in on-site operations and ensure the safety of workers through education and training, regular inspections, and risk control measures.

[0003] There are many types of on-site operations, and each operation requires different skills, tools, and equipment. The existing methods for identifying hidden dangers in on-site operations are often carried out through manual investigation and tracking in specific operation scenarios. They are unable to shield the differences in the types of on-site operations and to intelligently identify hidden dangers in on-site operations in any scenario and conduct risk assessments. In addition, existing technologies are unable to combine multiple on-site data to provide timely and accurate warnings of operational hazards. Summary of the invention

[0004] The purpose of the present invention is to provide a method for AI image recognition of hidden dangers in on-site operations, which is used to solve the problem that the existing technology cannot screen the different types of on-site operations to perform intelligent recognition and risk assessment of hidden dangers in on-site operations;

[0005] The technical problem to be solved by the present invention is: how to provide a method for intelligently identifying and assessing the risks of on-site operations in any scenario.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for AI image recognition of hidden dangers in on-site operations, comprising the following steps:

[0008] Step 1: Mark the field operation area as the monitoring area, mark the operation time period of each working day as the monitoring time period, obtain the image of the monitoring area and divide it into several monitoring units, including personnel unit r, equipment unit s and space unit k;

[0009] Step 2: During the monitoring period, AI images are used to identify each monitoring unit and classify them into personnel sets, equipment sets, and space sets;

[0010] Step 3: Evaluate the hidden dangers in the personnel set, equipment set and space set respectively: During the monitoring period, monitor the total number of personnel RY and personnel flow value RL in the personnel set, the number of correct identification times SZ and the number of incorrect identification times SC in the equipment set, and the image area Sk of the space unit k in the space set and the area occupied by the object image Wk in real time, analyze the hidden dangers in the monitoring area and give an alarm;

[0011] Step 4: During the monitoring period, the number of hidden danger alarm signals generated by the personnel set, equipment set and space set is obtained in real time, and marked as the number of hidden dangers RH, the number of hidden dangers SH and the number of hidden dangers KH respectively, and the hidden danger exceeding the standard is further analyzed and alarmed, and the hidden danger warning coefficient YX is generated, and the hidden danger warning analysis is carried out according to the hidden danger warning coefficient YX;

[0012] Step 5: Generate an early warning optimization analysis cycle, and establish a comprehensive early warning set based on the hidden danger warning coefficient YX at the moment when the hidden danger exceeds the standard alarm signal is generated within the early warning optimization analysis cycle. The elements in the set are recorded as hidden danger warning coefficients YXn, n=1, 2,…, i, i is a positive integer, and the hidden danger warning threshold YXmax is periodically updated based on the result after discrete value elimination from the set.

[0013] Furthermore, in step one, a camera is used to take a picture of the monitoring area during the monitoring period to obtain an image of the monitoring area, which is then enlarged and transformed into grayscale. The monitoring area is segmented by a binary method to obtain a number of monitoring units, the attributes of which include personnel, equipment, and space. The monitoring units with the attributes of personnel, equipment, and space are marked as personnel unit r, equipment unit s, and space unit k, respectively.

[0014] Furthermore, in step two, facial images of the staff and front images of the equipment for the operation are extracted, and the images and names are stored in a local database in a one-to-one correspondence; a training set is formed by the facial images of the staff and the front images of the equipment in the local database and the existing data of the AI ​​model; based on the training set, AI image recognition is used to identify and classify each monitoring unit in the monitoring area, and the monitoring units with the attribute of personnel are classified into a personnel set, the monitoring units with the attribute of equipment are classified into an equipment set, and the monitoring units with the attribute of space are classified into a space set.

[0015] Furthermore, in step three, the hidden danger occurrence assessment process within the personnel set includes: when a staff member enters the monitoring area, the facial image of the staff member collected by the AI ​​image recognition camera is used. If the recognition is successful, the total number of personnel in the monitoring area RY is increased by 1, and the personnel flow value RL is recorded as increased by 1. If the recognition fails, it is prompted that non-staff members have entered the monitoring area, and the on-site operation has hidden dangers. A hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; when the staff member leaves the monitoring area, the facial image of the staff member in the monitoring area is recognized by AI image recognition, and the total number of personnel units r in the monitoring area at the current time is obtained and recorded as the total number of personnel RY, and the difference with the total number of personnel RY obtained last time is recorded. Recorded as the personnel flow value RL; obtain the personnel number threshold RYmax of the monitoring area. When the total number of personnel RY is greater than or equal to the personnel number threshold RYmax, it is judged that the personnel in the monitoring area are redundant, and it is prompted that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; obtain the personnel flow threshold RLmax of the monitoring area, take the absolute value and sum of all the obtained personnel flow values ​​RL to obtain the personnel flow RYL. When RYL is greater than or equal to the personnel flow threshold RLmax, it is judged that the personnel flow in the monitoring area is large, and it is prompted that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager.

[0016] Furthermore, in step three, the hidden danger occurrence assessment process in the equipment set includes: using the front image of the equipment captured by the AI ​​image recognition camera, if the recognition is successful, the number of correct recognition times SZ is increased by 1, if the recognition fails, the number of incorrect recognition times SC is increased by 1, and the recognition accuracy Ps of the equipment unit s is obtained by numerical calculation through the formula Ps=SZ / (SZ+SC); the recognition accuracy Ps of all equipment units s in the equipment set are summed and averaged to obtain the average recognition accuracy P; the equipment abnormality rate threshold PY is obtained, if the average recognition accuracy P is greater than or equal to the equipment abnormality rate threshold PY, it is judged that the equipment is operating normally and there are no hidden dangers in the on-site operation; if the average recognition accuracy P is less than the equipment abnormality rate threshold PY, it is judged that the equipment is operating abnormally, indicating that there are hidden dangers in the on-site operation, generating a hidden danger alarm signal and sending the hidden danger alarm signal to the mobile phone terminal of the manager.

[0017] Furthermore, in step three, the hidden danger occurrence assessment process in the space set includes: using the AI ​​image recognition camera to collect the object image in the space unit k, recording the sum of the object image areas in the space unit k as the object image occupied area Wk, obtaining the image area Sk of the space unit k, numerically calculating the object image occupied area Wk and the image area Sk of the space unit k by the formula Mk=Wk / Sk, obtaining the density Mk of the space unit k, summing and averaging the density Mk of all the space units k in the space set to obtain the average density M; obtaining the space density threshold MY, if the average density M is greater than or equal to the space density threshold MY, then it is judged that the space in the monitoring area is dense and messy, indicating that the on-site operation has hidden dangers, generating a hidden danger alarm signal and sending the hidden danger alarm signal to the mobile phone terminal of the manager; if the average density M is less than the space density threshold, then it is judged that the space in the monitoring area is spacious and orderly, and the on-site operation does not have hidden dangers.

[0018] Furthermore, in step four, the personnel hidden danger threshold RHmax, the equipment hidden danger threshold SHmax and the space hidden danger threshold KHmax are obtained, and the number of personnel hidden dangers RH, the number of equipment hidden dangers SH and the number of space hidden dangers KH are compared with the corresponding personnel hidden danger threshold RHmax, the equipment hidden danger threshold SHmax and the space hidden danger threshold KHmax respectively: if at least one indicator exceeds the preset threshold, it is judged that the hidden danger of the on-site operation exceeds the standard, and a hidden danger exceeding standard alarm signal is generated and the hidden danger exceeding standard alarm signal is sent to the mobile phone terminal of the manager; within the monitoring time period, the number of personnel hidden dangers RH, the number of equipment hidden dangers SH and the number of space hidden dangers KH are numerically calculated by the formula YX=k1×RH+k2×SH+k3×KH to obtain the hidden danger warning coefficient YX, wherein k1, k2 and k3 are all proportional coefficients, and k3>k2>k1>1; the hidden danger warning coefficient YX is compared with the preset hidden danger warning threshold YXmax: if the hidden danger warning coefficient YX is greater than or equal to the hidden danger warning threshold YXmax, a hidden danger warning signal is generated and sent to the mobile phone terminal of the manager.

[0019] Furthermore, in step five, all elements in the comprehensive warning set are calculated to obtain the standard deviation S and the average value X, and the statistic Tn is calculated by the formula Tn=(YXn-X) / S; the critical value Ta is obtained by looking up the table, if Tn≥Ta, it means that the hidden danger warning coefficient YXn is a discrete value and needs to be discarded, otherwise it is retained; and the minimum element in the comprehensive warning set after discarding the discrete value is marked as the hidden danger warning threshold value YXmax for the next analysis cycle, and the updated hidden danger warning threshold value YXmax is directly called when performing hidden danger warning in the next analysis cycle.

[0020] The present invention has the following beneficial effects:

[0021] 1. The training set is composed of the facial images of the staff and the front images of the equipment in the local database and the existing data of the AI ​​model. Based on the training set, the AI ​​images can accurately identify the staff and equipment in the field operation and classify them, and then identify the hidden dangers for different types of field operations, ensuring the intelligence and versatility of the hidden danger identification of field operations;

[0022] 2. AI image recognition technology is used to monitor the total number of personnel, personnel flow, equipment operation status and space density at the work site in real time, and a variety of data are fully combined to conduct comprehensive alarms for hidden dangers in on-site operations, thereby improving the accuracy and timeliness of identifying hidden dangers in on-site operations;

[0023] 3. During the early warning optimization analysis cycle, the hidden danger warning coefficient YX at the time when the hidden danger exceeding the standard alarm signal is generated is counted and screened, and then the hidden danger warning threshold YXmax is periodically updated, thereby improving the accuracy of early warning of hidden dangers in on-site operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0025] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0026] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than 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.

[0027] like Figure 1 As shown, a method for AI image recognition of hidden dangers in on-site operations includes the following steps:

[0028] Step 1: Mark the on-site operation area as the monitoring area, mark the operation time period of each working day as the monitoring time period, take pictures of the monitoring area through the camera during the monitoring time period, obtain the image of the monitoring area, enlarge the image and transform the grayscale, and segment the monitoring area through the binary method to obtain several monitoring units. The attributes of the monitoring unit include personnel, equipment and space. The monitoring units with the attributes of personnel, equipment and space are marked as personnel unit r, equipment unit s and space unit k respectively;

[0029] Step 2: Identify and classify the attributes of each monitoring unit: extract the facial images of the staff and the front images of the equipment for the operation, and store the images and names in a one-to-one correspondence in the local database; the facial images of the staff and the front images of the equipment in the local database and the existing data of the AI ​​model constitute a training set; based on the training set, use AI image recognition to identify and classify the various monitoring units in the monitoring area, and the monitoring units with the attribute of personnel are classified as personnel sets, the monitoring units with the attribute of equipment are classified as equipment sets, and the monitoring units with the attribute of space are classified as space sets; the training set of the local database combined with the data of the existing AI model can identify hidden dangers for different types of on-site operations, ensuring the intelligence and versatility of on-site operation hidden danger identification;

[0030] Step 3: Evaluate the potential hazards in the personnel set, equipment set, and space set respectively:

[0031] Personnel set: During the monitoring period, the number of personnel units in the personnel set is monitored in real time: When a staff member enters the monitoring area, the facial image of the staff member collected by the AI ​​image recognition camera is used. If the recognition is successful, the total number of personnel in the monitoring area RY is increased by 1, and the personnel flow value RL is recorded as plus 1. If the recognition fails, it is prompted that non-staff members have entered the monitoring area, and the on-site operation has hidden dangers. A hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; when the staff member leaves the monitoring area, the facial image of the staff member in the monitoring area is used to recognize the AI ​​image, and the total number of personnel units r in the monitoring area at the current time is obtained and recorded as the total number of personnel RY, and it is compared with the total number of personnel RY obtained last time. The difference is recorded as the personnel flow value RL; the personnel number threshold RYmax of the monitoring area is obtained. When the total number of personnel RY is greater than or equal to the personnel number threshold RYmax, it is judged that the personnel in the monitoring area are redundant, and it is prompted that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; the personnel flow threshold RLmax of the monitoring area is obtained, and the absolute value of all the obtained personnel flow values ​​RL is summed to obtain the personnel flow RYL. When RYL is greater than or equal to the personnel flow threshold RLmax, it is judged that the personnel flow in the monitoring area is large, and it is prompted that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager;

[0032] Equipment set: During the monitoring period, the operation and use of equipment unit s are monitored regularly. The front image of the equipment is collected by the AI ​​image recognition camera. If the recognition is successful, the number of correct recognitions SZ is increased by 1. If the recognition fails, the number of incorrect recognitions SC is increased by 1. The recognition accuracy Ps of the equipment unit s is obtained by numerical calculation using the formula Ps=SZ / (SZ+SC); the recognition accuracy Ps of all equipment units s in the equipment set are summed and averaged to obtain the average recognition accuracy P; the equipment abnormality rate threshold PY is obtained. If the average recognition accuracy P is greater than or equal to the equipment abnormality rate threshold PY, it is judged that the equipment is operating normally and there are no hidden dangers in the on-site operation; if the average recognition accuracy P is less than the equipment abnormality rate threshold PY, it is judged that the equipment is operating abnormally, indicating that there are hidden dangers in the on-site operation, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager;

[0033] Spatial set: During the monitoring period, the occupied area of ​​objects in the spatial unit k is monitored in real time. In particular, the objects are common sundries other than staff and equipment, such as discarded materials, construction tools, cleaning supplies, etc.; the image of the object in the spatial unit k is collected by the AI ​​image recognition camera, and the sum of the area of ​​the object image in the spatial unit k is recorded as the object image occupied area Wk, and the image area Sk of the spatial unit k is obtained. The object image occupied area Wk and the image area Sk of the spatial unit k are numerically calculated by the formula Mk=Wk / Sk to obtain the density Mk of the spatial unit k. The density Mk of all the spatial units k in the spatial set is summed and averaged to obtain the average density Mk. The average density M is obtained; the spatial density threshold MY is obtained. If the average density M is greater than or equal to the spatial density threshold MY, the space in the monitoring area is judged to be dense and disorderly, indicating that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; if the average density M is less than the spatial density threshold, the space in the monitoring area is judged to be spacious and orderly, and the on-site operation does not have hidden dangers; the total number of personnel RY, personnel flow RL, average equipment recognition accuracy P and average density M of the space at the operation site are monitored in real time through AI image recognition technology, and multi-source data are fully combined to alarm the hidden dangers of on-site operations, thereby improving the accuracy and timeliness of identifying hidden dangers in on-site operations;

[0034] Step 4: According to step 3, the number of hidden danger alarm signals generated by personnel sets, equipment sets and space sets is counted in real time, and the number of hidden danger alarm signals generated by personnel sets, equipment sets and space sets during the monitoring period is marked as personnel hidden danger number RH, equipment hidden danger number SH and space hidden danger number KH respectively; obtain the personnel hidden danger threshold RHmax, equipment hidden danger threshold SHmax and space hidden danger threshold KHmax, and compare the personnel hidden danger number RH, equipment hidden danger number SH and space hidden danger number KH with the corresponding personnel hidden danger threshold RHmax, equipment hidden danger threshold SHmax and space hidden danger threshold KHmax respectively: if at least one indicator exceeds the preset threshold , it is judged that the hidden dangers of the on-site operation exceed the standard, and a hidden danger exceeding standard alarm signal is generated and sent to the mobile phone terminal of the manager; during the monitoring period, the number of personnel hidden dangers RH, the number of equipment hidden dangers SH and the number of space hidden dangers KH are numerically calculated by the formula YX=k1×RH+k2×SH+k3×KH to obtain the hidden danger warning coefficient YX, where k1, k2 and k3 are all proportional coefficients, and k3>k2>k1>1; the hidden danger warning coefficient YX is compared with the preset hidden danger warning threshold YXmax: if the hidden danger warning coefficient YX is greater than or equal to the hidden danger warning threshold YXmax, a hidden danger warning signal is generated and sent to the mobile phone terminal of the manager;

[0035] Step 5: Generate an early warning optimization analysis cycle with an irregular duration, and count the hidden danger warning coefficient YX at the time when the hidden danger exceeding alarm signal is generated within the early warning optimization analysis cycle to establish a comprehensive early warning set, in which the elements are recorded as hidden danger warning coefficients YXn, n=1, 2,..., i, i is a positive integer; calculate all elements in the comprehensive early warning set to obtain the standard deviation S and the average value X, and calculate the statistic Tn through the formula Tn=(YXn-X) / S; obtain the critical value Ta by looking up the table, if Tn≥Ta, it means that the hidden danger warning coefficient YXn is a discrete value and needs to be discarded, otherwise it is retained; and mark the minimum element in the comprehensive early warning set after discarding the discrete value as the hidden danger warning threshold YXmax of the next analysis cycle, and directly call the updated hidden danger warning threshold YXmax when performing hidden danger warning in the next analysis cycle; in the early warning optimization analysis cycle, by counting and screening the hidden danger warning coefficient YX at the time when the hidden danger exceeding alarm signal is generated, the hidden danger warning threshold YXmax is periodically updated, thereby improving the accuracy of early warning of hidden dangers in on-site operations.

[0036] A method for AI image recognition of hidden dangers in on-site operations, during operation, the on-site operation area is marked as a monitoring area, the image of the monitoring area is segmented to obtain a number of monitoring units, each monitoring unit is identified by AI image and classified into a personnel set, an equipment set and a space set; the operation time period of each working day is marked as a monitoring time period, and within the monitoring time period, the total number of personnel RY and the personnel flow value RL in the personnel set, the average recognition accuracy P of the equipment set and the average density M of the space set are monitored in real time, and the occurrence of hidden dangers in the monitoring area is analyzed and alarmed; then the number of hidden danger alarms generated by the personnel set, the equipment set and the space set within the monitoring time period is obtained, and the hidden danger exceeding the standard is further analyzed and alarmed, and a hidden danger warning coefficient YX is generated, and the hidden danger is analyzed according to the hidden danger warning coefficient YX; finally, a warning optimization analysis cycle is generated, and the hidden danger warning threshold YXmax is periodically updated.

[0037] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

[0038] The above formulas are obtained by collecting a large amount of data for software simulation and selecting a formula close to the real value. The coefficients in the formula are set by technicians in this field according to the actual situation; for example: formula YX=k1×RH+k2×SH+k3×KH; technicians in this field collect multiple groups of sample data and set corresponding hidden danger warning coefficients for each group of sample data; substitute the set hidden danger warning coefficients and the collected sample data into the formula, any three formulas constitute a three-variable linear equation group, screen the calculated coefficients and take the average, and obtain the values ​​of k1, k2 and k3 as 3.44, 2.17 and 1.85 respectively;

[0039] The size of the coefficient is to quantify each parameter to obtain a specific value for subsequent comparison. The size of the coefficient depends on the amount of sample data and the initial setting of the corresponding hidden danger warning coefficient for each set of sample data by technical personnel in this field; as long as it does not affect the proportional relationship between the parameter and the quantified value, such as the hidden danger warning coefficient is proportional to the value of the number of personnel hidden dangers.

[0040] 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 present 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.

[0041] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. 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 present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for AI image recognition of hidden dangers in on-site operations, characterized in that: The following steps are involved: Step 1: Mark the field operation area as the monitoring area, mark the operation time period of each working day as the monitoring time period, obtain the image of the monitoring area and divide it into several monitoring units, including personnel unit r, equipment unit s and space unit k; Step 2: During the monitoring period, AI images are used to identify each monitoring unit and categorize them into personnel sets, equipment sets, and space sets; Step 3: Evaluate the hidden dangers in the personnel set, equipment set and space set respectively: During the monitoring period, monitor the total number of personnel RY and personnel flow value RL in the personnel set, the number of correct identification times SZ and the number of incorrect identification times SC in the equipment set, and the image area Sk of the space unit k in the space set and the area occupied by the object image Wk in real time, analyze the hidden dangers in the monitoring area and give an alarm; Step 4: During the monitoring period, the number of hidden danger alarm signals generated by the personnel set, equipment set and space set is obtained in real time, and marked as the number of hidden dangers RH, the number of hidden dangers SH and the number of hidden dangers KH respectively, and the hidden danger exceeding the standard is further analyzed and alarmed, and the hidden danger warning coefficient YX is generated, and the hidden danger warning analysis is carried out according to the hidden danger warning coefficient YX; Step 5: Generate an early warning optimization analysis cycle, and establish a comprehensive early warning set with the hidden danger warning coefficient YX at the time when the hidden danger exceeding the standard alarm signal is generated within the early warning optimization analysis cycle. The elements in the set are recorded as hidden danger warning coefficients YXn, n=1, 2, ..., i, i is a positive integer, and the hidden danger warning threshold YXmax is periodically updated according to the result after discrete value elimination of the comprehensive early warning set; In step 1, a camera is used to take a picture of the monitoring area during the monitoring period to obtain an image of the monitoring area, which is then enlarged and gray-scaled. The monitoring area is segmented by a binary method to obtain a number of monitoring units. The attributes of the monitoring units include personnel, equipment, and space. The monitoring units with the attributes of personnel, equipment, and space are marked as personnel unit r, equipment unit s, and space unit k, respectively. In step 2, the facial images of the staff and the front images of the equipment for the operation are extracted, and the images and names are stored in a local database in a one-to-one correspondence; the facial images of the staff and the front images of the equipment in the local database and the existing data of the AI ​​model constitute a training set; based on the training set, AI image recognition is used to identify and classify the various monitoring units in the monitoring area, and the monitoring units with the attribute of personnel are classified into the personnel set, the monitoring units with the attribute of equipment are classified into the equipment set, and the monitoring units with the attribute of space are classified into the space set; In step three, the hidden danger occurrence assessment process in the personnel set includes: when a staff member enters the monitoring area, the facial image of the staff member collected by the AI ​​image recognition camera is used. If the recognition is successful, the total number of personnel in the monitoring area RY is increased by 1, and the personnel flow value RL is recorded as increased by 1. If the recognition fails, it is prompted that non-staff members have entered the monitoring area, and the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; when the staff member leaves the monitoring area, the facial image of the staff member in the monitoring area is used to recognize the AI ​​image, and the total number of personnel units r in the monitoring area at the current time is obtained and recorded as the total number of personnel RY, and the difference with the total number of personnel RY obtained last time is recorded as the personnel flow value RL; the threshold value RYmax of the number of personnel in the monitoring area is obtained. When the total number of personnel RY is greater than or equal to the threshold value RYmax of the number of personnel, it is judged that the personnel in the monitoring area is redundant, and it is prompted that the on-site operation has hidden dangers, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; Obtain the personnel flow threshold RLmax of the monitoring area, take the absolute value and sum all the obtained personnel flow values ​​RL, and obtain the personnel flow amount RYL. When RYL is greater than or equal to the personnel flow threshold RLmax, it is judged that the personnel flow in the monitoring area is large, and it is prompted that there are hidden dangers in the on-site operation, and a hidden danger alarm signal is generated and sent to the mobile phone terminal of the manager; In step 3, the hidden danger occurrence assessment process in the equipment set includes: using the front image of the equipment collected by the AI ​​image recognition camera, if the recognition is successful, the number of correct recognition times SZ is increased by 1, if the recognition fails, the number of incorrect recognition times SC is increased by 1, and the recognition accuracy Ps of the equipment unit s is obtained by numerical calculation through the formula Ps=SZ / (SZ+SC); the recognition accuracy Ps of all equipment units s in the equipment set are summed and averaged to obtain the average recognition accuracy P; The device abnormality rate threshold PY is obtained. If the average recognition accuracy rate P is greater than or equal to the device abnormality rate threshold PY, it is judged that the equipment is operating normally and there are no hidden dangers in the on-site operation. If the average recognition accuracy rate P is less than the device abnormality rate threshold PY, it is judged that the equipment is operating abnormally, indicating that there are hidden dangers in the on-site operation, generating a hidden danger alarm signal and sending the hidden danger alarm signal to the mobile phone terminal of the manager.

2. The method of AI image recognition of hidden dangers in on-site operations according to claim 1, characterized in that: In step three, the hidden danger occurrence assessment process in the space set includes: using the AI ​​image recognition camera to collect the object image in the space unit k, recording the sum of the object image areas in the space unit k as the object image occupied area Wk, obtaining the image area Sk of the space unit k, and numerically calculating the object image occupied area Wk and the image area Sk of the space unit k through the formula Mk=Wk / Sk to obtain the density Mk of the space unit k, summing and averaging the density Mk of all the space units k in the space set to obtain the average density M; obtaining the space density threshold MY, if the average density M is greater than or equal to the space density threshold MY, it is judged that the space in the monitoring area is dense and messy, indicating that the on-site operation has hidden dangers, generating a hidden danger alarm signal and sending the hidden danger alarm signal to the mobile phone terminal of the manager; if the average density M is less than the space density threshold, it is judged that the space in the monitoring area is spacious and orderly, and the on-site operation does not have hidden dangers.

3. The method of AI image recognition of hidden dangers in on-site operations according to claim 2, characterized in that: In step 4, the personnel hidden danger threshold RHmax, the equipment hidden danger threshold SHmax and the space hidden danger threshold KHmax are obtained, and the number of personnel hidden dangers RH, the number of equipment hidden dangers SH and the number of space hidden dangers KH are compared with the corresponding personnel hidden danger threshold RHmax, the equipment hidden danger threshold SHmax and the space hidden danger threshold KHmax respectively: if at least one indicator exceeds the preset threshold, it is judged that the hidden danger of the on-site operation exceeds the standard, and a hidden danger exceeding standard alarm signal is generated and the hidden danger exceeding standard alarm signal is sent to the mobile phone terminal of the manager; within the monitoring time period, the number of personnel hidden dangers RH, the number of equipment hidden dangers SH and the number of space hidden dangers KH are numerically calculated by the formula YX=k1×RH+k2×SH+k3×KH to obtain the hidden danger warning coefficient YX, where k1, k2 and k3 are all proportional coefficients, and k3>k2>k1>1; the hidden danger warning coefficient YX is compared with the preset hidden danger warning threshold YXmax: if the hidden danger warning coefficient YX is greater than or equal to the hidden danger warning threshold YXmax, a hidden danger warning signal is generated and sent to the mobile phone terminal of the manager.

4. The method of AI image recognition of hidden dangers in on-site operations according to claim 3, characterized in that: In step 5, all elements in the comprehensive warning set are calculated to obtain the standard deviation S and the average value X, and the statistic Tn is calculated by the formula Tn=(YXn-X) / S; the critical value Ta is obtained by looking up the table. If Tn≥Ta, it means that the hidden danger warning coefficient YXn is a discrete value and needs to be discarded, otherwise it is retained; The minimum element in the comprehensive warning set after discarding discrete values ​​is marked as the hidden danger warning threshold value YXmax of the next analysis cycle, and the updated hidden danger warning threshold value YXmax is directly called when performing hidden danger warning in the next analysis cycle.

Images