Multi-target intelligent cruise laser gas ignition and safety protection system
By using a unified multi-task visual analysis model, deep integration of flame recognition and intrusion detection is achieved, solving the problem of poor coordination in existing technologies, improving recognition accuracy and environmental adaptability, simplifying hardware composition and maintenance, and enhancing system reliability and adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU DAYOU PETROLEUM DRILLING & EXPLOITING ENGINEERING CO
- Filing Date
- 2026-01-28
- Publication Date
- 2026-04-24
AI Technical Summary
Existing laser ignition and security protection systems lack a unified vision platform, resulting in poor coordination between flame recognition and security area intrusion detection. The systems are complex to deploy, have high maintenance costs, and reduced reliability. They are unable to effectively identify flames and intrusion target types (people/vehicles/animals) or accurately assess the flame status.
By using artificial intelligence algorithms to process the same video stream in parallel, the system can accurately identify the flame status of the ignition target and detect intrusion into the security area around the laser beam path. Based on the dynamic interaction and collaboration of the analysis results, the system can intelligently control the start-up and shutdown, power adjustment, and gimbal cruise behavior of the laser ignition device, thus constructing a safe, efficient, and adaptive integrated intelligent ignition and protection system.
This system achieves deep integration of flame recognition and intrusion detection, improving the system's recognition accuracy and environmental adaptability, reducing the false positive rate, simplifying hardware composition and maintenance, and enhancing the system's reliability and adaptability.
Smart Images
Figure CN121921900A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas exploration and production safety technology, specifically to a multi-target intelligent cruise laser gas ignition and safety protection system. Background Technology
[0002] In the process of oil and gas exploration and production, combustible gases produced in the wellbore must be ignited and burned to ensure safe discharge. Traditional ignition methods, such as oil pan flames, are inefficient and unsafe. Laser ignition technology, due to its advantages such as long operating distance, remote non-contact operation, and ability to work continuously for extended periods, is gradually becoming the mainstream solution.
[0003] Existing laser ignition technologies often employ separate physical sensor systems to achieve specific functions, which has significant drawbacks. Flame detection typically relies on ultraviolet or infrared sensors, while security protection depends on independent electronic fence modules (such as laser rangefinders or infrared photodiodes). Data from these two systems is not shared, lacking a unified intelligent decision-making center and resulting in poor system coordination. Physical sensors are highly susceptible to interference from environmental factors such as sunlight (especially ultraviolet radiation), temperature changes, and welding arc light, leading to decreased detection accuracy and a high risk of false alarms or missed alarms. Existing solutions offer limited information dimensions, sensing only single physical quantities (such as beam obstruction or ultraviolet intensity), failing to acquire the visual context of the scene, accurately identify the type of intrusion target (person / vehicle / animal), or precisely assess the flame state. To achieve multi-target patrol and all-around protection, a large number of independent physical sensors must be deployed, resulting in complex system deployment, high maintenance costs, and reduced reliability.
[0004] Therefore, there is a lack of a new laser ignition and safety protection system based on a unified vision platform that can simultaneously and intelligently identify the intrusion status of flames and safe areas, and achieve coordinated control of the two. Summary of the Invention
[0005] This invention overcomes the shortcomings of existing technologies and provides a multi-target intelligent cruise laser gas ignition and safety protection system. It uses artificial intelligence algorithms to process the same video stream in parallel, and simultaneously achieves accurate identification of the flame state of the ignition target and intrusion detection of the safety area around the laser beam path. Based on the dynamic interaction and collaboration of the analysis results of the two, it intelligently controls the start and stop, power adjustment and gimbal cruise behavior of the laser ignition device, and constructs a safe, efficient and adaptive integrated intelligent ignition and protection system.
[0006] The technical solution adopted by this invention is as follows: This solution provides a multi-target intelligent cruise laser gas ignition and safety protection system, including an intelligent launch cabin, a control host, and a tablet terminal. The intelligent launch cabin is set high in the area to be protected and includes a laser emission unit, a visible light video acquisition unit, an indicator laser unit, a pan-tilt unit, and an alarm unit; the visible light video acquisition unit is coaxially or parallel to the laser emission unit in its optical path, used to acquire real-time video images containing the target area and the safety protection area; the control host is communicatively connected to the intelligent launch cabin and also to the tablet terminal. The control host includes an AI algorithm processing module, a task management module, a laser control module, and a communication module; the AI algorithm processing module receives and analyzes real-time video images, organizing continuous real-time video images into a video sequence; the AI algorithm processing module internally runs a unified multi-task visual analysis model, which includes a parallel flame recognition sub-model and an input... The intrusion detection sub-model and the flame recognition sub-model are used to identify flames within the target area from video images and output flame status information, which includes at least the presence, area, and height of the flame. The intrusion detection sub-model identifies human bodies or vehicles within a preset security protection area from video images and outputs intrusion status information, which includes at least the presence, type, and location of the intrusion target. The task management module acquires the flame status information and intrusion status information and generates comprehensive control commands based on preset collaborative decision-making logic. The task management module also outputs system operating parameters and alarm information. The laser control module controls the laser emitting unit to start / stop, control the power, and target point of laser emission according to the comprehensive control commands, and controls the pan-tilt unit to perform multi-target patrol.
[0007] Furthermore, the collaborative decision-making logic includes: when the intrusion status information indicates that an intrusion has been detected, regardless of the flame status, the integrated control command is to immediately stop laser emission and trigger the alarm unit to issue an audible and visual alarm; when the intrusion status information indicates that there is no intrusion, the integrated control command is determined based on the flame status information: if the flame exists and its area and height both reach the preset safe combustion threshold, the integrated control command is to stop laser emission or reduce the laser power to the maintenance mode; if the flame does not exist, or the flame area and height are lower than the preset safe combustion threshold, the integrated control command is to control the laser emission unit to emit laser at the current target point with a preset ignition power and control the gimbal to cruise along a preset cruise path.
[0008] Furthermore, the flame recognition sub-model and the intrusion detection sub-model share the feature encoding backbone network in the unified multi-task visual analysis model. The feature encoding backbone network is used to extract spatiotemporal features from the input video image. The feature encoding backbone network adopts a three-dimensional convolutional neural network architecture to capture the dynamic texture features of flames and the motion trajectory features of intrusion targets in the video sequence. Based on the shared feature map output by the feature encoding backbone network, the flame recognition sub-model processes the data through a flame-specific head branch network to output flame status information. The flame-specific head branch network includes a flame classification head, a flame segmentation head, and a flame regression head. Based on the same shared feature map, the intrusion detection sub-model processes the data through an intrusion detection-specific head branch network to output intrusion status information.
[0009] Furthermore, the AI algorithm processing module also includes an environmental perception submodule. Based on the statistical features of video images, including brightness, color distribution, and sharpness, the environmental perception submodule dynamically assesses the visibility of the current environment and outputs the environmental visibility level. The task management module is also used to dynamically adjust the detection confidence thresholds of the flame recognition submodel and the intrusion detection submodel in combination with the environmental visibility level: when the environmental visibility level is low, the detection confidence threshold of the flame recognition submodel is reduced to improve the flame detection rate, while the detection confidence threshold of the intrusion detection submodel is increased to reduce false alarms caused by environmental interference.
[0010] Furthermore, the AI algorithm processing module also includes a feature fusion and correlation analysis module. This module receives flame status information output by the flame recognition sub-model and intrusion status information output by the intrusion detection sub-model, performs spatial and temporal correlation analysis on the flame status information and intrusion status information, and outputs the correlation analysis results. If the detected burning position of the flame overlaps with or is close to a preset distance from the detected intrusion target in space, it is determined to be a high-risk event, and a high-risk alarm signal is output to the task management module. The task management module generates a comprehensive control command based on the high-risk alarm signal to immediately stop laser emission and enhance the alarm.
[0011] Furthermore, the intelligent launch cabin also includes a ranging unit, which is connected to the control host to measure the distance from the current target point to the intelligent launch cabin and output a ranging signal. The laser control module is also used to automatically adjust the focusing parameters of the laser emission unit according to the ranging signal to ensure that the size of the laser spot at the target point is minimized and the energy density is maximized. The mission management module also inputs the ranging signal into the flame recognition sub-model as a reference parameter to adjust the accuracy of flame height estimation.
[0012] Furthermore, the tablet terminal sends user commands to the control host, receives and displays real-time video images, flame status information, intrusion status information, system operating parameters, and alarm information from the control host; the tablet terminal is equipped with a security zone calibration interface, which allows users to draw the boundaries of preset security protection areas on the real-time video images, and sends the generated boundary data to the control host for storage, for use by the intrusion detection sub-model; the tablet terminal is also equipped with a target point setting and cruise path planning interface, which allows users to mark multiple target points on the real-time video images and set the pan-tilt unit's cruise sequence and dwell time between target points, and the cruise path plan is also sent to the control host.
[0013] Furthermore, after receiving the real-time video image, the AI algorithm processing module executes the following processing steps: Step S101: Preprocess the real-time video image, including scaling to a uniform resolution, normalizing pixel values and dynamic brightness compensation, to obtain a preprocessed real-time video image; combine multiple consecutive preprocessed real-time video images into a video sequence. Step S102: Input the video sequence into the feature encoding backbone network to extract multi-level and multi-dimensional spatiotemporal features; Step S103: The extracted spatiotemporal features are input into the flame recognition sub-model and the intrusion detection sub-model in parallel. Step S104: The flame recognition sub-model performs flame-related feature enhancement on the spatiotemporal features, determines whether there is a flame in the target area of the current real-time video image through the flame classification head, outputs a pixel-level mask of the flame through the flame segmentation head, and estimates the flame height through the flame regression head; finally, it outputs the flame state information. Step S105: The intrusion detection sub-model performs human and vehicle target detection based on spatiotemporal features, identifies targets within the preset security protection area, and outputs intrusion status information, including the target's category, location bounding box, and confidence level. Step S106: The feature fusion and correlation analysis module integrates the flame state information and the intrusion state information to determine the spatial relationship between the flame and the intrusion target, and outputs the correlation analysis results. Step S107: The environmental perception submodule assesses environmental visibility based on the statistical features of real-time video images and outputs the environmental level. In step S108, the task management module receives the correlation analysis results, environmental level, flame status information, and intrusion status information, executes collaborative decision-making logic, and generates comprehensive control commands.
[0014] Furthermore, the control host also includes a historical data storage module for long-term storage of analyzed flame status information, intrusion status information, system operating parameters, and real-time video images; the AI algorithm processing module also includes a model self-learning and optimization module, which periodically obtains real-time video images, flame status information, and intrusion status information from the historical data storage module to incrementally learn the unified multi-task visual analysis model in order to adapt to new scenarios, lighting conditions, or target types and improve the model's generalization ability.
[0015] Furthermore, the system also includes a backup power management module, which is connected to the laser emitting unit, the control host and the visible light video acquisition unit. It is used to automatically switch to backup power supply when the main power is interrupted and send a power failure alarm to the mission management module. After receiving the power failure alarm, the mission management module controls the intelligent launch cabin to safely shut down all launch functions and sends the current system status to the remote monitoring platform through the communication module.
[0016] Compared with the prior art, the beneficial effects of the present invention are: (1) Breaking away from the traditional design approach based on discrete physical sensors, this system integrates the two core security functions of flame recognition and intrusion detection into the same computation and decision-making process through a unified multi-task visual analysis model. The two sub-tasks share the underlying feature extraction, and their analysis results are used for collaborative decision-making within the task management module. For example, intrusion detection has the highest priority and can immediately cut off the laser in any state, but the system will record the flame state information and intelligently decide whether to restore ignition based on the flame state after the intrusion is eliminated; the failure of flame recognition (such as failure to ignite) can also reduce the detection confidence threshold of the intrusion detection sub-model or strengthen the monitoring of specific areas, thereby achieving a dynamic balance and mutual promotion between protection and ignition functions. (2) The unified multi-task visual analysis model utilizes the rich spatiotemporal information of video, and can comprehensively identify flames from multiple dimensions such as texture, color, and dynamic contours, effectively resisting interference from strong light and smoke, and reducing the false judgment rate. For intrusion detection, the system can identify based on the overall appearance and movement pattern of the human body and vehicle, rather than simply relying on beam blocking, so it can still maintain a high detection accuracy under complex lighting, occlusion, and rainy / foggy weather. The introduction of the environmental perception submodule enables the system to adaptively adjust the detection sensitivity, further enhancing its all-weather working capability; (3) The unified multi-task visual analysis model not only provides a simple "present / absent" result, but also provides quantitative information such as the area and height of the flame to help determine whether ignition is successful and to roughly estimate the amount of gas; it can also provide the type (person / vehicle) and precise location of the intrusion target, making security protection more precise and proactive. The feature fusion and correlation analysis module can determine the spatial relationship between the flame and the intrusion target, identify high-risk events, and thus trigger a higher level of emergency response; (4) A high-precision visible light video acquisition unit replaced multiple dedicated physical sensors, simplifying the hardware composition, wiring, installation, and debugging of the intelligent launch cabin. The AI algorithm processing module makes the maintenance and upgrade of the software system (such as algorithm updates) more centralized and efficient. It reduces the number of physical sensor failure points and improves the overall reliability of the system; (5) Through the tablet terminal, users can easily customize the safety protection area, set multiple target points and cruise paths, enabling the system to adapt to various complex blowout pool layouts and operational requirements. The incremental learning capability of the multi-task visual analysis model enables the system to continuously evolve during use, adapt to new environments, and extend the effective life cycle of the system; (6) The system not only realizes a real-time closed loop from perception (video analysis) to decision-making (collaborative control) and then to execution (laser, PTZ, alarm), but also introduces a historical data storage module and model self-learning and optimization functions, forming a continuous improvement closed loop of "perception-decision-execution-learning-optimization". This enables the system to have the ability to continuously evolve and improve itself while ensuring safe and efficient operation. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram of the operation of the AI algorithm processing module in the system of the present invention; The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation
[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0019] Example: Please see Figures 1-2 This embodiment presents a multi-target intelligent cruise laser gas ignition and safety protection system, applied in the oil and gas exploration and production process. It safely and intelligently ignites and combusts emitted combustible gases while simultaneously ensuring the safety of personnel and equipment at the ignition site. The system includes an intelligent launch cabin, a control host, and a tablet terminal.
[0020] The intelligent launch cabin is positioned high up near the blowhole; the control unit is housed in a protective enclosure on-site; and the tablet terminal is the mobile device used by the operators. All three are interconnected to enable data and command exchange.
[0021] The intelligent launch cabin includes a visible light video acquisition unit, a laser emission unit, an indicator laser unit, an alarm unit, a ranging unit, and a pan-tilt unit. All components of the intelligent launch cabin are ultimately connected into a single unit via an integrated shell and cables, providing a high level of protection to withstand harsh outdoor environments.
[0022] The visible light video acquisition unit includes a high-precision visible light camera with optical zoom capability. This unit is used to acquire real-time video images, which originate from the target area and the safety protection area. The imaging optical path of the visible light video acquisition unit and the main laser optical path ultimately output by the laser emission unit are strictly adjusted to be coaxial or parallel, maintaining a fixed, small spatial distance. In this embodiment, at a working distance of 100 meters, the distance between the two light points is less than 5 centimeters, facilitating precise aiming by the operator using real-time video images.
[0023] The laser emitting unit includes a laser generator, an optical fiber transmission line, and a laser focusing lens. The laser generator uses a 500W high-power air-cooled fiber laser to produce a high-energy laser beam. The laser generator transmits the laser beam to the laser focusing lens via a high-temperature resistant optical fiber transmission line. The laser focusing lens is an electrically adjustable zoom focusing lens that precisely focuses the incident parallel laser beam onto a distant target point, forming a high-temperature ignition point.
[0024] The indicator laser unit includes a low-power visible laser; in this embodiment, a 1W green laser is used. The emission path of the indicator laser unit is also arranged parallel to the main laser path. When the laser emission unit is working, the indicator laser unit emits light synchronously, forming a visible indicator line to warn and mark the laser path for on-site personnel. The indicator laser unit is connected to the control host and is controlled by a synchronous switch.
[0025] The alarm unit includes an audible alarm (high-decibel buzzer) and a visual alarm (flashing warning light) to issue local alerts when the system detects intrusion or other anomalies.
[0026] The ranging unit is a laser ranging module used to measure the distance from the intelligent launch cabin to the target point being aimed at, and outputs the ranging signal to the control host to assist in automatic focusing and improve the accuracy of flame height estimation.
[0027] The gimbal serves as a mechanical support and moving component. Its top and front end are equipped with a visible light video acquisition unit, an indicator laser unit, and a ranging unit via fixed brackets. A laser focusing lens is fixed at the bottom of the gimbal. The gimbal is connected to the control host to achieve multi-angle aiming and cruising.
[0028] The control host is the core computing and control unit of the system. It communicates not only with the intelligent launch cabin but also with the tablet terminal. The control host includes an AI algorithm processing module, a task management module, a laser control module, a communication module, a historical data storage module, and a backup power management module.
[0029] The AI algorithm processing module is the core innovation of this invention. This module receives and analyzes real-time video images from the visible light video acquisition unit, organizing the continuous real-time video images into a video sequence. Internally, the AI algorithm processing module runs a unified multi-task visual analysis model. This model includes parallel flame recognition and intrusion detection sub-models. The multi-task visual analysis model is based on a deep learning neural network architecture, with its core design lying in the fact that the flame recognition and intrusion detection sub-models share a unified feature encoding backbone network.
[0030] The feature encoding backbone network employs a three-dimensional convolutional neural network architecture. The input video sequence (e.g., 16 consecutive frames) undergoes multi-layer processing by the feature encoding backbone network, progressively extracting spatiotemporal features from low-level edges and textures to high-level semantics, and outputting shared feature maps. These feature maps are shared by the flame recognition sub-model and the intrusion detection sub-model, avoiding the need to extract features separately for each task, improving computational efficiency, and enabling the model to learn general visual features that are effective for both flames and intrusion targets.
[0031] The flame recognition sub-model is based on a shared feature map and is processed by a flame-specific head branch network to output flame state information. The flame-specific head branch network includes a flame classification head, a flame segmentation head, and a flame regression head. The flame classification head is used to focus on the color and dynamic texture of the flame. The flame segmentation head outputs a pixel-level mask of the flame for calculating its area. The flame regression head combines the position and ranging data of the flame in the real-time video image to estimate its actual height. The flame state information includes the presence, area, and height of the flame.
[0032] The intrusion detection sub-model, based on the same shared feature map, employs a target detection network to detect and identify human bodies and vehicles within a pre-defined security protection area, and outputs intrusion status information. This intrusion status information includes the presence or absence of an intrusion, the type of intrusion target (human body / vehicle), its location bounding box, and its confidence level.
[0033] The AI algorithm processing module also includes an environmental perception submodule. The environmental perception submodule dynamically assesses the visibility of the current environment (divided into four levels: excellent, good, medium, and poor) by calculating the average brightness, contrast, saturation, and sharpness indicators of the video image, and outputs the environmental visibility level. The output environmental visibility level provides a basis for subsequent dynamic threshold adjustment.
[0034] The AI algorithm processing module also includes a feature fusion and correlation analysis module. This module receives flame status information from the flame recognition sub-model and intrusion status information from the intrusion detection sub-model, performs spatial and temporal correlation analysis on the flame status information and intrusion status information, and outputs the correlation analysis results. If the detected flame location overlaps with or is close to the detected intrusion target location within a certain radius of the flame area, it is determined to be a high-risk event, and a high-risk alarm signal is output to the task management module. The task management module generates a comprehensive control command to immediately stop laser emission and enhance the alarm based on the preset collaborative decision-making logic and the high-risk alarm signal.
[0035] The AI algorithm processing module also includes a model self-learning and optimization module. The model self-learning and optimization module regularly (weekly) reads real-time video images, flame status information and intrusion status information from the historical data storage module to incrementally learn the unified multi-task visual analysis model in operation. This enables the system to continuously learn new knowledge, adapt to environmental changes, and improve long-term operating performance.
[0036] The task management module receives all outputs from the AI algorithm processing module, including flame status information, intrusion status information, environmental visibility level, and high-risk alarm signals. The task management module also receives user commands from the tablet terminal (including manual start / stop and mode switching). Internally, the task management module runs preset collaborative decision-making logic and generates comprehensive control commands, which are sent to the laser control module, alarm unit, etc. The task management module also outputs system operating parameters and alarm information.
[0037] The core principles of collaborative decision-making logic are: safety first, intelligent collaboration, and dynamic balance. Specifically, these include: Highest priority: Intrusion prevention. Once the intrusion status information indicates that an intrusion has been detected (a person or vehicle entering the preset security protection zone), regardless of the flame status, a "stop laser + start alarm" command is immediately generated. At the same time, the current flame status and patrol position are recorded.
[0038] After the intrusion is resolved, the system will resume or adjust its operation based on the flame status: When the intrusion signal disappears, the system will not immediately ignite automatically. The task management module will make a judgment based on the previously recorded flame status and the latest flame status information. If the flame is burning stably (area and height meet the requirements), the system will remain in a stopped state; if the flame is extinguished or unstable, the system will wait for manual confirmation via a tablet terminal or resume ignition in a cautious mode (lower power) after a set delay (safety confirmation).
[0039] When the intrusion status information indicates no intrusion, intelligent ignition control is executed: In the no-intrusion state, the system mainly controls the laser and pan / tilt unit based on the flame status information. If a flame exists and its area and height both reach the preset safe combustion threshold (indicating successful ignition and stable combustion), the integrated control command is "stop laser emission (or reduce laser power to an extremely low maintenance level) + continue gimbal cruise." This achieves energy saving and avoids unnecessary laser irradiation.
[0040] If the flame is absent, or the flame area and height are below the preset safe combustion threshold (indicating no ignition or the flame is about to extinguish), the integrated control command is "to fire a laser at the current target point with a preset ignition power + control the gimbal to cruise along a preset cruise path." The cruise logic ensures that the laser irradiates multiple target points one by one, improving the success rate of gas ignition.
[0041] Dynamic environmental adaptation: The task management module dynamically adjusts the detection confidence thresholds of the flame recognition sub-model and the intrusion detection sub-model based on the environmental visibility level. When the environmental visibility level is low (foggy weather), the detection confidence threshold of the flame recognition sub-model is lowered to improve the flame detection rate and prevent missed flame detections. Conversely, the detection confidence threshold of the intrusion detection sub-model is appropriately increased to prevent false alarms caused by fog or changes in light conditions. This dynamic adjustment ensures balanced system performance under different environments.
[0042] High-risk event handling: When a high-risk alarm signal is received from the feature fusion and correlation analysis module, the task management module immediately generates an instruction to "stop laser + enhance alarm (such as double the volume, flashing red warning light)" and pushes the details of this high-risk event to the tablet terminal and remote monitoring platform to request rapid manual intervention.
[0043] The laser control module is responsible for converting the comprehensive control commands generated by the task management module into specific hardware control signals. Specifically, this includes: Laser power supply and modulation control: Based on the "emit / stop / power adjustment" command, control the power switch and output current of the laser generator, thereby controlling the switching and power of the laser.
[0044] Gimbal Motion Control: Based on the "aiming coordinates" or "cruise path" commands, the gimbal motors are driven via serial port to achieve horizontal and pitch movements and position locking. It is responsible for managing the execution of preset positions and cruise sequences.
[0045] Lens focal length control: Based on the ranging signal provided by the ranging unit and the instructions from the task management module, a control signal is sent to the motorized zoom focusing lens to automatically adjust the focal length so that the laser spot is focused to the minimum at the target point.
[0046] The communication module is responsible for data transmission, connecting to the tablet terminal for bidirectional transmission of video, control commands, and status information. The communication module can also communicate with a remote monitoring platform to enable remote monitoring and data reporting.
[0047] The historical data storage module uses a large-capacity memory to store analyzed flame status information, intrusion status information, system operating parameters, and real-time video images for long-term use in accident tracing and model optimization.
[0048] The backup power management module monitors the status of the main power supply and automatically and seamlessly switches to the internal backup power supply in the event of a power outage, immediately notifying the task management module. Upon receiving the power outage notification, the task management module executes a safe shutdown procedure: stopping the laser, shutting down non-essential equipment, and sending a power outage alarm and final status to the remote platform via the communication module.
[0049] The tablet terminal sends user commands to the control host, receives and displays real-time video images, flame status information, intrusion status information, system operating parameters, and alarm information from the control host. Specifically, this includes: Real-time video image display area: Displays real-time video images from the visible light video acquisition unit. Flame status information and intrusion status information can be overlaid on the video: detected human bodies / vehicles are marked with red boxes, detected flames are marked with yellow outlines or color blocks, and the estimated flame height is displayed.
[0050] Security Zone Calibration Interface: Allows operators to draw polygons of arbitrary shapes on real-time video images as the boundaries of preset security protection areas, and sends the generated boundary coordinate data to the control host for storage, which is used for the detection range constraints of the intrusion detection sub-model.
[0051] Target setting and cruise path planning interface: Allows operators to mark multiple target points on real-time video images. The cruise sequence of these target points and the dwell time (laser irradiation duration) at each point can be set. This cruise planning data is sent to the task management module of the control host.
[0052] Control and Status Display Area: Provides manual / automatic mode switching buttons, forced ignition / stop buttons, emergency stop buttons, etc. Real-time display of system status: current target point, laser power, flame status, intrusion status, battery level, etc. When the system alarms, this area will highlight the alarm information and provide a reset option.
[0053] The specific implementation method of this embodiment is as follows: Workflow in automatic mode: The system starts up, the control host initializes, and the AI algorithm processing module loads the multi-task visual analysis model. The task management module reads the security protection zone and cruise route planning data set on the tablet terminal.
[0054] The gimbal moves along the planned path to the first target point B1. The ranging unit measures the distance, and the laser control module automatically focuses.
[0055] The AI algorithm processing module continuously analyzes the video captured by the visible light video acquisition unit. The flame recognition sub-model determines whether there is a flame at the first target point B1. If not, the task management module generates a "fire laser at ignition power" command. The laser control module executes the command, and the laser is directed towards the first target point B1.
[0056] Meanwhile, the intrusion detection sub-model continuously monitors the preset security protection area.
[0057] Scenario A: The flame is ignited, and its area and height rapidly increase, reaching the safety threshold. The task management module determines that ignition was successful and generates a "stop laser" command, shutting down the laser. The gimbal continues to move to the next target point B2.
[0058] Scenario B: During the ignition process of the first target point B1, the intrusion detection sub-model detects a pedestrian entering the security protection area. The task management module immediately generates a "stop laser + start alarm" command, the laser is cut off, and the alarm unit emits an audible and visual warning. The intrusion alarm interface pops up on the tablet terminal. After the pedestrian leaves, the task management module, based on the flame status (the flame is now extinguished) and preset logic, decides whether to wait for confirmation or attempt to reignite the first target point B1 in cautious mode after a delay.
[0059] Scenario C: The weather turns into dense fog, and the environmental perception submodule outputs poor visibility. The task management module dynamically lowers the flame recognition threshold and raises the intrusion detection threshold. The unified multi-task visual analysis model recognizes flames under more lenient conditions while being more cautious in determining intrusions, balancing security and efficiency.
[0060] The system cruises along the planned path in a loop, continuously executing the above steps until it receives a stop command or switches to manual mode.
[0061] Workflow in manual mode: Operators manually control the gimbal for aiming in real time via the tablet terminal's manual control interface. Clicking the "fire" button on the screen generates a launch command from the task management module. Meanwhile, the intrusion detection sub-model continues to run in the background with the highest priority; upon detecting an intrusion, it automatically cuts off the laser to ensure safety.
[0062] This invention, by constructing a unified platform based on artificial intelligence visual analysis, achieves deep integration and collaboration between flame recognition and security intrusion detection at the algorithm, data, and decision levels, effectively solving problems such as functional fragmentation, poor environmental adaptability, and high maintenance costs in traditional technical solutions. The system of this invention can not only reliably and intelligently complete multi-target gas ignition tasks, but also proactively and accurately ensure on-site safety, representing an advanced direction for the intelligent and integrated development of laser ignition and safety protection technologies.
[0063] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.
Claims
1. A multi-target intelligent cruise laser gas ignition and safety protection system, characterized in that, The system includes an intelligent launch cabin, a control host, and a tablet terminal. The intelligent launch cabin includes a visible light video acquisition unit that acquires real-time video images. The control host communicates with both the intelligent launch cabin and the tablet terminal. The control host includes an AI algorithm processing module that receives and analyzes real-time video images. Internally, the AI algorithm processing module runs a unified multi-task visual analysis model and outputs flame status information and intrusion status information. The control host also includes a task management module that acquires flame status information and intrusion status information and generates comprehensive control commands based on preset collaborative decision-making logic. The intelligent launch cabin also includes a laser emission unit and a gimbal. The control host also includes a laser control module. Based on the comprehensive control commands, the laser control module controls the laser emission unit to emit lasers and controls the gimbal to perform multi-target cruise.
2. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 1, characterized in that: The multi-task visual analysis model includes a parallel flame recognition sub-model and an intrusion detection sub-model. The flame recognition sub-model and the intrusion detection sub-model share a unified feature coding backbone network. The feature coding backbone network adopts a three-dimensional convolutional neural network architecture to extract spatiotemporal features from real-time video images and output a shared feature map. The flame recognition sub-model is based on the shared feature map and is processed through a flame-specific head branch network to output flame status information. The intrusion detection sub-model is based on the same shared feature map and is processed by a specific head branch network for intrusion detection to output intrusion status information.
3. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 2, characterized in that: The flame-specific head branch network includes a flame classification head, a flame segmentation head, and a flame regression head. The flame recognition sub-model determines the presence of flames through the flame classification head, outputs a pixel-level mask of the flames through the flame segmentation head, estimates the flame height through the flame regression head, and finally outputs flame status information. The intrusion detection-specific head branch network includes a target detection network that detects and identifies human bodies and vehicles within a preset security protection area and finally outputs intrusion status information.
4. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 3, characterized in that: The collaborative decision-making logic specifically includes: when the intrusion status information indicates that an intrusion has been detected, the integrated control command is to immediately stop laser emission and trigger an alarm; when the intrusion status information indicates that there is no intrusion, the integrated control command is determined based on the flame status information: if a flame exists and the flame area and height both reach the preset safe combustion threshold, the integrated control command is to stop laser emission or reduce the laser power to the maintenance mode; if a flame does not exist, or the flame area and height do not reach the preset safe combustion threshold, the integrated control command is to control the laser emission unit to emit laser at a preset ignition power and control the pan-tilt unit to cruise along a preset cruise path.
5. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 3, characterized in that: The AI algorithm processing module also includes an environmental perception submodule; the environmental perception submodule assesses the environmental visibility based on the statistical features of real-time video images and outputs the environmental visibility level; the task management module dynamically adjusts the detection confidence thresholds of the flame recognition submodel and the intrusion detection submodel in combination with the environmental visibility level.
6. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 5, characterized in that: The AI algorithm processing module also includes a feature fusion and correlation analysis module. The feature fusion and correlation analysis module receives flame status information and intrusion status information, performs spatial and temporal correlation analysis on the flame status information and intrusion status information, and outputs the correlation analysis results. If it is determined that the burning position of the flame overlaps with or is close to the position of the intrusion target at a preset distance, a high-risk alarm signal is output to the task management module. The task management module generates a comprehensive control command to immediately stop laser emission and enhance the alarm based on the high-risk alarm signal.
7. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 1, characterized in that: The intelligent launch cabin also includes a ranging unit, which measures the distance from the current target point to the intelligent launch cabin and outputs a ranging signal; the laser control module adjusts the focusing parameters of the laser emission unit according to the ranging signal.
8. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 6, characterized in that: The tablet terminal sends user commands to the control host and displays real-time video images, flame status information, and intrusion status information. The tablet terminal has a security zone calibration interface for drawing the boundaries of the security protection area. The tablet terminal also has a target point setting and cruise path planning interface for calibrating multiple target points and setting cruise paths.
9. The multi-target intelligent cruise laser gas ignition and safety protection system according to claim 8, characterized in that: The control host also includes a historical data storage module and a model self-learning and optimization module; the historical data storage module is used to store real-time video images, flame status information and intrusion status information; the model self-learning and optimization module periodically obtains real-time video images, flame status information and intrusion status information from the historical data storage module to perform incremental learning on the unified multi-task visual analysis model.