Field operation scene recognition and operation behavior intelligent analysis system based on deep learning

Abstract

The invention discloses a field operation scene recognition and operation behavior intelligent analysis system based on deep learning, and belongs to the technical field of power grid operation digital management. Carrying out multi-direction camera shooting and image processing and analysis on the field operation area to obtain target recognition features, and carrying out data processing and calculation on the target recognition features to obtain the abnormality of the operation area; according to the method, monitoring and data processing are carried out from an operation specification dimension, all job operation abnormal data obtained through processing are integrated and calculated, data analysis is carried out on job operation anomalies to obtain corresponding operation identification marks and operation state instructions, and risk analysis data of different dimensions are integrated and calculated, so that the risk analysis accuracy is improved. And according to the operation risk identifier obtained through analysis, risk dynamic prompt and safety dynamic management and control are carried out on field operation, so that field operation scene identification and operation behavior automatic judgment on whether the operation is in a specified interval and whether the operation accords with an operation sequence are realized, and illegal behaviors are automatically identified.

Description

Technical Field

[0001] This invention relates to the field of digital management technology for power grid operations, specifically to a deep learning-based system for identifying on-site operation scenarios and intelligently analyzing operational behaviors. Background Technology

[0002] With the development of technology, the power grid field operation process needs to be perceived, understood, judged and warned in real time, so as to improve the safety, standardization and efficiency of the operation.

[0003] Existing power operation monitoring schemes cannot use deep learning models to analyze on-site video streams to automatically identify and match elements such as equipment intervals, switch positions, personnel postures, and operational actions. They also cannot automatically determine whether operations are within specified intervals or conform to the operational sequence, or automatically identify violations. Summary of the Invention

[0004] The purpose of this invention is to provide a deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior, which solves the technical problems of poor autonomous recognition and evaluation of on-site operations and operational behavior and poor autonomous control in existing solutions.

[0005] The objective of this invention can be achieved through the following technical solutions: A deep learning-based system for identifying on-site work scenarios and intelligently analyzing operational behaviors includes: The image data acquisition and processing module for the work area is used to take comprehensive monitoring point videos of the target area where the on-site operation is carried out, obtain monitoring images of the corresponding locations, sort and combine them to obtain a regional monitoring combination set, and preprocess and match the monitoring images in the regional monitoring combination set to obtain regional risk analysis data. The work procedure image data acquisition and processing module is used to capture frontal images of operators performing on-site operations to obtain corresponding operation images, preprocess and match the operation images, and preprocess and analyze the operators' voice data to obtain operation risk analysis data. The Operation Status Integration Assessment and Management Module is used to integrate and calculate the regional risk analysis data corresponding to the target area where the on-site operation is carried out and the operation risk analysis data corresponding to the operator to obtain the corresponding operation risk integration value. The operation risk integration value is analyzed to determine the overall operation risk corresponding to the on-site operation and to implement targeted safety control.

[0006] Preferably, several monitoring images of the regional monitoring combination are preprocessed and feature extracted, and all the identification features obtained from the processing of all monitoring images are sorted and combined to obtain an identification feature set; When performing anomaly impact analysis on the identified feature set, the total number of identified features is counted and analyzed. If the total number of identified features is 0, a normal operation instruction is generated for the operation area and the impact flag of the abnormal feature is set to 0.

[0007] Preferably, if the total number of identification features is not 0, an abnormal operation area instruction is generated and several identification features in the identification feature set are traversed with a preset sample feature database. If there is a sample feature in the sample feature database that is the same as the identification feature, the corresponding identification feature is marked as the target identification feature. If no sample feature with the same recognition feature exists in the sample feature database, the corresponding recognition feature will be marked as a normal recognition feature. The total number n of target recognition features is counted and analyzed. If the total number of target recognition features is 0, a normal operation area instruction is generated and a prompt is displayed.

[0008] Preferably, if the total number of target identification features is not 0, a data processing instruction is generated, and the feature influence coefficient yi associated with different target identification features is obtained according to the data processing instruction, i=1, 2, 3, ..., n; n is a positive integer; i is different target identification features, and the anomaly degree of the work area corresponding to all target identification features is obtained by calculation. Analyze the anomaly degree of the work area to determine the corresponding anomaly impact of the work area and obtain the work area anomaly low impact instruction or work area anomaly high impact instruction. The data for regional risk analysis consists of normal instructions for the work area, the degree of abnormality in the work area, and instructions with low or high impact from abnormalities in the work area obtained through analysis.

[0009] Preferably, standard operation features are obtained according to the type of on-site operation, operation images are preprocessed and feature recognition is performed to obtain operation recognition features, all operation recognition features obtained by recognition are numbered and marked, and all numbered and marked operation recognition features are sequentially matched with the sample operation feature set, and sample operation features that are the same as operation recognition features are marked as valid features; and sample operation features that do not match in the sample operation feature set are marked as abnormal features. The total number of identification features N1 corresponding to all operation identification features and the total number of sample features N2 corresponding to all sample operation features in the sample operation feature set are counted. The data is then analyzed using the operation feature identification piecewise function, and the operation identification identifier is output.

[0010] Preferably, when determining the corresponding operation status of the operator based on the operation identification mark, the operation identification mark is analyzed. If the operation identification mark is 0, a normal operation instruction is generated and prompted. If the operation identification identifier is not 0, the operation identification identifier is compared and analyzed with the operation identification threshold. If the operation identification identifier is less than or equal to the operation identification threshold, a minor operation abnormality instruction is generated and a prompt is displayed. If the operation identification value is greater than the operation identification threshold, a severe operation anomaly command will be generated and a prompt will be displayed. Operational anomaly levels, operation identification markers, and the analysis of normal operation instructions, slightly abnormal operation instructions, or severely abnormal operation instructions constitute operational risk analysis data.

[0011] Preferably, the regional risk analysis data and operational risk analysis data corresponding to the on-site operation are obtained and analyzed in a traversal manner. If the results of the traversal statistics do not simultaneously contain normal instructions for the work area and normal instructions for operation, an integrated instruction is generated. Based on the integrated instruction, the abnormality degree of the work area in the regional risk analysis data and the abnormality degree of the operation in the operational risk analysis data are obtained, and the integrated value of the operational risk corresponding to the on-site operation is obtained by calculation.

[0012] Preferably, when determining the overall risk of the on-site operation based on the integrated operation risk value, the integrated operation risk value is input into the operation risk identification segmentation function for data analysis and outputs the operation risk identifier. Based on the operation risk identifier, a corresponding overall operation risk warning is generated, and targeted safety control is implemented for the on-site operation.

[0013] Preferably, a corresponding low-risk or high-risk overall operation prompt is generated based on the operation risk indicator with a value of 1 or 2; and a remote on-site operation management prompt is given to the safety management personnel based on the operation risk indicator with a value of 1 or 2.

[0014] Preferably, the operator's voice data is preprocessed, and the preprocessed voice features are matched against a preset standard voice feature set. Speech features that do not exist in the standard speech feature set are marked as valid features, while speech features that exist in the standard speech feature set are marked as abnormal features.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention obtains target recognition features by taking multi-directional photos of the on-site work area and processing and analyzing the images. It then processes and calculates the anomaly degree of the work area through data processing of the target recognition features. This not only provides a digital representation of the impact of anomalies in the on-site work area, but also provides reliable work area risk data support for the processing and analysis of the overall risk of subsequent operations, thereby improving the diversity of automatic identification, processing and utilization of dimensional risks in the on-site work area.

[0016] This invention monitors and processes data from the perspective of operational standards. It integrates and calculates all abnormal operational data to obtain the operational abnormality degree, and analyzes the operational abnormality degree to obtain corresponding operational identification marks and operational status instructions. This not only monitors and digitally represents the operational standardization of operators, but also provides reliable operational standard data support for the subsequent analysis and handling of overall operational risks, thereby improving the diversity and comprehensiveness of automatic risk identification and handling at the operational standard level.

[0017] This invention integrates and calculates risk analysis data from different dimensions, enabling a digital representation of the overall operational risk of on-site operations. Based on the operational risk identifiers obtained from the analysis, it provides dynamic risk alerts and dynamic safety control for on-site operations, thereby improving the effectiveness of autonomous identification, assessment, and control of on-site operations and behaviors. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the operation of the deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, the present invention is a deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior, including: an operation area image data acquisition and processing module, an operation specification image data acquisition and processing module, and an operation status integration evaluation and management module; The work area image data acquisition and processing module is used to capture comprehensive images of the target area where the work is being carried out, obtain monitoring images of the corresponding locations, sort and combine them to obtain a regional monitoring combination set, and perform preprocessing and matching analysis on several monitoring images in the regional monitoring combination set to obtain regional risk analysis data; including: The target area is the area where on-site operations will be carried out. The monitoring points are a number of monitoring points set in different directions at a preset interval, with the midpoint of the target area as the monitoring points. The different directions include, but are not limited to, due east, due west, due north, and due south. The interval is determined according to the existing operation area requirements data corresponding to the on-site operations. The purpose of setting up a number of monitoring points is to obtain all environmental images of the target area during the on-site operations. The environmental images obtained by the cameras can provide reliable image data support for the automatic identification and analysis of risks in the subsequent operation area. It is important to note that combustible materials should be removed from the area around the work site to prevent the spread of open flames or ignition of surrounding objects. Several monitoring images from the regional monitoring combination are preprocessed and feature extracted. All the identification features obtained from the processing of all monitoring images are sorted and combined to obtain an identification feature set. It should be noted that the image preprocessing and feature extraction of several monitoring images are existing conventional technical solutions, and the specific steps will not be elaborated here; in addition, the feature extraction is the feature of foreign objects on the ground in the monitoring images. When performing anomaly impact analysis on the identification feature set, the identification feature set is traversed to count the total number of identification features and analyzed. If the total number of identification features is 0, a normal operation instruction is generated for the operation area and the impact flag of the abnormal feature is set to 0. If the total number of identification features is not 0, an abnormal operation area instruction is generated and several identification features in the identification feature set are traversed with the preset sample feature database. If there is a sample feature in the sample feature database that is the same as the identification feature, the corresponding identification feature is marked as the target identification feature. The sample feature database has several pre-set sample features, which are determined based on existing on-site operational environment requirements data; the sample features are specifically the features corresponding to different combustibles. If no sample feature with the same recognition feature exists in the sample feature database, the corresponding recognition feature will be marked as a normal recognition feature. The total number n of target identification features is counted and analyzed. If the total number of target identification features is 0, a normal operation area instruction is generated and a prompt is given. It can be understood that although there are foreign objects in the target area, if the foreign objects are not flammable, the environmental requirements for on-site operation are met. If the total number of target recognition features is not 0, a data processing instruction is generated, and the feature influence coefficient yi associated with different target recognition features is obtained according to the data processing instruction, i=1, 2, 3, ..., n; n is a positive integer; i represents different target recognition features, and the coefficient is obtained through the formula... Calculate the anomaly degree Zy of the work area corresponding to all target identification features; where NJj is the detection concentration of different combustible gases in the on-site work area, which can be obtained by existing gas detection sensors; NJj0 is the standard concentration of different combustible gases in the on-site work area, which is determined according to existing detection standard requirements; j = 1, 2, 3, ..., m; m is a positive integer; j represents different combustible gases; specifically, combustible gases include, but are not limited to, methane, propane, and acetylene, which can be obtained by existing detection instruments; Among them, different sample features are pre-set with a corresponding feature influence coefficient. The feature influence coefficient is used to digitally and differentiate the abnormal influence of different sample features. The specific value of the feature influence coefficient can be set by professionals in this field based on their work experience and work requirements data. In addition, the anomaly degree of the work area is used to integrate and calculate different abnormal data from the image recognition and analysis of the work area on site to digitally represent its abnormal impact; the larger the anomaly degree of the work area, the greater the abnormal impact of the work area. Analyze the anomaly degree of the work area to determine the corresponding anomaly impact; If the anomaly degree of the work area is less than or equal to 1, a low impact instruction for the work area anomaly will be generated and a prompt will be displayed. If the anomaly degree of the work area is greater than 1, a high impact command for the work area anomaly will be generated and a prompt will be displayed. The normal instructions for the work area, the abnormality of the work area, and the low-impact or high-impact instructions for the work area obtained through analysis constitute the regional risk analysis data. Unlike existing technical solutions that rely on manual identification and assessment of risks in on-site work areas, this invention uses multi-directional video recording, image processing, and analysis to obtain target identification features. These features are then processed and calculated to determine the anomaly degree of the work area. This approach not only digitally represents the impact of anomalies in the work area but also provides reliable risk data support for subsequent overall risk management and analysis, thus enhancing the versatility of automatic risk identification and utilization in on-site work areas.

[0022] The work procedure image data acquisition and processing module is used to capture frontal images of operators performing on-site operations, obtaining corresponding operation images. This can be achieved using on-site surveillance video equipment, which records the entire on-site operation process. The module preprocesses and matches the operation images, and preprocesses and analyzes the operators' voice data to obtain operation risk analysis data; including: Standard operation features are obtained based on the type of on-site operation. Operation images are preprocessed and feature recognition is performed to obtain operation recognition features. All operation recognition features obtained are numbered and marked. All numbered and marked operation recognition features are sequentially matched with the sample operation feature set. Sample operation features that are the same as the operation recognition features are marked as valid features. Sample operation features that do not match in the sample operation feature set are marked as abnormal features. The preprocessing and feature extraction of several monitoring images are conventional existing technical solutions, and the specific steps will not be elaborated here. In addition, the operator's voice data is preprocessed, and the preprocessed voice features are matched against a preset standard voice feature set. Speech features that are not present in the standard speech feature set are marked as valid features, while speech features that are present in the standard speech feature set are marked as abnormal features. The processing and feature extraction of the operator's speech data are existing conventional technical solutions, and the specific implementation steps are not described here. The total number of identification features corresponding to all operation identification features is N1, and the total number of sample features corresponding to all sample operation features in the sample operation feature set is N2. The data is then analyzed using the operation feature identification piecewise function, and the operation identification identifier is output. The expression for the piecewise function for operation feature recognition is as follows: ; in, In the formula, Zk is the feature influence factor corresponding to different abnormal features; k is different abnormal features, k=1, 2, 3, ..., N2-N1; Cy is the degree of abnormality of operation; It should be noted that different sample operation characteristics are pre-set with a corresponding feature impact factor. The feature impact factor is used to digitally and differentiate the feature impact corresponding to the sample operation characteristics. The specific value of the feature impact factor can be determined based on the total number of accidents caused by different sample operation characteristics in history, or it can be determined by the staff in this field based on their work experience and work requirements data. Operation anomaly rate is used to integrate and calculate all operation anomaly data to digitally represent its overall impact on the operation; operation identification label is used to classify and digitally represent the operation anomaly rate. When determining the operator's corresponding operation status based on the operation identification mark, the operation identification mark is analyzed. If the operation identification mark is 0, a normal operation instruction is generated and prompted. If the operation identification identifier is not 0, the operation identification identifier is compared and analyzed with the operation identification threshold. If the operation identification identifier is less than or equal to the operation identification threshold, a minor operation abnormality instruction is generated and prompted. The operation identification threshold is determined based on the total number of abnormal operations due to non-standard operation in the past. If the operation identification value is greater than the operation identification threshold, a severe operation anomaly command will be generated and a prompt will be displayed. Operational anomaly level, operation identification marks, and normal operation instructions, slightly abnormal operation instructions, or severely abnormal operation instructions obtained through analysis constitute operational risk analysis data; In this embodiment of the invention, by monitoring and processing data from the perspective of operational standards, all abnormal operation data obtained through processing are integrated and calculated to obtain the operational abnormality degree. Data analysis is then performed on the operational abnormality degree to obtain the corresponding operation identification mark and operation status instruction. This not only monitors and digitally represents the operator's operational standardization, but also provides reliable operational standard data support for the subsequent overall risk processing and analysis of operations, thereby improving the diversity and comprehensiveness of automatic risk identification and processing utilization from the perspective of operational standards.

[0023] The operational status integration assessment and management module is used to integrate and calculate the regional risk analysis data corresponding to the target area where on-site operations are carried out, as well as the operational risk analysis data corresponding to the operators, to obtain the corresponding operational risk integration value. It then analyzes this integrated risk value to determine the overall operational risk for the on-site operation and implements targeted safety controls. This includes: Obtain and analyze the regional risk analysis data and operational risk analysis data corresponding to the on-site operation. If the results of the traversal and statistics show both normal instructions for the work area and normal instructions for operation, then generate a normal instruction. If the results of the traversal statistics do not simultaneously contain both normal work area instructions and normal operation instructions, an integrated instruction is generated. Based on the integrated instruction, the work area anomaly degree Zy in the regional risk analysis data and the operation anomaly degree Cy in the operation risk analysis data are obtained. The integrated operation risk value Zf corresponding to the on-site operation is calculated using the formula Zf=α×Zy+β×Cy; where α and β are different weighting coefficients, and 0<β<α. It should be noted that the operational risk integration value is a numerical value that is a digital representation of the overall risk of on-site operations by integrating and calculating risk analysis data from different dimensions. When determining the overall risk of a field operation based on the integrated risk value, the integrated risk value is input into the operation risk identification segmentation function for data analysis and outputs an operation risk identifier. Based on the operation risk identifier, a corresponding overall operation risk warning is generated, and targeted safety control is implemented for the field operation. The expression for the piecewise function for identifying operational risks is as follows: In the formula, Y is the job risk classification value, and the specific value of the job risk classification value is determined based on the median of the integrated values ​​of all job risks tested in the previous period. Based on the operation risk indicator with a value of 1 or 2, generate corresponding overall low risk or overall high risk warnings for the operation. Additionally, based on the operational risk indicator with a value of 1 or 2, remote on-site operation management prompts are provided to safety management personnel.

[0024] In this embodiment of the invention, by integrating and calculating risk analysis data from different dimensions, the overall operational risk of on-site operations can be digitally represented. Based on the operational risk identifiers obtained from the analysis, dynamic risk warnings and dynamic safety management of on-site operations can be provided, thereby improving the autonomous identification and assessment and autonomous management effects of on-site operations and behaviors.

[0025] Furthermore, the formulas mentioned above are all numerical calculations obtained by removing dimensions and using simulation software to obtain a formula that is closest to the real situation, based on the collection of a large amount of data.

[0026] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0027] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0028] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0029] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based system for identifying on-site work scenarios and intelligently analyzing operational behavior, characterized in that: include: The image data acquisition and processing module for the work area is used to take comprehensive monitoring point videos of the target area where the on-site operation is carried out, obtain monitoring images of the corresponding locations, sort and combine them to obtain a regional monitoring combination set, and preprocess and match the monitoring images in the regional monitoring combination set to obtain regional risk analysis data. The work procedure image data acquisition and processing module is used to capture frontal images of operators performing on-site operations to obtain corresponding operation images, preprocess and match the operation images, and preprocess and analyze the operators' voice data to obtain operation risk analysis data. The Operation Status Integration Assessment and Management Module is used to integrate and calculate the regional risk analysis data corresponding to the target area where the on-site operation is carried out and the operation risk analysis data corresponding to the operator to obtain the corresponding operation risk integration value. The operation risk integration value is analyzed to determine the overall operation risk corresponding to the on-site operation and to implement targeted safety control.

2. The deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior as described in claim 1, characterized in that, Several monitoring images from the regional monitoring combination are preprocessed and feature extracted. All the identification features obtained from the processing of all monitoring images are sorted and combined to obtain an identification feature set. When performing anomaly impact analysis on the identified feature set, the total number of identified features is counted and analyzed. If the total number of identified features is 0, a normal operation instruction is generated for the operation area and the impact flag of the abnormal feature is set to 0.

3. The deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior according to claim 2, characterized in that, If the total number of identification features is not 0, an abnormal operation area instruction is generated and several identification features in the identification feature set are traversed with the preset sample feature database. If there is a sample feature in the sample feature database that is the same as the identification feature, the corresponding identification feature is marked as the target identification feature. If no sample feature with the same recognition feature exists in the sample feature database, the corresponding recognition feature will be marked as a normal recognition feature. The total number n of target recognition features is counted and analyzed. If the total number of target recognition features is 0, a normal operation area instruction is generated and a prompt is displayed.

4. The on-site operation scene recognition and intelligent analysis system based on deep learning according to claim 3, characterized in that, If the total number of target identification features is not 0, a data processing instruction is generated, and the feature influence coefficient yi associated with different target identification features is obtained according to the data processing instruction, i=1,2,3,...,n; n is a positive integer; i is a different target identification feature, and the anomaly degree of the work area corresponding to all target identification features is obtained by calculation. Analyze the anomaly degree of the work area to determine the corresponding anomaly impact of the work area and obtain the work area anomaly low impact instruction or work area anomaly high impact instruction. The data for regional risk analysis consists of normal instructions for the work area, the degree of abnormality in the work area, and instructions with low or high impact from abnormalities in the work area obtained through analysis.

5. The on-site operation scene recognition and intelligent analysis system based on deep learning according to claim 1, characterized in that, Standard operation features are obtained based on the type of on-site operation. Operation images are preprocessed and feature recognition is performed to obtain operation recognition features. All operation recognition features obtained are numbered and marked. All numbered and marked operation recognition features are sequentially matched with the sample operation feature set. Sample operation features that are the same as the operation recognition features are marked as valid features. Sample operation features that do not match in the sample operation feature set are marked as abnormal features. The total number of identification features N1 corresponding to all operation identification features and the total number of sample features N2 corresponding to all sample operation features in the sample operation feature set are counted. The data is then analyzed using the operation feature identification piecewise function, and the operation identification identifier is output.

6. The deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior according to claim 5, characterized in that, When determining the operator's corresponding operation status based on the operation identification mark, the operation identification mark is analyzed. If the operation identification mark is 0, a normal operation instruction is generated and prompted. If the operation identification identifier is not 0, the operation identification identifier is compared and analyzed with the operation identification threshold. If the operation identification identifier is less than or equal to the operation identification threshold, a minor operation abnormality instruction is generated and a prompt is displayed. If the operation identification value is greater than the operation identification threshold, a severe operation anomaly command will be generated and a prompt will be displayed. Operational anomaly levels, operation identification markers, and the analysis of normal operation instructions, slightly abnormal operation instructions, or severely abnormal operation instructions constitute operational risk analysis data.

7. The deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior according to claim 6, characterized in that, Obtain and analyze the regional risk analysis data and operational risk analysis data corresponding to the on-site operation. If the results of the traversal statistics do not simultaneously contain normal operation instructions and normal operation instructions for the operation area, generate an integrated instruction. Based on the integrated instruction, obtain the anomaly degree of the operation area in the regional risk analysis data and the anomaly degree of the operation in the operational risk analysis data, and calculate the integrated value of the operation risk corresponding to the on-site operation.

8. The on-site operation scene recognition and intelligent analysis system based on deep learning according to claim 7, characterized in that, When determining the overall risk of a field operation based on the integrated risk value, the integrated risk value is input into the operation risk identification segmentation function for data analysis and output of the operation risk identifier. Based on the operation risk identifier, a corresponding overall operation risk warning is generated, and targeted safety control is implemented for the field operation.

9. The on-site operation scene recognition and intelligent analysis system based on deep learning according to claim 8, characterized in that, Based on the operation risk indicator with a value of 1 or 2, generate corresponding overall low risk or overall high risk warnings for the operation. Additionally, based on the operational risk indicator with a value of 1 or 2, remote on-site operation management prompts are provided to safety management personnel.

10. The deep learning-based on-site operation scene recognition and intelligent analysis system for operational behavior according to claim 5, characterized in that, The operator's voice data is preprocessed, and the preprocessed voice features are matched against a preset standard voice feature set. Speech features that do not exist in the standard speech feature set are marked as valid features, while speech features that exist in the standard speech feature set are marked as abnormal features.

Images