Blasting operation site behavior compliance identification method, system, device and medium

Abstract

The application discloses a blasting operation site behavior compliance identification method, system, device and medium, and relates to the technical field of computers; the method comprises the following steps: acquiring multi-source video streams and environment monitoring data of a blasting operation site; identifying a current operation stage according to the multi-source video streams; identifying scene risk factors according to the multi-source video streams and the environment monitoring data; inputting the multi-source video streams into a pre-trained deep learning model to analyze identity information, behavior sequences, equipment wearing states and position information of each operator; calling a compliance rule set matched with the current operation stage and the scene risk factors in a pre-constructed safety rule library; performing compliance verification on the identity information, the behavior sequences, the equipment wearing states and the position information based on the compliance rule set; and if there is a rule violation in the verification, generating a rule violation judgment result containing a rule violation type, a risk level, associated personnel and a spatial area identifier. The application can realize intelligent monitoring and rule violation identification of the whole process of blasting operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology, in particular to a blasting operation site behavior compliance identification method, system, device and medium. BACKGROUND

[0002] Blasting operation is indispensable in the fields of mine, tunnel, building demolition and other engineering fields, but because it involves the transportation, storage, filling and detonation of dangerous goods such as explosives, and is often carried out in complex and high-risk environments, safety compliance is crucial. The state and industry standards have formulated strict specifications for the behavior of operating personnel, equipment wearing, regional control, etc. Any slight violation may cause serious accidents, resulting in loss of life and property and environmental damage.

[0003] However, the existing supervision means still has significant defects. Key links such as personnel responsibility separation (such as taking, issuing, and blasting by the same person) and identification clothing are often difficult to implement due to management oversight or resource allocation difficulties, and violations such as personnel gathering, unlicensed work, and equipment mismatch occur frequently. Traditional manual inspection or video monitoring has low efficiency and strong subjectivity, and it is difficult to achieve all-weather and non-missing monitoring. In addition, in the face of sudden scene changes (such as heavy rain, leakage), static safety rules cannot dynamically adjust the protection requirements, resulting in a lag in risk response. SUMMARY

[0004] To solve the problems in the prior art, the present application provides a blasting operation site behavior compliance identification method, system, device and medium, which can realize intelligent monitoring and violation identification of the whole process of blasting operation.

[0005] To achieve the above purpose, the present application provides a blasting operation site behavior compliance identification method, comprising:

[0006] Obtain multi-source video stream and environmental monitoring data of the blasting operation site; the environmental monitoring data includes temperature and humidity, wind speed and dangerous gas concentration;

[0007] Identify the current operation stage according to the multi-source video stream;

[0008] Identify scene risk factors according to the multi-source video stream and environmental monitoring data;

[0009] Input the multi-source video stream into a pre-trained deep learning model to parse the identity information, behavior sequence, equipment wearing state and position information of each operating personnel;

[0010] Call the compliance rule set in the pre-constructed safety rule library that matches the current operation stage and scene risk factors;

[0011] Based on the compliance rule set, the identity information, behavior sequence, equipment wearing state and location information are checked for compliance;

[0012] If there is a violation, a violation judgment result including the violation type, risk level, associated personnel and spatial area identifier is generated.

[0013] Optionally, the current work stage is identified according to the multi-source video stream, including:

[0014] The multi-source video stream is processed for spatio-temporal synchronization to generate a calibrated multi-view video sequence;

[0015] The device operation features and personnel cooperation features in the multi-view video sequence are extracted;

[0016] Based on the preset work stage conversion rule, the device operation features and personnel cooperation features are analyzed for time sequence correlation;

[0017] According to the correlation analysis result, the current work stage is determined.

[0018] Optionally, according to the multi-source video stream and environmental monitoring data, scene risk factors are identified, including:

[0019] The environmental monitoring data is compared with a preset safety threshold to generate an environmental risk level;

[0020] The multi-source video stream is parsed to extract environmental interference factor features of smoke density, rain and fog interference intensity and ground muddy area coverage range;

[0021] Based on a pre-constructed risk mapping rule, the environmental risk level and environmental interference factor features are fused to generate scene risk factors.

[0022] Optionally, the compliance rule set includes qualification rules, behavior rules, equipment rules and space rules; based on the compliance rule set, the identity information, behavior sequence, equipment wearing state and location information are checked for compliance, including:

[0023] According to the qualification rules, the identity information is checked, and when the work personnel qualification is invalid or the post permission is out of bounds, it is determined that there is a violation;

[0024] According to the behavior rules, the behavior sequence is compared, and when a key action is missing, operation is out of order or high-risk redundant action is identified, it is determined that there is a violation;

[0025] According to the equipment rules, the equipment wearing state is checked, and when the protective equipment is missing, invalid or does not match the current work stage requirement, it is determined that there is a violation;

[0026] According to the space rule comparison of the position information, when the job personnel is in a prohibited area, exceeds the limit of gathering, or does not evacuate according to the clearing requirement, it is determined that there is a violation behavior.

[0027] Optionally, the method further comprises:

[0028] When the violation judgment result is generated, a real-time video segment corresponding to the violation type, associated personnel and space area identifier in the multi-source video stream is extracted;

[0029] A timestamp and a space position mark are added to the real-time video segment, and are associated and bound with the violation judgment result, environmental monitoring data and current operation stage to generate an electronic evidence data package;

[0030] A digital signature and a hash value are added to the electronic evidence data package to generate an electronic evidence record that cannot be tampered with;

[0031] The electronic evidence record is stored in a tamper-proof storage system to form a traceable electronic evidence chain.

[0032] Optionally, the multi-source video stream is input to a pre-trained deep learning model, and before that, the method further comprises:

[0033] Point cloud data and on-site audio stream of the blasting operation site are acquired;

[0034] According to the environmental monitoring data and the multi-source video stream, the environmental interference intensity is determined;

[0035] When the environmental interference intensity exceeds a preset threshold, the multi-source video stream is input to a pre-trained deep learning model, specifically, the multi-source video stream, point cloud data and on-site audio stream are input to a pre-trained deep learning model;

[0036] The point cloud data is used to supplement the space positioning information, and the on-site audio stream is used to extract acoustic event features associated with standard operation actions.

[0037] The application also provides a blasting operation site behavior compliance identification system, comprising:

[0038] An acquisition unit is configured to acquire multi-source video stream and environmental monitoring data of a blasting operation site; the environmental monitoring data includes temperature and humidity, wind speed and dangerous gas concentration;

[0039] An operation stage identification unit is configured to identify a current operation stage according to the multi-source video stream;

[0040] A scene risk analysis unit is configured to identify scene risk factors according to the multi-source video stream and environmental monitoring data;

[0041] The personnel behavior analysis unit is configured to input the multi-source video stream into a pre-trained deep learning model to analyze identity information, behavior sequence, equipment wearing state and position information of each worker;

[0042] The rule dynamic scheduling unit is configured to call a compliance rule set matched with the current work stage and scene risk factor from a pre-constructed safety rule library;

[0043] The compliance verification unit is configured to verify the identity information, behavior sequence, equipment wearing state and position information based on the compliance rule set.

[0044] The violation decision generation unit is configured to generate a violation judgment result including a violation type, a risk level, associated personnel and a spatial area identifier if a violation behavior is found.

[0045] The present application also provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned identification method when executing the program.

[0046] The present application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the above-mentioned identification method.

[0047] According to the embodiments of the present application, the following technical effects are achieved:

[0048] The explosion work site behavior compliance identification method provided by the present application realizes full-dimensional real-time perception of the work scene by obtaining multi-source video stream and environment monitoring data (including temperature and humidity, wind speed and dangerous gas concentration) of the explosion work site, and completely solves the problems of low efficiency of traditional manual inspection and inability of video monitoring to automatically identify. The current work stage is identified by using multi-source video stream, and the scene risk factor is dynamically identified in combination with the environment monitoring data, which significantly improves the situation analysis capability in complex environments (such as rain, fog and smoke), and overcomes the defects of low identification accuracy and poor robustness caused by environmental interference in traditional methods.

[0049] By inputting the multi-source video stream into a pre-trained deep learning model, the identity information, behavior sequence, equipment wearing state and position information of the personnel are accurately analyzed, the automatic conversion from the original video to the structured compliance elements is realized, the instantaneous violation behaviors such as unlicensed work, equipment missing and operation disorder are effectively captured, and the problems of personnel responsibility confusion, hidden violation and timely discovery in the prior art are solved. The compliance rule set matched with the current work stage and scene risk factor is dynamically called, so that the safety supervision requirements are adjusted in real time according to the work progress and environmental threats (such as automatically strengthening the verification of anti-skid equipment when sudden heavy rain occurs), which breaks through the limitation of static rules that cannot dynamically respond to scene changes. BRIEF DESCRIPTION OF DRAWINGS

[0050] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which like reference characters refer to like parts throughout the figures, and wherein:

[0051] Figure 1 A method flow diagram of the blasting site behavior compliance identification method according to an embodiment of the present application is shown in FIG. 1.

[0052] Figure 2 A module structure diagram of the blasting site behavior compliance identification system according to an embodiment of the present application is shown in FIG. 2.

[0053] Figure 3 A structure diagram of the electronic device according to an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0055] Please refer to Figure 1 , Figure 1 A method flow diagram of the blasting site behavior compliance identification method is shown in FIG. 1.

[0056] The blasting site behavior compliance identification method comprises:

[0057] S101: Obtain multi-source video streams and environment monitoring data of a blasting site.

[0058] The environment monitoring data comprises temperature and humidity, wind speed, and dangerous gas concentration.

[0059] In the application, to establish a comprehensive perception of the safety state of the blasting site, first, multi-dimensional real-time data of the site are systematically obtained. Specifically, by deploying multiple visible light and infrared video monitoring devices in key areas of the site (such as temporary storage points of explosives, charging areas, connection areas, detonation stations, warning boundaries, and personnel passageways, etc.), dynamic video streams covering different angles and different operation links are continuously collected to form multi-source video streams. These video streams complement each other, aiming to eliminate the visual blind area of a single angle, and provide basic visual information for subsequent identification of personnel behavior, equipment state, operation stage, and some environmental interference factors.

[0060] At the same time, in order to accurately capture the environmental physical conditions that may affect the safety of the operation or increase the risk of violating regulations, environmental monitoring data needs to be collected simultaneously. This is achieved by deploying a corresponding sensor network at key locations in the operation site. The collected data includes: temperature and humidity data reflecting the comfort level and potential static risk at the site; wind speed data affecting the accumulation and diffusion of harmful gases and the trajectory prediction of blasting flying objects; and dangerous gas concentration data (such as methane, carbon monoxide, hydrogen sulfide, etc.) directly related to explosion risk and personnel health.

[0061] The synchronization of multi-source video streams and environmental monitoring data (temperature and humidity, wind speed, and dangerous gas concentration) is the data cornerstone of the entire compliance identification method. Video streams provide direct visual observation of personnel activities, object status, and operation processes, while environmental monitoring data quantifies the physical and chemical risk factors present at the site.

[0062] S102: Identify the current operation stage according to the multi-source video stream.

[0063] In the application, after obtaining the multi-source video stream of the blasting operation site, it is necessary to accurately identify the current operation stage of the site, which is a key prerequisite for dynamic compliance supervision. To achieve this goal, first, the collected multi-channel video stream is processed for time and space synchronization, through timestamp alignment and spatial coordinate calibration, eliminating the visual misalignment caused by differences in device deployment location or clock bias, generating a time and space unified multi-view video sequence, providing consistent visual input for subsequent feature analysis.

[0064] Based on the calibrated video sequence, the video content is deeply analyzed to extract key operation state features. Key device operation features (such as drill operation, explosive loading device activation, detonator preparation state, and specific device action patterns) and personnel collaboration features (such as safety officer demarcation, blaster loading and connecting, clearing personnel evacuation, and multi-person collaborative handling) are extracted. These features are the core visual identifiers of different operation stages.

[0065] The extracted device operation features and personnel collaboration features are input into the preset operation stage conversion rule library for time sequence correlation analysis. The rule library clearly defines the logical conversion conditions and typical feature combinations between each operation stage (such as preparation stage, loading stage, connecting stage, warning stage, detonation stage, and warning removal and inspection stage) based on standards such as the "Blasting Safety Regulations". The analysis process dynamically tracks the appearance order, duration, and combination of features on the time axis to determine the specific stage node of the current operation process.

[0066] Specifically, the above-mentioned identification of the current operation stage according to the multi-source video stream includes the following steps:

[0067] spatiotemporal synchronization processing on the multi-source video stream to generate a calibrated multi-view video sequence;

[0068] extracting device operation features and personnel collaboration features in the multi-view video sequence;

[0069] based on a preset job phase transition rule, performing temporal correlation analysis on the device operation features and the personnel collaboration features;

[0070] determining the current job phase according to the correlation analysis result.

[0071] The current job phase automatic identification realized through the above steps replaces the traditional inefficient mode relying on manual experience judgment, significantly improving the objectivity and real-time performance of phase judgment. On the one hand, it provides accurate basis for subsequent dynamic scheduling to match the safety rule set of this phase (such as strictly checking anti-static equipment during charging phase and ensuring clearing during detonation phase); on the other hand, by real-time monitoring of the operation progress, the system can predict possible risk points in the next phase (such as strengthening personnel clearing verification in advance when entering the detonation phase), so as to realize the advance of safety supervision and effectively avoid compliance risks caused by phase misjudgment or response lag.

[0072] S103: identifying scene risk factors according to the multi-source video stream and the environmental monitoring data.

[0073] In the application, in order to dynamically evaluate the overall safety situation of the blasting operation site, potential risk inducements need to be identified by comprehensively analyzing the video and environmental data. First, the real-time collected environmental monitoring data (including temperature, humidity, wind speed, and dangerous gas concentration) are automatically compared with the preset safety threshold. For example, when the methane concentration exceeds 20% of the lower explosive limit, the wind speed is higher than the safe operation limit, or the humidity is too low to cause static risk, a quantitative environmental risk level is generated according to the preset grading standard (such as low, medium, high, and extremely high), directly reflecting the threat degree of the physical environment to the operation safety.

[0074] At the same time, computer vision technology is used to deeply analyze the multi-source video stream and extract environmental interference features in the visual layer. Three key factors are identified: first, the smoke diffusion mode and visibility change are analyzed to calculate the smoke concentration, and the interference of the smoke concentration on the field of view of the operation personnel and the identification of the device is evaluated; second, the raindrop trajectory, fog density, and light attenuation degree are detected to quantify the rain and fog interference intensity; third, based on the ground reflection characteristics, water accumulation area, and personnel walking posture, the coverage range of the muddy area on the ground is identified and calculated. These features together constitute the environmental interference factor feature set.

[0075] The generated environmental risk level is fused with the environmental interference factor feature set in multiple dimensions through a pre-constructed risk mapping rule. The rule library defines the coupling effect of risk factors under different environmental conditions (e.g., high wind speed superimposed on high smoke concentration will significantly increase the risk of flying stones, and heavy rain may expand the muddy area, which may cause personnel to slip or equipment to malfunction). The fusion process uses weighted aggregation or a decision tree model to finally output a dynamic scene risk factor that not only quantifies the danger of the environment itself but also assesses the impact of the environment on the accuracy of the monitoring system.

[0076] Specifically, the identification of the scene risk factor according to the multi-source video stream and the environmental monitoring data includes the following steps:

[0077] Comparing the environmental monitoring data with the preset safety threshold generates an environmental risk level;

[0078] Analyzing the multi-source video stream extracts environmental interference factor features such as smoke concentration, rain and fog interference intensity, and ground muddy area coverage;

[0079] Based on the pre-constructed risk mapping rule, the environmental risk level and the environmental interference factor features are fused to generate a scene risk factor.

[0080] The above steps can break through the limitations of traditional static safety assessment. On the one hand, by fusing multi-source heterogeneous data (sensor data and video features), it can perceive complex risks that the human eye cannot detect (such as flammable gas leakage accompanied by low visibility); on the other hand, the dynamically generated scene risk factor provides a scientific basis for subsequent steps (such as adaptively adjusting the compliance rule strictness, triggering specific warning mechanisms). For example, when identifying "high rain and fog interference + ground muddy area exceeding limit", the verification intensity of personnel anti-slip equipment can be automatically enhanced, or the fault tolerance threshold of spatial positioning algorithm can be increased, thereby maintaining the effectiveness and robustness of supervision in harsh environments, significantly reducing the risk of missed judgment due to environmental mutations.

[0081] S104: Input the multi-source video stream into the pre-trained deep learning model to analyze the identity information, behavior sequence, equipment wearing state, and position information of each worker.

[0082] In the application, to realize the accurate extraction of individualized compliance elements of the workers, the multi-source video stream processed synchronously needs to be deeply semantically analyzed. This process is completed by inputting the video stream into a pre-trained deep learning model. The model is trained based on a large-scale blasting operation scene dataset and adopts a multi-task learning architecture, which can output four types of key information in parallel: identity information (by fusing face recognition, uniform color coding recognition, and ID card detection technology, to distinguish roles such as blaster, safety officer, and guard, and verify their qualification validity), behavior sequence (by a spatiotemporal action segmentation model, to continuously capture and label standard actions such as “handling explosives”, “loading drill holes”, “connecting detonators”, and “leaving the warning area” and their temporal relationships), equipment wearing state (based on a target detection and attribute classification model, to identify the wearing integrity and effectiveness of protective equipment such as safety helmet, anti-static clothing, and goggles, such as detecting whether the helmet strap is buckled tightly and whether the protective clothing is damaged), and position information (combining multi-view geometric reconstruction and real-time positioning algorithm, to accurately output the coordinates of the personnel in the three-dimensional operation space and associate them to the pre-divided functional areas such as “explosive loading area”, “outside the warning line”, and “explosion avoidance point”).

[0083] The model design is optimized for the special challenges of the blasting site. On the one hand, the spatial complementarity of multi-source video streams is utilized. When recognition is difficult due to obstruction (such as equipment obstruction) or environmental interference (such as smoke), the reliable features of other views are automatically switched or weighted to ensure the robustness of the analysis results. On the other hand, the model uses a transfer learning mechanism to fine-tune the pre-trained general features for specific blasting scenes (such as low light in underground and strong reflection in open air), significantly improving the recognition accuracy of identifying elements (such as the reflective vest identification and warning zone number required by the “General Rules for Warning and Registration Identification of Civil Explosives”), and overcoming the high false detection rate of traditional methods in complex environments.

[0084] The deep learning model analyzes the personnel identity, behavior, equipment, and location in all dimensions, providing data support for subsequent rule checking and completely replacing the inefficient mode of manual frame-by-frame review. Secondly, the analysis process is real-time and high-precision, which can instantly capture instantaneous violations such as “unlicensed personnel entering the explosive loading area” and “not wearing anti-static gloves during the connection phase”. Finally, the multi-view fusion mechanism effectively solves the single-camera blind area problem, ensuring full coverage of supervision in hidden corners or densely populated areas, and avoiding compliance risks caused by visual omission from the root.

[0085] As to the deep learning model, it adopts an integrated architecture based on multi-task learning. The core of the architecture is a shared visual feature extraction backbone network; for example, the backbone network can adopt an I3D structure pre-trained on large-scale video datasets such as Kinetics, which internally contains multiple three-dimensional convolutional layers, three-dimensional pooling layers and nonlinear activation layers, and can effectively extract space-time features in the video. In order to process multi-source video streams, the system is equipped with a structure identical to the above-mentioned backbone network for each video source for independent feature extraction, and then the multi-view features are integrated through a feature fusion layer; the fusion layer can include channel splicing operations and one or more subsequent three-dimensional convolutional layers for feature dimension reduction and enhancement, and finally generate a unified multi-view joint feature representation.

[0086] The joint feature representation integrates visual information from different perspectives, providing a common and rich data basis for subsequent multiple recognition tasks. On this basis, the model parallelly constructs four specific task analysis branches: identity analysis branch, behavior analysis branch, equipment detection branch and position analysis branch.

[0087] Specifically, the identity analysis branch can include a region proposal network (RPN), an ROI alignment layer and two fully connected layers for performing person detection, bounding box regression and identity classification, and can fuse the output of a dedicated face recognition subnetwork. The behavior analysis branch includes a time series modeling module (such as a long short-term memory network layer or a space-time graph convolution layer), followed by a fully connected classification layer, for time series segmentation and identification of continuous behaviors. The equipment detection branch adopts an anchor-based object detection structure (such as Faster R-CNN or YOLO variants), including several two-dimensional convolutional layers, up-sampling layers and prediction layers, for locating and classifying various safety protection equipment and determining their wearing state. The position analysis branch is constructed as an encoder-decoder structure semantic segmentation network (such as U-Net); the encoder part extracts features through multiple convolution and pooling layers, and the decoder part restores the spatial resolution through deconvolution layers or up-sampling operations, and finally outputs the spatial region label of the person through a pixel-by-pixel classification layer.

[0088] The training of the deep learning model adopts a phased strategy, first pre-training based on large-scale general datasets such as Kinetics and COCO, initializing network parameters to build basic feature extraction capability; then fine-tuning using exclusive data for blasting operations. The exclusive data set required for fine-tuning comes from multi-view videos of real blasting sites, covering different light, weather interference and operation stages, and is annotated by professional personnel with four types of true value data of identity, behavior, equipment and position.

[0089] The training uses a multi-task joint loss function, where identity recognition uses cross-entropy loss, behavior sequence uses time focus loss to focus on key actions, equipment detection combines bounding box regression and classification loss, and position resolution uses Euclidean distance loss and region classification loss. The optimization process selects the adaptive moment estimation optimizer, and is supplemented by a learning rate dynamic adjustment strategy. Data augmentation techniques such as random occlusion simulation and illumination disturbance are introduced during training, effectively improving the model's generalization ability to complex disturbances in the field.

[0090] In addition, to meet the deployment requirements, the model can be compressed through knowledge distillation technology to generate a lightweight version to adapt to edge computing nodes and ensure real-time processing performance. In the model verification stage, an independent test set is used for evaluation, and the key indicators are required to reach the set threshold (such as identity recognition accuracy > 95%, behavior sequence F1 score > 0.92, equipment detection mAP > 0.89, and positioning error < 0.3 meters), and based on the misjudgment samples in the real scene, continuous incremental learning is carried out, so as to ensure the adaptability and accuracy of the model in different blasting environments, and provide reliable data support for compliance verification.

[0091] S105: Call the compliance rule set matched with the current job stage and scene risk factor in the pre-built safety rule library.

[0092] In the application, in order to verify the compliance of the blasting operation site, the safety rules that adapt to the current situation need to be dynamically scheduled. Based on the pre-built safety rule library, which is formed by structured analysis of standards such as "Blasting Safety Regulations" and "General Rules for Warning and Registration Marks of Civil Explosive Articles", contains hundreds of atomized rules covering the whole process, and is stored in multiple levels according to the job stage (preparation, charging, wiring, alert, detonation, and removal) and risk dimension (environment, behavior, equipment, and space).

[0093] When the current job stage and scene risk factor are obtained, the rule library is quickly searched through a double matching mechanism; specifically, first, according to the job stage, the baseline rule set that must be enforced in this stage is filtered (for example, the charging stage must include rules such as "anti-static equipment verification" and "single-person operation time limit"), and then according to the scene risk factor, incremental rules are dynamically loaded (for example, when the "high wind speed" risk factor is identified, rules such as "heavy object reinforcement check" and "flystone protection zone personnel prohibited entry" are added), and finally the real-time effective compliance rule set is generated.

[0094] The matching process can adopt a hybrid retrieval strategy based on hash table and decision tree, the job stage is used as the primary key to quickly locate the rule subset, and the scene risk factor is used to activate the associated rules through the weight matrix (for example, the "methane concentration exceeds the standard" factor will activate the "interrupt the job" and "evacuation instruction" high-risk rules, and the "rain and fog interference medium level" only triggers the "enhance positioning verification" and "lower behavior recognition confidence threshold" adaptive rules). In order to ensure the accuracy of rule scheduling, each rule is pre-installed with an effective condition expression (such as "wind speed > 8 m / s, stage = alert"), which is parsed and executed by the rule engine in real time.

[0095] The present application can dynamically load rules driven by the current job stage and scene risk factors, which can adapt to the actual job progress and environmental threats (such as automatically increasing the anti-skid equipment verification intensity when a rainstorm suddenly occurs in the connection stage), solving the defect that the static rules in the prior art cannot dynamically adjust the protection requirements; and the fine tailoring of the rule set avoids redundant verification, significantly improving the processing efficiency (such as retaining only core rules in low-risk stages). More importantly, this mechanism provides flexible adaptation capability for compliance requirements in different risk scenarios - for example, when the scene risk factor identifies "high gas concentration + low visibility", the compliance rule set will simultaneously strengthen the priority of "personnel positioning accuracy verification" and "automatic shutdown instruction", building an adaptive safety protection network from the rule level, effectively preventing the occurrence of complex risk events.

[0096] S106: Based on the compliance rule set, the identity information, behavior sequence, equipment wearing state and position information are checked for compliance.

[0097] Among them, the compliance rule set includes: qualification rules, behavior rules, equipment rules and space rules.

[0098] In the application, based on the dynamically generated compliance rule set, the structured information obtained by the above S104 is checked. The checking process is divided into four categories according to the rule type, and the specific implementation is as follows.

[0099] Qualification rule checking: compare the identity information (such as personnel ID, role category, certificate number) with the pre-set permission list and permission matrix in the rule set. If it is detected that the qualification certificate is expired, the job personnel role conflicts with the current stage forbidden permission (such as non-safety personnel appearing in the detonation area in the alert stage), or it is detected that there is no certificate personnel (such as illegal face or work clothes code), it is determined that the qualification is violated.

[0100] Behavior rule verification: match the behavior sequence (e.g., "carry explosives → load drill hole → connect detonator → evacuate") with the standard operation procedure (SOP) timing template of the current stage in the rule set. Detect missing key actions (e.g., "static discharge operation" not detected after loading), operation disorder (e.g., "trigger detonator detection signal before completing the wiring"), or high-risk redundant actions (e.g., "return to loading area during detonation phase") in real time and mark behavior deviations through the dynamic time warping algorithm (DTW).

[0101] Equipment rule verification: verify whether the equipment wearing state meets the requirements of the stage based on the target detection results. For example, in the loading phase, the integrity of the anti-static clothing (including no damage to the clothes and connection of the grounding belt) and the wearing of safety glasses are strictly verified. In the detonation phase, the disabled state of wireless communication equipment is focused on. The rule engine analyzes the equipment attributes and rule conditions (e.g., "protective clothing type = A level, current phase = loading") in real time, and triggers a violation determination when it identifies missing, failed, or type-mismatched equipment (e.g., helmet strap not tightened or ordinary work clothes instead of anti-static clothing).

[0102] Space rule verification: compare the location information (three-dimensional coordinates and associated area labels) with the electronic fence database. The verification content includes prohibited area intrusion (e.g., unauthorized personnel entering the explosive temporary storage area), personnel gathering exceeding the limit (e.g., the number of people in the warning area > the preset capacity), and clearing the scene and staying (e.g., there are still personnel heat signals outside the blast shelter after the detonation countdown ends). Space verification combines real-time positioning and historical trajectory prediction to ensure the accuracy of capturing transient violations (e.g., quickly entering the restricted area).

[0103] The verification process adopts a multi-level confidence mechanism: for ambiguous data caused by environmental interference (e.g., behavior recognition confidence of 60% in rain and fog), automatically associate scene risk factors to dynamically adjust the determination threshold (e.g., lower the confidence threshold to 50% in high-risk scenarios). At the same time, the rule engine supports composite condition triggering (e.g., "identity = blaster, location = wiring area, and anti-static gloves not detected"), enabling multi-dimensional violation correlation analysis.

[0104] Specifically, the above compliance rule set verifies the identity information, behavior sequence, equipment wearing state, and location information for compliance, including the following steps:

[0105] Verify the identity information according to the qualification rules. When the work personnel's qualification is invalid or the post permission exceeds the boundary, it is determined that there is a violation behavior;

[0106] Compare the behavior sequence according to the behavior rules. When key actions are missing, operations are disordered, or high-risk redundant actions are detected, it is determined that there is a violation behavior;

[0107] Verify the equipment wearing state according to the equipment rules. When protective equipment is missing, failed, or does not match the requirements of the current operation phase, it is determined that there is a violation behavior;

[0108] Based on spatial rules and location information comparison, when workers are in prohibited areas, exceed gathering limits, or fail to evacuate as required, it is determined that there is a violation.

[0109] S107: If a violation is found during the verification, a violation judgment result containing the violation type, risk level, associated personnel, and spatial area identifier will be generated.

[0110] In the application, when a violation is detected during the compliance verification process, a structured violation judgment result will be generated. This result first precisely identifies the violation type, and its classification strictly corresponds to the four-dimensional architecture of the rule base: qualification violation (such as unlicensed operation or overreach of authority), behavioral violation (such as missing key actions or high-risk redundant operations), equipment violation (such as malfunctioning or incompatible protective equipment), and spatial violation (such as trespassing into restricted areas or exceeding gathering limits). Each type is associated with a preset risk level assessment model—this model is trained based on the nature of the violation (such as whether it directly triggered an explosion), scenario risk factors (such as not wearing anti-static clothing in a high-concentration flammable gas environment), and historical accident data, and outputs four levels of quantitative risk labels: low, medium, high, and emergency; for example, "personnel were not evacuated during the detonation phase" is automatically labeled as "emergency".

[0111] The judgment result is simultaneously linked to the information of related personnel. By extracting the identity identifiers (such as facial ID, employee number) and role attributes (blaster / safety officer) parsed in S104, the responsible person is accurately identified, and their behavioral sequence fragments are linked as corroborating evidence. Spatial area identification is based on the electronic fence system and real-time positioning data, marking the specific functional area where the violation occurred (such as "northeast corner of the explosive loading area" or "5 meters outside the warning line"), supplemented by three-dimensional coordinates and regional risk attributes (such as "high-risk gas accumulation area"). Finally, the four elements (violation type, risk level, related personnel, and spatial area identification) are encapsulated into standardized data objects for use in early warning and evidence chain construction.

[0112] This invention enables precise output and rapid response in risk decision-making. First, structured judgment results replace traditional vague textual descriptions, allowing the command center to instantly understand the nature of violations (e.g., "Spatial violation - Emergency level: Zhang San is lingering in the detonation zone"), significantly improving emergency decision-making efficiency. Second, automated risk level assessment (e.g., combining "high wind speed" scenario factors to upgrade "personnel gathering in the fly rock protection zone" to high risk) solves the problem of strong subjectivity in manual judgment, ensuring that resources are prioritized for truly high-risk events. Finally, binding personnel and spatial information provides core elements for subsequent electronic evidence chains, fundamentally avoiding disputes during liability tracing and significantly strengthening the authority and enforceability of safety supervision.

[0113] In one embodiment, before inputting the multi-source video streams into the pre-trained deep learning model, the method further includes:

[0114] acquire point cloud data and on-site audio stream of the blasting operation site;

[0115] determine the environmental interference intensity according to the environmental monitoring data and the multi-source video stream;

[0116] when the environmental interference intensity exceeds the preset threshold, input the multi-source video stream into the pre-trained deep learning model, specifically, input the multi-source video stream, the point cloud data and the on-site audio stream into the pre-trained deep learning model;

[0117] wherein, the point cloud data is used to supplement the spatial positioning information, and the on-site audio stream is used to extract acoustic event features associated with standard operation actions.

[0118] In the application, before the multi-source video stream is input into the pre-trained deep learning model for analysis, a multi-modal data enhancement acquisition and adaptive triggering process can also be performed. Specifically, first, source sensing devices are added in key areas of the blasting operation site, and point cloud data generated by a millimeter wave radar and on-site audio stream captured by a high-sensitivity acoustic sensor array are synchronously acquired. The point cloud data provides spatial depth information that is not constrained by visible light conditions and can penetrate certain degree of smoke and water mist interference; the on-site audio stream records various sound events in the operation environment, including human voice, mechanical operation sound, and blasting equipment assembly sound.

[0119] The environmental monitoring data and the multi-source video stream are continuously analyzed jointly to comprehensively judge the current environmental interference intensity. Specifically, by comparing multi-dimensional indexes such as real-time wind speed, rainfall, smoke visual concentration, and visibility estimation value with corresponding preset values, and introducing a fusion evaluation model based on machine learning, a quantitative environmental interference intensity score is output. The score reflects the degree of perception challenge currently faced by the visible light visual sensor.

[0120] When it is determined according to the score that the environmental interference intensity exceeds the preset threshold, i.e. the visual perception condition is significantly deteriorated to possibly affect the recognition accuracy, the multi-modal fusion analysis mechanism is automatically triggered. Under this mechanism, the step of inputting the multi-source video stream into the pre-trained deep learning model is specifically expanded to synchronously input the multi-source video stream, the point cloud data and the on-site audio stream into the deep learning model. The point cloud data is mainly used to supplement and provide more robust three-dimensional spatial positioning and motion trajectory information of personnel and large equipment in the case of visual obstruction or low visibility; the on-site audio stream is processed through the acoustic event recognition branch of the model, aiming to extract acoustic event features closely related to the standard operation process of blasting, such as the knocking sound of a specific tool, the repetition of a security password, and the specific friction sound during explosive loading, etc. These acoustic features can be used as an effective supplement and cross verification of visual behavior sequences.

[0121] By introducing the above-mentioned multi-modal data adaptive fusion mechanism, the accuracy of behavior analysis in harsh environments is significantly improved, ensuring the continuity and reliability of the compliance monitoring process.

[0122] Further, the deep learning model analyzes the identity information, behavior sequence, equipment wearing state, and position information of each worker according to multi-source video streams, point cloud data, and live audio streams. Its process deeply integrates multi-source heterogeneous data to improve the robustness and accuracy of the analysis. The model first processes the point cloud data, extracts and tracks the dynamic contours of each worker through point cloud clustering and three-dimensional target tracking algorithms, and then generates their accurate three-dimensional spatial coordinates and motion trajectory information. This effectively compensates for the possible failure of pure visual positioning in cases of visual obstruction or extremely low visibility.

[0123] Synchronously, the model analyzes the live audio stream in real time, uses acoustic event detection and sound source positioning technology to separate and identify specific acoustic events that are strongly related to standard blasting operation actions from complex background noise, such as the crisp tapping sound when connecting detonators, the roaring sound of a specific type of drilling machine, or the standard safety password sound between personnel, and associates these acoustic events with their approximate directions.

[0124] After that, the cross-modal feature alignment module built into the model begins to work. This module is responsible for aligning and mapping the spatial positioning information from the point cloud data, the acoustic event features from the audio stream, and the multi-view joint feature representation generated after feature fusion from the multi-source video stream (which contains visual information such as personnel appearance, action posture, equipment shape, etc.) on a unified spatio-temporal reference. On this basis, the module uses a weighted fusion strategy based on an attention mechanism to adaptively fuse the features of the above three modalities, and finally generates a unified and enhanced multi-modal feature representation.

[0125] It should be noted that the weight coefficients in this weighted fusion process are not fixed but are dynamically adjusted by the model according to the real-time evaluation of the environmental interference intensity. For example, when the environmental interference intensity is extremely high (such as thick smoke), causing the quality of visual features to decline significantly, the model will automatically increase the fusion weight of point cloud data and audio stream features, relying on the spatial perception ability of radar and the timing characteristics of acoustic events to maintain the extraction ability of key information; conversely, when the visual conditions are good, more emphasis is placed on rich visual features. Finally, based on this dynamically fused multi-modal feature representation, the model outputs the identity information, continuous behavior sequence, complete equipment wearing state, and accurate position information of each worker through its parallel multiple task analysis branches with stable and high precision.

[0126] In one embodiment, the above method further comprises:

[0127] When generating the violation determination result, real-time video clips corresponding to the violation type, associated personnel, and spatial region identifier in the multi-source video stream are extracted;

[0128] A timestamp and a spatial position marker are added to the real-time video clip, and are associated and bound with the violation determination result, environmental monitoring data, and the current work stage to generate an electronic evidence data package;

[0129] A digital signature and a hash value are added to the electronic evidence data package to generate an electronic evidence record that cannot be tampered with;

[0130] The electronic evidence record is stored in a tamper-proof storage system to form a traceable electronic evidence chain.

[0131] In the application, when the violation determination result is generated, the construction process of the electronic evidence chain is automatically started. Specifically, first, the original real-time video clip accurately corresponding to the violation type, the identity of the associated personnel, and the spatial region identifier where the violation occurred in the multi-source video stream is retrieved and extracted in reverse. This extraction process ensures that the video evidence is completely synchronized in time with the moment when the violation occurs and covers the specific area perspective of the violation, thereby truly and completely restoring the scene of the violation.

[0132] Subsequently, the extracted original real-time video clip is subjected to enhanced marking processing. High-precision timestamp information is superimposed on it; at the same time, spatial position markers based on two-dimensional or three-dimensional electronic maps of the scene are superimposed to clearly mark the specific geographic location or functional area where the violation occurred. After completing the marking, the video clip is structurally associated and bound with the currently generated violation determination result, environmental monitoring data snapshot at the time of the violation, and identified current work stage information, and is encapsulated to form an electronic evidence data package that is complete in content and associated in elements. The internal logic of the data package is unified, and all other associated information can be retrieved and traced through any element.

[0133] To ensure the authenticity and integrity of the evidence and prevent it from being tampered with during transmission, storage, or review, cryptography technology can be used to reinforce the electronic evidence data package. A digital signature of the data package is generated using an asymmetric encryption algorithm, and a unique hash value of the data package content is generated through a hash operation. The digital signature is used to verify the legality of the evidence source and the integrity of the content, and the hash value serves as the "digital fingerprint" of the data package, and any minor changes will cause the hash value to be invalid. Thus, an electronic evidence record that cannot be tampered with is generated.

[0134] Finally, the generated electronic evidence record is transmitted and stored in a dedicated tamper-proof storage system in real time. The storage system can use write-once-read-many (WORM) technology or distributed storage technology based on blockchain to prevent the record from being modified or deleted afterwards from the physical medium or system architecture. All records are stored in chronological order and indexed by event type, forming a complete, clear and traceable electronic evidence chain.

[0135] Corresponding to the foregoing application function implementation method embodiments, the present application also provides a blasting operation site behavior compliance identification system and corresponding embodiments.

[0136] Please refer to Figure 2 , Figure 2 for a schematic diagram of the module structure of the blasting operation site behavior compliance identification system.

[0137] The blasting operation site behavior compliance identification system comprises:

[0138] An acquisition unit 21 is configured to acquire multi-source video streams and environmental monitoring data of a blasting operation site; the environmental monitoring data comprises temperature and humidity, wind speed, and dangerous gas concentration;

[0139] An operation stage identification unit 22 is configured to identify a current operation stage according to the multi-source video streams;

[0140] A scene risk analysis unit 23 is configured to identify scene risk factors according to the multi-source video streams and the environmental monitoring data;

[0141] A personnel behavior analysis unit 24 is configured to input the multi-source video streams into a pre-trained deep learning model to analyze identity information, behavior sequences, equipment wearing states, and position information of each operation personnel;

[0142] A rule dynamic scheduling unit 25 is configured to call a compliance rule set in a pre-constructed safety rule library that matches the current operation stage and the scene risk factors;

[0143] A compliance verification unit 26 is configured to perform compliance verification on the identity information, the behavior sequences, the equipment wearing states, and the position information based on the compliance rule set;

[0144] A violation decision generation unit 27 is configured to generate a violation judgment result comprising a violation type, a risk level, associated personnel, and a spatial region identifier if there is a violation behavior in the verification.

[0145] In one embodiment, in terms of identifying the current operation stage according to the multi-source video streams, the operation stage identification unit 22 is specifically configured to:

[0146] perform spatio-temporal synchronization processing on the multi-source video streams to generate a calibrated multi-view video sequence;

[0147] extracting device operation features and personnel cooperation features in the multi-view video sequence;

[0148] performing time sequence correlation analysis on the device operation features and the personnel cooperation features based on preset operation stage conversion rules;

[0149] determining a current operation stage according to a result of the correlation analysis.

[0150] In one embodiment, in terms of identifying scene risk factors based on multi-source video streams and environmental monitoring data, the scene risk analysis unit 23 is specifically configured to:

[0151] comparing the environmental monitoring data with a preset safety threshold to generate an environmental risk level;

[0152] analyzing the multi-source video streams to extract environmental interference factor features of smoke density, rain and fog interference intensity, and ground muddy area coverage range;

[0153] generating the scene risk factors by fusing the environmental risk level and the environmental interference factor features based on a pre-constructed risk mapping rule.

[0154] In one embodiment, in terms of performing compliance verification on identity information, behavior sequence, equipment wearing state, and location information based on a compliance rule set, the compliance verification unit 26 is specifically configured to:

[0155] verifying the identity information according to a qualification rule, and determining that there is a violation when the qualification of the operation personnel is invalid or the post permission is out of bounds;

[0156] comparing the behavior sequence according to a behavior rule, and determining that there is a violation when a key action is missing, an operation is out of order, or a high-risk redundant action is identified;

[0157] verifying the equipment wearing state according to an equipment rule, and determining that there is a violation when protective equipment is missing, invalid, or does not match the requirements of the current operation stage;

[0158] comparing the location information according to a space rule, and determining that there is a violation when the operation personnel is in a prohibited area, exceeds the gathering limit, or does not evacuate according to the clearing requirements.

[0159] In one embodiment, the acquisition unit 21 is further configured to:

[0160] acquire point cloud data and on-site audio streams of a blasting operation site;

[0161] The system further includes:

[0162] an environmental interference determination unit configured to determine environmental interference intensity based on environmental monitoring data and multi-source video streams;

[0163] In the aspect of inputting the multi-source video stream into the pre-trained deep learning model, the personnel behavior analysis unit 24 is specifically used for:

[0164] When the environmental interference intensity exceeds the preset threshold, inputting the multi-source video stream, the point cloud data and the live audio stream into the pre-trained deep learning model;

[0165] The point cloud data is used for supplementing the spatial positioning information, and the live audio stream is used for extracting acoustic event features associated with the standard operation action.

[0166] In one embodiment, the system further comprises:

[0167] The evidence chain generation unit is used for:

[0168] When the violation determination result is generated, extracting real-time video clips in the multi-source video stream corresponding to the violation type, the associated personnel and the spatial region identifier;

[0169] Adding a timestamp and a spatial position mark to the real-time video clips, and binding the real-time video clips with the violation determination result, the environmental monitoring data and the current operation stage to generate an electronic evidence data package;

[0170] Adding a digital signature and a hash value to the electronic evidence data package to generate an electronic evidence record that is not tamperable;

[0171] Storing the electronic evidence record into a tamper-proof storage system to form a traceable electronic evidence chain.

[0172] As to the system in the above embodiment, the specific manner in which the various unit modules perform operations has been described in detail in the embodiment relating to the method, and will not be described in detail here.

[0173] In the embodiment of the application, an electronic device is also provided, which comprises a processor and a memory, and the memory stores a computer program.

[0174] Please refer to Figure 3 , the electronic device 3000 comprises a memory 3010 and a processor 3020.

[0175] The processor 3020 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can be any conventional processor.

[0176] The memory 3010 can include various types of storage units, such as a system memory, a read-only memory (ROM), and a permanent storage device. Among them, the ROM can store static data or instructions required by the processor 3020 or other modules of the computer. The permanent storage device can be a read-write storage device. The permanent storage device can be a non-volatile storage device that does not lose stored instructions and data even after the computer is powered off. In some embodiments, the permanent storage device uses a mass storage device (such as a magnetic or optical disk, a flash memory) as a permanent storage device. In some other embodiments, the permanent storage device can be a removable storage device (such as a floppy disk, an optical drive). The system memory can be a read-write storage device or a volatile read-write storage device, such as a dynamic random access memory. The system memory can store some or all of the instructions and data required by the processor during runtime. In addition, the memory 3010 can include a combination of any computer-readable storage media, including various types of semiconductor memory chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic disks and / or optical disks. In some embodiments, the memory 3010 can include a read and / or write removable storage device, such as a compact disc (CD), a read-only digital versatile disc (such as DVD-ROM, double-layer DVD-ROM), a read-only Blu-ray disc, an ultra-density optical disc, a flash memory card (such as an SD card, a minSD card, a Micro-SD card, etc.), a magnetic floppy disk, etc. The computer-readable storage medium does not include a carrier wave and a transient electronic signal transmitted through a wireless or wired transmission.

[0177] The memory 3010 stores executable code, which, when processed by the processor 3020, can cause the processor 3020 to perform part or all of the above-mentioned methods.

[0178] Furthermore, the method according to the present application can also be implemented as a computer program or a computer program product, which comprises computer program code instructions for performing some or all of the steps of the above-mentioned method of the present application.

[0179] Alternatively, the present application can also be implemented as a computer readable storage medium (or a non-transitory machine readable storage medium or a machine readable storage medium) having stored thereon executable codes (or computer programs or computer instruction codes) which, when executed by a processor of an electronic device (or a server, etc.), cause the processor to perform some or all of the steps of the above-mentioned method according to the present application.

[0180] The above has described embodiments of the present application, the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The selection of the terms used herein is intended to best explain the principles of the embodiments, practical application or improvement of the technology in the market, or to enable other ordinary skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for identifying compliance in blasting operations, characterized in that, include: Acquire multi-source video streams and environmental monitoring data from the blasting operation site; The environmental monitoring data includes temperature, humidity, wind speed, and concentration of hazardous gases; The current operation stage is identified based on the multi-source video stream; Based on the multi-source video streams and environmental monitoring data, identify scene risk factors; The multi-source video stream is input into a pre-trained deep learning model to parse out the identity information, behavior sequence, equipment wearing status and location information of each operator; during the parsing process, when there is occlusion or environmental interference in a certain perspective of the multi-source video stream, the deep learning model switches or weightedly fuses the features of the other perspectives. Call the set of compliance rules in the pre-built security rule base that matches the current operation stage and scenario risk factors; The compliance rule set includes: qualification rules, behavior rules, equipment rules, and space rules; Based on the aforementioned compliance rule set, compliance verification is performed on the identity information, behavior sequence, equipment wearing status, and location information. If a violation is found during the verification, a violation determination result is generated, which includes the violation type, risk level, associated personnel, and spatial area identifier. Identifying the current operation stage based on the multi-source video streams includes: The multi-source video streams are spatiotemporally synchronized to generate a calibrated multi-view video sequence; Extract equipment operation features and personnel collaboration features from the multi-view video sequence; Based on preset work phase transition rules, a time-series correlation analysis is performed on the equipment operation characteristics and personnel collaboration characteristics; Based on the correlation analysis results, the current operation stage is determined; Before inputting the multi-source video stream into a pre-trained deep learning model, the method further includes: Acquire point cloud data and on-site audio stream from the blasting operation site; The intensity of environmental interference is determined based on the environmental monitoring data and the multi-source video stream; When the intensity of the environmental interference exceeds a preset threshold, the multi-source video stream is input into a pre-trained deep learning model. Specifically, the multi-source video stream, point cloud data, and on-site audio stream are input into the pre-trained deep learning model. The point cloud data is used to supplement spatial positioning information, and the on-site audio stream is used to extract acoustic event features associated with standard operating procedures. The cross-modal feature alignment module built into the deep learning model aligns and maps the spatial positioning information from the point cloud data, the acoustic event features from the live audio stream, and the multi-view joint feature representation generated by feature fusion of the multi-source video streams onto a unified spatiotemporal reference. Then, an attention-based weighted fusion strategy is used for adaptive fusion to generate multimodal feature representations. The weight coefficients in the weighted fusion process are dynamically adjusted by the deep learning model based on the real-time evaluation of the intensity of environmental interference.

2. The method for identifying compliance of blasting operation site behavior according to claim 1, characterized in that, Based on the multi-source video streams and environmental monitoring data, scene risk factors are identified, including: The environmental monitoring data is compared with a preset safety threshold to generate an environmental risk level. Analyze the multi-source video streams to extract environmental interference factors such as smoke and dust concentration, rain and fog interference intensity, and the coverage area of ​​muddy areas on the ground; Based on pre-constructed risk mapping rules, the environmental risk level and environmental disturbance factor characteristics are integrated to generate scenario risk factors.

3. The method for identifying compliance of blasting operation site behavior according to claim 1, characterized in that, Based on the aforementioned compliance rule set, compliance verification is performed on the identity information, behavior sequence, equipment wearing status, and location information, including: The identity information is verified according to the qualification rules. When the operator's qualification is invalid or the job authority is exceeded, it is determined that there is a violation. According to the behavior rules, the behavior sequence is compared. When a key action is missing, the operation is out of order, or a high-risk redundant action is identified, it is determined that there is a violation. The equipment wearing status is verified according to the equipment rules. When the protective equipment is missing, malfunctioning, or does not match the requirements of the current operation stage, it is determined that there is a violation. Based on the spatial rules and the location information, if the workers are in a prohibited area, exceed the gathering limit, or fail to evacuate as required, it is determined that there is a violation.

4. The method for identifying compliance of blasting operation site behavior according to claim 1, characterized in that, The method further includes: When the violation determination result is generated, real-time video segments corresponding to the violation type, associated personnel, and spatial area identifier are extracted from the multi-source video stream; Add timestamps and spatial location markers to the real-time video clips, and associate and bind them with the violation judgment results, environmental monitoring data and the current operation stage to generate an electronic evidence data package; Add a digital signature and hash value to the electronic evidence data packet to generate an immutable electronic evidence record; The electronic evidence records are stored in an anti-tampering storage system to form a traceable chain of electronic evidence.

5. A compliance identification system for blasting operation site behavior, characterized in that, include: The acquisition unit is used to acquire multi-source video streams and environmental monitoring data from the blasting operation site. The environmental monitoring data includes temperature, humidity, wind speed, and concentration of hazardous gases; A job phase identification unit is used to identify the current job phase based on the multi-source video stream; The scene risk analysis unit is used to identify scene risk factors based on the multi-source video streams and environmental monitoring data; The personnel behavior analysis unit is used to input the multi-source video stream into a pre-trained deep learning model to analyze the identity information, behavior sequence, equipment wearing status and location information of each operator; during the analysis process, when there is occlusion or environmental interference in a certain perspective of the multi-source video stream, the deep learning model switches or weightedly fuses the features of the other perspectives. The rule dynamic scheduling unit is used to call the compliance rule set in the pre-built security rule library that matches the current operation stage and scenario risk factors; the compliance rule set includes: qualification rules, behavior rules, equipment rules and space rules; The compliance verification unit is used to perform compliance verification on the identity information, behavior sequence, equipment wearing status and location information based on the compliance rule set. The violation decision generation unit is used to generate a violation judgment result containing the violation type, risk level, related personnel, and spatial area identifier if a violation is found during verification. In identifying the current job stage based on the multi-source video stream, the job stage identification unit is specifically used for: The multi-source video streams are spatiotemporally synchronized to generate a calibrated multi-view video sequence; Extract equipment operation features and personnel collaboration features from the multi-view video sequence; Based on preset work phase transition rules, a time-series correlation analysis is performed on the equipment operation characteristics and personnel collaboration characteristics; Based on the correlation analysis results, the current operation stage is determined; The acquisition unit is also used to acquire point cloud data and on-site audio streams from the blasting operation site; An environmental interference determination unit is used to determine the intensity of environmental interference based on the environmental monitoring data and the multi-source video stream. In inputting the multi-source video streams into a pre-trained deep learning model, the personnel behavior parsing unit is specifically used for: When the intensity of the environmental interference exceeds a preset threshold, the multi-source video stream is input into a pre-trained deep learning model. Specifically, the multi-source video stream, point cloud data, and on-site audio stream are input into the pre-trained deep learning model. The point cloud data is used to supplement spatial positioning information, and the on-site audio stream is used to extract acoustic event features associated with standard operating procedures. The cross-modal feature alignment module built into the deep learning model aligns and maps the spatial positioning information from the point cloud data, the acoustic event features from the live audio stream, and the multi-view joint feature representation generated by feature fusion of the multi-source video streams onto a unified spatiotemporal reference. Then, an attention-based weighted fusion strategy is used for adaptive fusion to generate multimodal feature representations. The weight coefficients in the weighted fusion process are dynamically adjusted by the deep learning model based on the real-time evaluation of the intensity of environmental interference.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the identification method as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When a computer program is executed by a processor, it implements the identification method as described in any one of claims 1-4.

Images