Ship production workshop safety monitoring device and method and computer equipment

Abstract

The invention relates to a ship production workshop safety monitoring device and method and computer equipment. According to the application, the video stream, the sound data and the environmental parameters are acquired through the multi-mode monitoring terminal, and the procedure exclusive behavior recognition module performs special illegal behavior recognition, so that precise monitoring of procedures such as welding, coating, hoisting and the like is realized, and the recognition accuracy of the special illegal behaviors is improved. When a suspected illegal behavior is detected, the system automatically triggers the authority face recognition camera to carry out secondary verification of the identity and the operation authority of the person, so that closed-loop control of behavior recognition-identity verification-authority verification is realized, and the safety risk caused by operation of an unauthorized person is effectively avoided. The industrial-grade super computer carries out centralized processing on identification results and sends violation information to the grading early warning module, and the grading early warning module carries out three-stage grading early warning and corresponding processing measures according to violation risk grades, so that automation and timeliness of a response mechanism are realized, and safety management is promoted.

Description

Technical Field

[0001] This application relates to the field of monitoring and identification technology, and in particular to a safety monitoring device, method and computer equipment for a shipbuilding production workshop. Background Technology

[0002] In recent years, the shipbuilding industry has flourished. Ship production workshops, as core work areas, encompass diverse and complex processes such as welding, painting, hoisting, and assembly of core hull components. These workshops are characterized by spacious and clearly defined areas, dense equipment and complex working conditions, frequent personnel movement, and significant differences in operational authority.

[0003] Currently, safety management in shipbuilding workshops primarily relies on traditional methods, namely manual inspections combined with general-purpose monitoring equipment. Manual inspections require significant manpower and are limited by time intervals and personnel availability, making it difficult to achieve full-time, all-around coverage of the workshop. While general-purpose monitoring equipment can collect video to some extent, it has several limitations: Firstly, it lacks the ability to specifically identify risks associated with different processes, such as the risk of welding sparks igniting fires in the welding area, the risk of VOC concentration exceeding standards due to the lack of respirators in the painting area, and the risk of improper hand signals and personnel entering dangerous areas in the hoisting area. It often only identifies basic violations such as smoking and using mobile phones, failing to meet the needs of precise control at each process level. Secondly, due to the complex environment of shipbuilding workshops, such as strong light in the welding area, fog in the painting area, and equipment obstructions, general-purpose monitoring suffers from low recognition rates and high false alarm rates, severely impacting control effectiveness.

[0004] Meanwhile, regarding personnel and access control, existing monitoring can only perform basic identity verification for personnel entering the premises, and cannot verify whether personnel have the operational authority for the current confidential process (such as the assembly of core ship components). This can easily lead to unauthorized personnel entering the premises, posing a security risk. Moreover, there is no effective linkage and traceability mechanism between material requisition and personnel operations. When violations such as non-painting personnel requisitioning paint or welding personnel requisitioning excessive amounts of welding wire occur, it is difficult to detect and handle them in a timely manner.

[0005] In terms of early warning mechanisms, traditional early warnings are mostly in the form of simple voice broadcasts, without tiered linkage based on risk levels. For high-risk hazards, such as welding sparks igniting materials, the system cannot be activated in a timely manner (e.g., fire doors close, sprinklers start) or emergency information cannot be pushed to the mobile terminals of management personnel, making it difficult to deal with high-risk hazards in a timely manner and easily leading to safety accidents.

[0006] In summary, shipbuilding workshops urgently need an intelligent monitoring technology that adapts to the characteristics of multiple processes to achieve integrated management of process-specific identification, personnel access verification, material correlation traceability, and hierarchical early warning linkage. This would solve the problems of insufficient accuracy and low response efficiency in safety control under the current complex environment, and promote the upgrading of shipbuilding workshop safety management towards intelligence and refinement. Summary of the Invention

[0007] Based on this, a safety monitoring device, method, and computer equipment for shipbuilding production workshops are provided to solve the technical problems of lack of specificity in safety management and control, disconnect between personnel authority and material management, and insufficient early warning linkage in the prior art.

[0008] On the one hand, a safety monitoring device for a shipbuilding workshop is provided, the device including an industrial-grade supercomputer, an edge server, a multimodal monitoring terminal, a permission-based facial recognition camera, an industrial switch, a facial recognition module, a process-specific behavior recognition module, and a graded early warning module; The multimodal monitoring terminal is directly connected to the process-specific behavior recognition module; the authorized face recognition camera is directly connected to the face recognition module; the industrial-grade supercomputer establishes connections with the face recognition module, the process-specific behavior recognition module, the industrial switch, and the hierarchical early warning module respectively; the edge server is connected to the industrial switch; The process-specific behavior recognition module is used to identify whether there are specific violations in each operating area of ​​the ship production workshop based on the video stream, sound data and environmental parameters collected by the multimodal monitoring terminal and output the judgment result; when a specific violation is identified, an instruction is sent to the face recognition module to trigger the authorized face recognition camera to perform secondary verification of the current operator's permissions; The access control facial recognition camera is linked to the facial recognition module. It retrieves access control data stored in the edge server through the industrial switch to verify the identity of the current operator and the access control permissions of personnel involved in confidential processes. When unauthorized personnel are detected to have entered, or when access control is found to be inconsistent after secondary verification, the violation information is sent to the hierarchical early warning module through the industrial-grade supercomputer. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information based on the violation risk level, and implements preset processing measures.

[0009] Furthermore, the multimodal monitoring terminal integrates a high-definition camera, an infrared sensor, a sound sensor, and a temperature sensor to collect video streams, sound data, and environmental parameters; The process-specific behavior recognition module includes a regional behavior judgment unit and a multimodal verification unit. The regional behavior judgment unit incorporates welding violation recognition algorithms, painting violation recognition algorithms, and hoisting violation recognition algorithms. The welding violation recognition algorithm is used to identify specific violations in the workshop welding area, such as not wearing fireproof clothing, improper operation of welding torches, and the risk of ignition from welding sparks. The painting violation recognition algorithm is used to identify specific violations in the painting area, such as not wearing gas masks, improper stacking of painting materials, or excessive concentration of volatile gases. The hoisting violation recognition algorithm is used to identify specific violations in the hoisting area, such as incorrect command gestures or personnel entering dangerous areas. The multimodal verification unit is used to cross-verify the identified specific violations by combining video streams, audio data, and environmental parameters collected by the multimodal monitoring terminal.

[0010] Furthermore, the process-specific behavior recognition module is equipped with a trained ship scene convolutional neural network to identify whether specific violations exist in each operating area of ​​the ship production workshop and output the judgment result. The ship scene convolutional neural network includes: a multi-source data input layer, a feature fusion convolutional layer, an attention mechanism pooling layer, a process classification fully connected layer, and a risk level output layer. The training steps of the ship scene convolutional neural network include: collecting violation samples from each process in the ship workshop, labeling the samples with behavior type and risk level, initializing model parameters through transfer learning, inputting samples for iterative training, optimizing the loss function, and stopping training when the model recognition accuracy is greater than a preset threshold.

[0011] Furthermore, the device also includes a mobile management terminal and a central control display screen, which are connected to the industrial switch. The mobile management terminal obtains early warning information through the industrial switch and supports managers to remotely view on-site videos and send control commands. The central control display screen displays real-time images and videos of violations, heat maps of risk areas, information on involved personnel, violation risk levels, and records of related material requisitions.

[0012] Furthermore, the device also includes a positioning component, a dynamic trajectory tracking module, and a material association verification module; the positioning component is integrated into the multimodal monitoring terminal, the positioning component is connected to the dynamic trajectory tracking module, and the dynamic trajectory tracking module and the material association verification module are connected to the industrial-grade supercomputer. The dynamic trajectory tracking module is used to track the movement trajectory of operators and hoisting equipment in real time through the mobile pan-tilt control of the multimodal monitoring terminal; the dynamic trajectory tracking module adopts a hybrid model of extended Kalman filter (EKF) and particle filter to perform frame parsing and fusion of the video stream from the multimodal monitoring terminal and the positioning signal from the positioning component to construct the motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory.

[0013] Furthermore, the device also includes a scene adaptation parameter configuration module, which is used to adjust the photosensitivity, sampling frequency, and recognition threshold of the multimodal monitoring terminal according to the environmental characteristics of each area of ​​the ship workshop.

[0014] Furthermore, the device also includes a fire protection system, an access control system, and an industrial-grade explosion-proof speaker; the fire protection system, the access control system, and the industrial-grade explosion-proof speaker are respectively connected to the graded early warning module; the graded early warning module uses industrial-grade smart speakers deployed in different areas of the workshop to provide directional voice broadcasts based on the level of violation risk, and triggers the fire protection system to close fire doors and activate sprinklers; the graded early warning module is also used to continuously monitor the execution status of the linked fire protection equipment after the early warning information is issued, and if no feedback of successful execution of the fire protection equipment is received within a preset time, the early warning level is raised and management personnel are notified to handle the situation.

[0015] Furthermore, the graded early warning module includes a three-level response unit, a regional targeted broadcasting unit, and an equipment linkage interface; the three-level response unit is used to trigger preset handling measures according to the level of violation risk; the regional targeted broadcasting unit achieves regional broadcasting through industrial-grade explosion-proof speakers deployed in the welding area, painting area, and hoisting area of ​​the shipbuilding workshop; the equipment linkage interface is used to connect the fire protection system, the emergency stop device of the lifting equipment, and the access control system.

[0016] On the other hand, a method for safety monitoring in a shipbuilding workshop is provided, the method comprising: Install the safety monitoring device for the ship production workshop as described above; Train a ship scene convolutional neural network adapted to multiple process scenarios in a ship workshop, and set the trained ship scene convolutional neural network in the process-specific behavior recognition module; Multimodal monitoring terminals are deployed in various areas of the shipbuilding workshop according to process zones to collect video streams, audio data and environmental parameters in real time. The data is transmitted to an industrial-grade supercomputer via an industrial switch and stored synchronously on an edge server. Shipyard personnel information is stored in the edge server's permission database to ensure real-time data synchronization; The industrial-grade supercomputer performs frame parsing and fusion of the video stream transmitted by the multimodal monitoring terminal and the positioning signal of the positioning component to construct a motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory. Based on the identified potential risky behavior intent, a violation information is sent to the violation behavior. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information according to the violation risk level, and implements preset processing measures.

[0017] In another aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of a safety monitoring method for ship production workshops.

[0018] The aforementioned safety monitoring devices, methods, and computer equipment for shipbuilding workshops, through the structured collaborative design of multimodal monitoring terminals, process-specific behavior recognition modules, facial recognition modules, and hierarchical early warning modules, achieve intelligent full-chain safety management and control of shipbuilding workshops, thereby obtaining the following technical effects: First, the multimodal monitoring terminals acquire video streams, audio data, and environmental parameters, and the process-specific behavior recognition modules identify specific violations, achieving precise monitoring of processes such as welding, painting, and hoisting, thus improving the accuracy of identifying specific violations. Second, when a suspected violation is detected, the system automatically triggers authorized facial recognition cameras to conduct secondary verification of personnel identity and operating permissions, realizing closed-loop control of "behavior recognition—identity verification—permission verification," effectively avoiding security risks caused by unauthorized personnel operations. Third, an industrial-grade supercomputer centrally processes the recognition results and sends the violation information to the hierarchical early warning module, which implements three-level hierarchical early warnings and corresponding handling measures based on the violation risk level, thereby achieving automation and timeliness of the response mechanism. Ultimately, through the aforementioned multi-module linkage, the device can significantly improve the real-time performance, accuracy, and reliability of violation identification, access control, and risk response in shipbuilding workshops, promoting the transformation of safety management from manual reliance to intelligent and proactive approaches. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic block diagram of a safety monitoring device for a ship production workshop in one embodiment of this application; Figure 2 This is a flowchart illustrating a convolutional neural network training method for a ship scene in one embodiment of this application. Figure 3 This is a flowchart illustrating the implementation steps of an industrial-grade supercomputer in one embodiment of this application. Figure 4 This is an internal structural diagram of a computer device in one embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] In one embodiment, such as Figure 1 As shown, a safety monitoring device for a shipbuilding workshop is provided. The device includes an industrial-grade supercomputer, an edge server, a multimodal monitoring terminal, a permission-based facial recognition camera, an industrial switch, a facial recognition module, a process-specific behavior recognition module, and a graded early warning module. The multimodal monitoring terminal is directly connected to the process-specific behavior recognition module; the authorized face recognition camera is directly connected to the face recognition module; the industrial-grade supercomputer establishes connections with the face recognition module, the process-specific behavior recognition module, the industrial switch, and the hierarchical early warning module respectively; the edge server is connected to the industrial switch; The process-specific behavior recognition module is used to identify whether there are specific violations in each operating area of ​​the ship production workshop based on the video stream, sound data and environmental parameters collected by the multimodal monitoring terminal and output the judgment result; when a specific violation is identified, an instruction is sent to the face recognition module to trigger the authorized face recognition camera to perform secondary verification of the current operator's permissions; The access control facial recognition camera is linked to the facial recognition module. It retrieves access control data stored in the edge server through the industrial switch to verify the identity of the current operator and the access control permissions of personnel involved in confidential processes. When unauthorized personnel are detected to have entered, or when access control is found to be inconsistent after secondary verification, the violation information is sent to the hierarchical early warning module through the industrial-grade supercomputer. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information based on the violation risk level, and implements preset processing measures.

[0023] like Figure 1 As shown, Figure 1 This is a schematic diagram of the safety monitoring device for a shipbuilding workshop according to the present invention. An industrial switch serves as the data transmission hub. Multimodal monitoring terminals (including high-definition cameras, various sensors, etc.) collect multi-source data such as video, sound, and environmental data. Material barcode scanning components transmit material verification data, and authorized facial recognition cameras transmit personnel facial data. This data is sent to the material verification module, personnel identification module, and process-specific behavior identification module, respectively. After each module analyzes and judges personnel permissions, material compliance, and process behavior, the results are transmitted to an industrial supercomputer. The industrial supercomputer integrates and processes the data, calls an improved convolutional neural network model for further analysis, and then sends the results to the hierarchical early warning module. Based on the risk level, the hierarchical early warning module coordinates with access control, area anti-riot smart speakers, fire protection systems, etc., to execute corresponding early warning and response operations. Simultaneously, edge servers store various data, a central control display screen shows personnel, material, and risk information, and mobile management terminals support remote viewing and command sending, achieving intelligent monitoring and safety management throughout the entire process.

[0024] Furthermore, the multimodal monitoring terminal integrates a high-definition camera, an infrared sensor, a sound sensor, and a temperature sensor to collect video streams, sound data, and environmental parameters; The process-specific behavior recognition module includes a regional behavior judgment unit and a multimodal verification unit. The regional behavior judgment unit incorporates welding violation recognition algorithms, painting violation recognition algorithms, and hoisting violation recognition algorithms. The welding violation recognition algorithm is used to identify specific violations in the workshop welding area, such as not wearing fireproof clothing, improper operation of welding torches, and the risk of ignition from welding sparks. The painting violation recognition algorithm is used to identify specific violations in the painting area, such as not wearing gas masks, improper stacking of painting materials, or excessive concentration of volatile gases. The hoisting violation recognition algorithm is used to identify specific violations in the hoisting area, such as incorrect command gestures or personnel entering dangerous areas. The multimodal verification unit is used to cross-verify the identified specific violations by combining video streams, audio data, and environmental parameters collected by the multimodal monitoring terminal.

[0025] The audio data includes welding noises, and the environmental parameters include VOC concentration and temperature. By cross-validating the identified specific violations, false alarms caused by a single data source are reduced.

[0026] Furthermore, the process-specific behavior recognition module is equipped with a trained ship scene convolutional neural network to identify whether specific violations exist in each operating area of ​​the ship production workshop and output the judgment result. The ship scene convolutional neural network includes: a multi-source data input layer, a feature fusion convolutional layer, an attention mechanism pooling layer, a process classification fully connected layer, and a risk level output layer. The training steps of the ship scene convolutional neural network include: collecting violation samples from each process in the ship workshop, labeling the samples with behavior type and risk level, initializing model parameters through transfer learning, inputting samples for iterative training, optimizing the loss function, and stopping training when the model recognition accuracy is greater than a preset threshold.

[0027] The preset threshold is preferably 95%.

[0028] like Figure 2 As shown, Figure 2 This is a flowchart of a convolutional neural network training method for ship scenarios. First, samples of violations from multiple processes such as welding, painting, and hoisting are collected (including video, audio, and environmental parameter data). After multi-source data preprocessing and alignment, sample labels (including behavior type and risk level) are manually assigned. Next, model parameters are initialized through transfer learning, constructing a dedicated network structure including a multi-source data input layer, a feature fusion convolutional layer, an attention mechanism pooling layer, a process classification fully connected layer, and a risk level output layer. During iterative training, the loss function is optimized in conjunction with ship industry safety regulations until the model's recognition accuracy reaches ≥95%, at which point training stops and the model is saved. Finally, the model is deployed to predict output behavior information and risk levels, or enters evaluation mode for continuous verification and iterative optimization.

[0029] Furthermore, the device also includes a mobile management terminal and a central control display screen, which are connected to the industrial switch. The mobile management terminal obtains early warning information through the industrial switch and supports managers to remotely view on-site videos and send control commands. The central control display screen displays real-time images and videos of violations, heat maps of risk areas, information on involved personnel, violation risk levels, and records of related material requisitions.

[0030] Specifically, when high-risk behavior or unauthorized personnel enter a confidential area, the central control display screen automatically switches to full-screen warning mode and issues a buzzer alert; and when the mobile management terminal receives a high-risk alarm, the central control display screen automatically enters split-screen warning mode, with the left window displaying real-time violation footage and the right window displaying information about the personnel involved, their operating permissions, and related material requisition records.

[0031] Furthermore, the device also includes a positioning component, a dynamic trajectory tracking module, and a material association verification module; the positioning component is integrated into the multimodal monitoring terminal, the positioning component is connected to the dynamic trajectory tracking module, and the dynamic trajectory tracking module and the material association verification module are connected to the industrial-grade supercomputer. The dynamic trajectory tracking module is used to track the movement trajectory of operators and hoisting equipment in real time through the mobile pan-tilt control of the multimodal monitoring terminal; the dynamic trajectory tracking module adopts a hybrid model of extended Kalman filter (EKF) and particle filter to perform frame parsing and fusion of the video stream from the multimodal monitoring terminal and the positioning signal from the positioning component to construct the motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory.

[0032] The positioning components include RFID positioning components and UWB positioning components. The dynamic trajectory tracking module adopts a hybrid model of extended Kalman filter (EKF) and particle filter to fuse the video stream from the multimodal monitoring terminal, the positioning signals from the RFID positioning components and the UWB positioning components to construct a high-precision motion trajectory. It also introduces an LSTM (Long Short-Term Memory) neural network to model the trajectory sequence and identify preset potential risk behaviors, including "personnel loitering in dangerous areas", "rapidly crossing restricted areas", and "abnormal gathering".

[0033] The multimodal monitoring terminal integrates an anti-fog and dustproof lens, while the authorized facial recognition camera integrates an explosion-proof housing and a wide dynamic range lighting component, adapting to the complex environment of a shipyard. The multimodal monitoring terminal uses an anti-fog and dustproof lens; the authorized facial recognition camera integrates an explosion-proof housing and a wide dynamic range lighting component, adapting to the complex environment of a shipyard, and can accurately capture facial features even in dimly lit corners and dusty areas.

[0034] Furthermore, the device also includes a scene adaptation parameter configuration module, which is used to adjust the photosensitivity, sampling frequency, and recognition threshold of the multimodal monitoring terminal according to the environmental characteristics of each area of ​​the ship workshop.

[0035] Furthermore, the device also includes a fire protection system, an access control system, and an industrial-grade explosion-proof speaker; the fire protection system, the access control system, and the industrial-grade explosion-proof speaker are respectively connected to the graded early warning module; the graded early warning module uses industrial-grade smart speakers deployed in different areas of the workshop to provide directional voice broadcasts based on the level of violation risk, and triggers the fire protection system to close fire doors and activate sprinklers; the graded early warning module is also used to continuously monitor the execution status of the linked fire protection equipment after the early warning information is issued, and if no feedback of successful execution of the fire protection equipment is received within a preset time, the early warning level is raised and management personnel are notified to handle the situation.

[0036] Furthermore, the graded early warning module includes a three-level response unit, a regional targeted broadcasting unit, and an equipment linkage interface; the three-level response unit is used to trigger preset handling measures according to the level of violation risk; the regional targeted broadcasting unit achieves regional broadcasting through industrial-grade explosion-proof speakers deployed in the welding area, painting area, and hoisting area of ​​the shipbuilding workshop; the equipment linkage interface is used to connect the fire protection system, the emergency stop device of the lifting equipment, and the access control system.

[0037] The three-level response unit is used to trigger preset handling measures according to the level of violation risk, including: high risk (fire hazard, personnel fall) triggering sound and light alarm + fire equipment linkage + emergency push to management personnel; medium risk (failure to wear protective equipment as required) triggering area voice broadcast + large screen prompt; low risk (improper placement of materials) triggering text reminder.

[0038] The aforementioned safety monitoring device for shipbuilding workshops achieves intelligent safety management and control across the entire production chain through a structured and collaborative design of multimodal monitoring terminals, process-specific behavior recognition modules, facial recognition modules, and hierarchical early warning modules. This results in the following technical effects: First, the multimodal monitoring terminals acquire video streams, audio data, and environmental parameters, while the process-specific behavior recognition modules identify specific violations, enabling precise monitoring of processes such as welding, painting, and hoisting, thereby improving the accuracy of identifying specific violations. Second, when a suspected violation is detected, the system automatically triggers a facial recognition camera to conduct a secondary verification of personnel identity and operating permissions, achieving closed-loop control of "behavior recognition—identity verification—permission verification," effectively avoiding security risks caused by unauthorized personnel. Third, an industrial-grade supercomputer centrally processes the identification results and sends the violation information to the hierarchical early warning module, which implements a three-tiered early warning system and corresponding handling measures based on the violation risk level, thereby achieving automation and timeliness of the response mechanism. Ultimately, through the aforementioned multi-module linkage, the device can significantly improve the real-time performance, accuracy, and reliability of violation identification, access control, and risk response in shipbuilding workshops, promoting the transformation of safety management from manual reliance to intelligent and proactive approaches.

[0039] In one embodiment, a method for safety monitoring in a shipbuilding workshop is provided, comprising the following steps: Install the safety monitoring device for the ship production workshop as described above; Train a ship scene convolutional neural network adapted to multiple process scenarios in a ship workshop, and set the trained ship scene convolutional neural network in the process-specific behavior recognition module; Multimodal monitoring terminals are deployed in various areas of the shipbuilding workshop according to process zones to collect video streams, audio data and environmental parameters in real time. The data is transmitted to an industrial-grade supercomputer via an industrial switch and stored synchronously on an edge server. Shipyard personnel information is stored in the edge server's permission database to ensure real-time data synchronization; The industrial-grade supercomputer performs frame parsing and fusion of the video stream transmitted by the multimodal monitoring terminal and the positioning signal of the positioning component to construct a motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory. Based on the identified potential risky behavior intent, a violation information is sent to the violation behavior. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information according to the violation risk level, and implements preset processing measures.

[0040] like Figure 3 As shown, the specific steps of the implementation process of the industrial-grade supercomputer are as follows.

[0041] (a) Install intelligent security monitoring devices According to the characteristics of the shipbuilding workshop processes, complete the installation and connection of each piece of equipment.

[0042] Multimodal monitoring terminal deployment: one unit is deployed every 50 square meters in the welding area, one unit is deployed in the painting area according to the ventilation zone, and one unit is arranged in a ring around the working radius in the hoisting area to ensure full coverage of the work areas of each process.

[0043] Access control facial recognition camera deployment: Deployed at key locations such as workshop entrances and entrances to areas involving confidential processes (such as the assembly area for core ship components) for preliminary verification of personnel identity and permissions.

[0044] Other equipment configuration: Configure an appropriate number of mobile management terminals and central control display screens according to management needs, and build a network transmission architecture through industrial switches to ensure connectivity of all devices.

[0045] (II) Training a convolutional neural network for ship scenes Data Acquisition: Using multimodal monitoring terminals and other specialized equipment, multimodal sample data (video, audio, environmental parameters, etc.) are collected in the welding, painting, hoisting and other process areas of the shipbuilding workshop under different working conditions and time periods. This includes violations (such as not wearing fireproof clothing in the welding area, not wearing gas masks in the painting area, and non-standard hand gestures in the hoisting area) and normal work behaviors.

[0046] Sample labeling: Experts familiar with ship workshop safety regulations and procedures were organized to manually label the collected samples according to ship industry safety regulations. The labeling content included the behavior type, risk level and the relevant process scenario.

[0047] Model Training: Using transfer learning, a suitable pre-trained convolutional neural network model is selected. The model parameters are initialized based on the convolutional neural network structure for the ship scene (multi-source data input layer, feature fusion convolutional layer, attention mechanism pooling layer, process classification fully connected layer, and risk level output layer). The labeled sample data is input into the model for iterative training. The loss function is optimized in combination with ship industry safety regulations until the model recognition accuracy is ≥95%. The trained model parameters are then saved.

[0048] (III) Deploy multimodal monitoring terminals and synchronously collect dynamic data from various areas. Multi-source data acquisition: After the multi-modal monitoring terminals are deployed in the process area, they collect video streams (25fps), sound data (44.1kHz sampling rate), and environmental parameters (temperature, VOC concentration, sampling interval 1s) in real time. The data is transmitted to an industrial-grade supercomputer through an industrial switch, and the edge server stores the real-time data synchronously to ensure data integrity and traceability.

[0049] Access control database establishment: Organize the information of shipyard personnel (name, work group, and operable procedures) into a standard format and enter it into the access control database on the edge server. When new employees join the company or when employee positions or permissions change, the database is updated in a timely manner to ensure real-time data synchronization.

[0050] Dynamic trajectory modeling: Industrial-grade supercomputers perform frame parsing and fusion of video streams, RFID and UWB data transmitted from multimodal monitoring terminals. A hybrid model of extended Kalman filter (EKF) and particle filter is used to construct high-precision motion trajectories. A pre-trained LSTM neural network is used to identify potential risky behavioral intentions in the trajectory, track the motion status in real time, and mark abnormal stay situations (such as personnel staying in the hoisting danger zone for more than 10 seconds) and risky movement patterns, providing data support for subsequent behavior recognition and risk warning.

[0051] (iv) Implement the hierarchical identification and linkage early warning process Multimodal fusion recognition: The process-specific behavior recognition module receives multi-source data in real time and calls specialized algorithms for analysis based on the corresponding process area. In the welding area, the risk of spark ignition is determined by "flame temperature > 300℃ + lack of fireproof clothing color characteristics"; in the painting area, the risk of exceeding the standard is triggered by "lack of gas mask outline + VOC concentration > 800ppm"; in the hoisting area, the risk of command violations is identified by "gesture matching degree with standard library < 60%", and the behavior type and risk level are output.

[0052] Dynamic access control verification: When personnel enter a confidential work area, or when the work area's dedicated behavior recognition module identifies high-risk behavior and issues an instruction, the access control facial recognition camera automatically captures facial image data and transmits it to the facial recognition module. The facial recognition module first performs facial feature comparison to confirm the personnel's basic identity, and then retrieves the corresponding personnel's operation permission information from the edge server's permission database for verification. If the permissions do not match, a signal is immediately sent to the tiered early warning module, triggering a high-level warning and locking the area's access control.

[0053] Tiered Response Execution: The tiered early warning module activates corresponding measures based on the risk level. High risk (such as fire hazards or falls) triggers audible and visual alarms, fire equipment activation, and emergency notifications to management personnel; medium risk (such as failure to wear protective equipment as required) triggers area voice broadcasts and large-screen prompts; low risk (such as improper material placement) triggers text reminders. Simultaneously, the module monitors the status of activated equipment; if no feedback is received within a preset time, the warning is escalated and management personnel are notified.

[0054] Emergency Tracing: When a security incident occurs, the industrial-grade supercomputer automatically retrieves video, audio, personnel trajectory records, and equipment response logs for 10 minutes before and after the incident to generate an "Incident Tracing Report," which includes the time of the violation, the personnel involved, the early warning response process, and the equipment execution status, for managers to review and analyze, and optimize security control measures.

[0055] This application achieves accurate identification of specific violations in processes such as welding, painting, and hoisting by using process-specific identification algorithms and convolutional neural networks for ship scenarios; it strictly controls personnel operation permissions by combining facial recognition and permission verification; it achieves efficient response by linking multiple devices based on a risk level-based hierarchical early warning mechanism; and in particular, it realizes early identification and closed-loop management of safety risks through the collaborative working mechanism between modules and innovative dynamic trajectory analysis, promoting the transformation of safety management in ship workshops from "passive response" to "proactive prevention", and effectively improving the safety management level of ship production workshops.

[0056] The core advantages of this application are reflected in: Precise identification of specific processes: By using regional behavior judgment and multimodal verification, dedicated identification logic is designed for processes such as welding, painting, and hoisting. Combined with convolutional neural networks for ship scenarios, the accuracy rate of identifying specific violations reaches over 95%.

[0057] Stricter personnel access control: Deeply link facial recognition with access verification for confidential processes, and introduce a behavior recognition-triggered secondary verification mechanism to achieve triple control of "identity + access + behavior" and eliminate the risk of unauthorized operation.

[0058] Intelligent early warning response: Based on a three-level risk level-based targeted broadcasting and equipment linkage mechanism, and by introducing closed-loop control logic, the entire chain from risk identification to emergency response is automated, reducing the response time for high-risk hazards to within 10 seconds, effectively improving the safety management level of ship production workshops.

[0059] Forward-looking trajectory analysis: Through innovative multi-source trajectory fusion algorithms and LSTM behavioral intent recognition, early detection and warning of potential risks can be achieved, promoting the transformation of safety management from "passive response" to "proactive prevention".

[0060] The aforementioned safety monitoring method for shipbuilding workshops achieves intelligent safety management and control across the entire process by employing a structured collaborative design of multimodal monitoring terminals, process-specific behavior recognition modules, facial recognition modules, and hierarchical early warning modules. This results in the following technical effects: First, the multimodal monitoring terminals acquire video streams, audio data, and environmental parameters, while the process-specific behavior recognition modules identify specific violations, enabling precise monitoring of processes such as welding, painting, and hoisting, thereby improving the accuracy of identifying specific violations. Second, when a suspected violation is detected, the system automatically triggers a facial recognition camera to conduct secondary verification of personnel identity and operational permissions, achieving closed-loop control of "behavior recognition—identity verification—permission verification," effectively avoiding security risks caused by unauthorized personnel. Third, an industrial-grade supercomputer centrally processes the identification results and sends the violation information to the hierarchical early warning module, which implements a three-tiered early warning system and corresponding handling measures based on the violation risk level, thus achieving automation and timeliness of the response mechanism. Ultimately, through the aforementioned multi-module linkage, the device can significantly improve the real-time performance, accuracy, and reliability of violation identification, access control, and risk response in shipbuilding workshops, promoting the transformation of safety management from manual reliance to intelligent and proactive approaches.

[0061] For specific limitations on safety monitoring methods in ship production workshops, please refer to the limitations on safety monitoring devices in ship production workshops mentioned above, which will not be repeated here.

[0062] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores safety monitoring data for the shipbuilding production workshop. The network interface communicates with external terminals via a network. When the computer program is executed by the processor, it implements a safety monitoring method for the shipbuilding production workshop.

[0063] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0064] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Install the safety monitoring device for the ship production workshop as described above; Train a ship scene convolutional neural network adapted to multiple process scenarios in a ship workshop, and set the trained ship scene convolutional neural network in the process-specific behavior recognition module; Multimodal monitoring terminals are deployed in various areas of the shipbuilding workshop according to process zones to collect video streams, audio data and environmental parameters in real time. The data is transmitted to an industrial-grade supercomputer via an industrial switch and stored synchronously on an edge server. Shipyard personnel information is stored in the edge server's permission database to ensure real-time data synchronization; The industrial-grade supercomputer performs frame parsing and fusion of the video stream transmitted by the multimodal monitoring terminal and the positioning signal of the positioning component to construct a motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory. Based on the identified potential risky behavior intent, a violation information is sent to the violation behavior. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information according to the violation risk level, and implements preset processing measures.

[0065] For specific limitations on the steps a processor takes when executing a computer program, please refer to the limitations on the methods for safety monitoring in shipbuilding workshops mentioned above, which will not be repeated here.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A safety monitoring device for a shipbuilding workshop, characterized in that, The device includes an industrial-grade supercomputer, an edge server, a multimodal monitoring terminal, a permission-based facial recognition camera, an industrial switch, a facial recognition module, a process-specific behavior recognition module, and a graded early warning module; The multimodal monitoring terminal is directly connected to the process-specific behavior recognition module; the authorized face recognition camera is directly connected to the face recognition module; the industrial-grade supercomputer establishes connections with the face recognition module, the process-specific behavior recognition module, the industrial switch, and the hierarchical early warning module respectively; the edge server is connected to the industrial switch; The process-specific behavior recognition module is used to identify whether there are specific violations in each operating area of ​​the ship production workshop based on the video stream, sound data and environmental parameters collected by the multimodal monitoring terminal and output the judgment result; when a specific violation is identified, an instruction is sent to the face recognition module to trigger the authorized face recognition camera to perform secondary verification of the current operator's permissions; The access control facial recognition camera is linked to the facial recognition module. It retrieves access control data stored in the edge server through the industrial switch to verify the identity of the current operator and the access control permissions of personnel involved in confidential processes. When unauthorized personnel are detected to have entered, or when access control is found to be inconsistent after secondary verification, the violation information is sent to the hierarchical early warning module through the industrial-grade supercomputer. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information based on the violation risk level, and implements preset processing measures.

2. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The multimodal monitoring terminal integrates a high-definition camera, infrared sensor, sound sensor, and temperature sensor to collect video streams, sound data, and environmental parameters. The process-specific behavior recognition module includes a regional behavior judgment unit and a multimodal verification unit. The regional behavior judgment unit incorporates welding violation recognition algorithms, painting violation recognition algorithms, and hoisting violation recognition algorithms. The welding violation recognition algorithm is used to identify specific violations in the workshop welding area, such as not wearing fireproof clothing, improper operation of welding torches, and the risk of ignition from welding sparks. The painting violation recognition algorithm is used to identify specific violations in the painting area, such as not wearing gas masks, improper stacking of painting materials, or excessive concentration of volatile gases. The hoisting violation recognition algorithm is used to identify specific violations in the hoisting area, such as incorrect command gestures or personnel entering dangerous areas. The multimodal verification unit is used to cross-verify the identified specific violations by combining video streams, audio data, and environmental parameters collected by the multimodal monitoring terminal.

3. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The process-specific behavior recognition module is equipped with a trained ship scene convolutional neural network to identify whether specific violations exist in each operating area of ​​the ship production workshop and output the judgment result. The ship scene convolutional neural network includes: a multi-source data input layer, a feature fusion convolutional layer, an attention mechanism pooling layer, a process classification fully connected layer, and a risk level output layer. The training steps of the ship scene convolutional neural network include: collecting violation samples from each process in the ship workshop, labeling the samples with behavior type and risk level, initializing model parameters through transfer learning, inputting samples for iterative training, optimizing the loss function, and stopping training when the model recognition accuracy is greater than a preset threshold.

4. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The device also includes a mobile management terminal and a central control display screen, which are connected to the industrial switch. The mobile management terminal obtains early warning information through the industrial switch and supports managers to remotely view on-site videos and send control commands. The central control display screen displays real-time images and videos of violations, heat maps of risk areas, information on involved personnel, violation risk levels, and records of related material requisitions.

5. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The device further includes a positioning component, a dynamic trajectory tracking module, and a material association verification module. The positioning component is integrated into the multimodal monitoring terminal and connected to the dynamic trajectory tracking module. The dynamic trajectory tracking module and the material association verification module are connected to the industrial-grade supercomputer. The dynamic trajectory tracking module is used to track the movement trajectory of operators and hoisting equipment in real time through the mobile pan-tilt control of the multimodal monitoring terminal. The dynamic trajectory tracking module adopts a hybrid model of extended Kalman filtering and particle filtering to perform frame parsing and fusion of the video stream from the multimodal monitoring terminal and the positioning signal from the positioning component to construct the movement trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the movement trajectory.

6. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The device also includes a scene adaptation parameter configuration module, which is used to adjust the photosensitivity, sampling frequency, and recognition threshold of the multimodal monitoring terminal according to the environmental characteristics of each area of ​​the ship workshop.

7. The safety monitoring device for shipbuilding workshops according to claim 1, characterized in that, The device also includes a fire protection system, an access control system, and an industrial-grade explosion-proof speaker; the fire protection system, the access control system, and the industrial-grade explosion-proof speaker are respectively connected to the graded early warning module; the graded early warning module uses industrial-grade smart speakers deployed in different areas of the workshop to provide directional voice broadcasts based on the level of violation risk, and triggers the fire protection system to close fire doors and activate sprinklers; the graded early warning module is also used to continuously monitor the execution status of the linked fire protection equipment after the early warning information is issued, and if no feedback of successful execution of the fire protection equipment is received within a preset time, the early warning level is raised and management personnel are notified to handle the situation.

8. The safety monitoring device for shipbuilding workshops according to claim 7, characterized in that, The graded early warning module includes a three-level response unit, a regional targeted broadcasting unit, and an equipment linkage interface. The three-level response unit is used to trigger preset handling measures according to the level of violation risk. The regional targeted broadcasting unit achieves regional broadcasting through industrial-grade explosion-proof speakers deployed in the welding area, painting area, and hoisting area of ​​the shipbuilding workshop. The equipment linkage interface is used to connect the fire protection system, the emergency stop device of the lifting equipment, and the access control system.

9. A method for safety monitoring in a shipbuilding workshop, characterized in that, include: The ship production workshop safety monitoring device as described in any one of claims 1 to 8 is provided; Train a ship scene convolutional neural network adapted to multiple process scenarios in a ship workshop, and set the trained ship scene convolutional neural network in the process-specific behavior recognition module; Multimodal monitoring terminals are deployed in various areas of the shipbuilding workshop according to process zones to collect video streams, audio data and environmental parameters in real time. The data is transmitted to an industrial-grade supercomputer via an industrial switch and stored synchronously on an edge server. Shipyard personnel information is stored in the edge server's permission database to ensure real-time data synchronization; The industrial-grade supercomputer performs frame parsing and fusion of the video stream transmitted by the multimodal monitoring terminal and the positioning signal of the positioning component to construct a motion trajectory, and uses a long short-term memory neural network to identify potential risky behavioral intentions in the motion trajectory. Based on the identified potential risky behavior intent, a violation information is sent to the violation behavior. The graded early warning module identifies the violation risk level corresponding to the violation information, sends early warning information according to the violation risk level, and implements preset processing measures.

10. A computer 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 computer program, it implements the steps of the method of claim 9.

Images