An industrial fire early warning device based on image recognition

Abstract

The application relates to the technical field of industrial safety monitoring, and discloses an industrial fire early warning device based on image recognition, which comprises a multi-dimensional physical quantity cooperative sensing module, which is used for collecting visible light images containing different polarization angles, key frame images, infrared temperature field images, displacement pulse sequences and load data; and a space-time reference unified alignment module, which is used for performing space mapping on the visible light images and the infrared temperature field images and generating material segment identification according to the displacement pulse sequences and the load data. The application adopts near-infrared frequency flash differential calculation and environmental interference index dynamic correction technology, achieves the technical effect of quantifying the backscattering intensity of dust and water vapor, realizes self-adaptive adjustment of a judgment threshold with the damaged degree of visibility, and solves the problems of high false alarm rate and increased line stop cost of traditional equipment caused by the inability to effectively distinguish environmental suspended particles from real smoke.

Description

Technical Field

[0001] This invention relates to the field of industrial safety monitoring technology, specifically to an industrial fire early warning device, electronic device, and storage medium based on image recognition. Background Technology

[0002] With the acceleration of urbanization and the expansion of industrial production, the solid waste treatment, recycling, and large-scale logistics sorting industries are experiencing a period of rapid development. In large-scale industrial settings, belt conveyors, as the core logistics transport carriers, play a crucial role in the continuous and efficient transport of bulk materials to various processing stages. However, due to the extreme complexity of the transported materials and the unavoidable mechanical friction, heat accumulation and spontaneous combustion, and electrical faults during transport, conveyor lines have become high-risk areas for fire accidents. Once a fire breaks out on the material on the conveyor belt, under high-speed operating conditions, the fire can quickly spread to upstream and downstream workshops, easily igniting surrounding stacks and equipment, causing huge economic losses and even casualties.

[0003] To mitigate such risks, existing industrial fire protection systems typically employ traditional point-type smoke / heat detectors, linear heat-sensing cables, or conventional video surveillance systems. However, in harsh environments such as solid waste treatment plants, waste-to-energy plants, and recycling workshops, conventional detection methods face significant challenges. Traditional smoke detectors are limited by the airflow diffusion in large spaces, resulting in extremely slow response times; linear heat-sensing cables require contact with flames or reach specific high temperatures to trigger, often missing the optimal time for fire suppression; and while conventional video surveillance incorporates basic image recognition technology to attempt to capture flame or smoke characteristics through cameras, its accuracy, anti-interference capabilities, and integration with the response system all exhibit significant deficiencies in practical industrial applications, making it difficult to meet the requirements for all-weather, high-reliability early warning systems.

[0004] Specifically, existing fire early warning technologies face the following technical bottlenecks and shortcomings in practical applications:

[0005] Solid waste treatment sites often contain high concentrations of dust, water vapor, and vehicle exhaust, which are physically very similar to the smoke at the beginning of a fire. This makes it difficult for traditional monitoring equipment to distinguish between environmental suspended particles and real smoke, leading to frequent false alarms, increasing the company's downtime investigation costs, and reducing the actual credibility of the early warning system.

[0006] The frequent direct sunlight, vehicle headlights, and specular reflections from metal materials in the sorting workshop create bright areas on the monitoring screen. Existing image recognition technology easily misidentifies dynamically flickering light spots as firelight. Because it cannot effectively eliminate light fluctuations caused by non-fire sources, the recognition accuracy is severely affected in complex photoelectric environments.

[0007] Due to the massive size and overlapping nature of waste piles, the initial source of fire often lies hidden inside the pile or beneath the materials, manifesting as a slow rise in temperature or a weak smoldering. Because of the lack of obvious open flame characteristics, technologies relying solely on visible light for identification cannot penetrate the surface materials to perceive the internal heat evolution, making it difficult to detect abnormal signs in the early stages of a fire.

[0008] During the continuous operation of a belt conveyor line, the position of the material is constantly changing. When a fire is detected, the fire source will be rapidly displaced by the conveying equipment, causing a spatial deviation between the alarm location and the actual fire point. In addition, due to the lack of real-time correlation evidence between abnormal signs and material batches, it is difficult for subsequent personnel to quickly locate and verify the fire point location. Summary of the Invention

[0009] In view of the shortcomings of the prior art, the present invention provides an industrial fire early warning device, electronic device and storage medium based on image recognition, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] In a first aspect, the present invention provides an industrial fire early warning device based on image recognition, comprising:

[0012] The multi-dimensional physical quantity collaborative sensing module is used to acquire visible light images, keyframe images, infrared temperature field images, displacement pulse sequences, and load data containing different polarization angles.

[0013] The spatiotemporal reference unification and alignment module is used to perform spatial mapping on the visible light image and the infrared temperature field image, and to generate material fragment identifiers based on the displacement pulse sequence and load data.

[0014] An environmental baseline adaptive adjustment module is used to generate an interference index based on the brightness difference of the fill light switch frame and to generate a dynamic hot spot discrimination threshold by combining it with the thermal imaging background temperature baseline.

[0015] The heterogeneous feature fusion judgment module is used to generate a reflection suppression mask based on polarization difference. In the non-reflective effective area, it combines the thermal imaging background temperature baseline, dynamic hot spot discrimination threshold and the highest temperature sequence to generate early warning events. The early warning events include suspected fire point areas, event time and highest temperature sequence.

[0016] The early warning package generation and closed-loop processing module is used to respond to early warning events. It generates early warning coordinates based on the early warning event and spatial mapping results, and encapsulates material fragment identifiers, early warning coordinates, maximum temperature sequence and key frame images to form an early warning package. It performs hash digest solidification on the early warning package and issues safe operation instructions.

[0017] Preferably, the material segment identifier is generated by splicing together the conveyor line number, the start and end values ​​of the warning time window, the displacement range of the material in the conveying direction, and the weight range of the material.

[0018] Preferably, when the early warning package generation and closed-loop processing module performs hash digest solidification, it uses a hash algorithm to calculate a digest value for the complete content of the early warning package and writes the digest value into a local unwritable log.

[0019] Preferably, the early warning package generation and closed-loop processing module is used for:

[0020] Sending control pulses to the field controller to slow down the conveyor line and execute a shutdown action after a preset delay, or activating the on-site directional spraying device and atomizing dust suppression device to identify the material segments. The corresponding areas were physically treated.

[0021] Preferably, the multidimensional physical quantity collaborative sensing module includes:

[0022] The timing-triggered and polarization imaging unit is used to generate a global synchronization clock signal and a frame index. According to the frame index The drive-controlled rotating mechanism switches the linear polarizer to the target polarization angle. And when the polarization angle is stable, the visible light imaging component is triggered to expose to obtain a resolution of Visible light image matrix ,in, and Pixel coordinates, target polarization angle The control logic satisfies the formula:

[0023] ,

[0024] in, The first polarization angle reference value, This is the angular difference between the first polarization angle and the second polarization angle;

[0025] Near-infrared strobe and temperature field capture unit, referencing the frame index Used to control the radiation intensity of near-infrared supplementary lighting components And simultaneously trigger the thermal imaging component to acquire the infrared temperature field image matrix. ,in, and For thermal imaging pixel coordinates, radiation intensity The control logic satisfies the formula:

[0026] ,

[0027] in, This refers to the high-level radiated power when the supplementary light is turned on. The ambient light noise floor power when the supplementary lighting is off is given by this formula, which causes the supplementary lighting state to form an alternating bright and dark strobe sequence between odd and even frames, and the infrared temperature field image matrix... Each pixel value in the image represents the radiation temperature of the target object.

[0028] The motion increment and load synchronization unit references the global synchronization clock signal for frame indexing. The cumulative pulse value of the rotary encoder is read at the corresponding exposure time. and the instantaneous weight value collected by the weighing sensor. Instantaneous weight value As load data, the physical displacement increment of the transport line in the current sampling period is calculated based on the pulse difference between the current frame and the previous frame. Physical displacement increment The calculation satisfies the formula:

[0029] ,

[0030] in, The physical diameter of the drive roller of the conveyor line. This represents the total number of pulses output by the rotary encoder in one revolution. This represents the cumulative number of pulses at the current frame time. The physical displacement increment is the cumulative pulse count of the previous frame. With the instantaneous weight value Packed as a matrix corresponding to visible light images The motion attribute metadata.

[0031] Preferably, the spatiotemporal reference unification and alignment module includes:

[0032] Heterogeneous image pixel-level registration unit, used to establish pixel coordinates of visible light images. and infrared temperature field image pixel coordinates The homography mapping relationship between them is determined, and a registered temperature field matrix with the same resolution as the visible light image is generated. The registration logic satisfies the following formula: ,

[0033] in, for The homography transformation matrix describes the rotation, translation, and perspective projection relationships of the thermal imaging component relative to the visible light imaging component. The heterogeneous image pixel-level registration unit applies the homography mapping relationship to the infrared temperature field image matrix. Perform backsampling interpolation to obtain the pixel coordinates corresponding to the visible light image. The radiation temperature value is used to achieve pixel-level alignment between visible light images and infrared temperature field images in the spatial dimension;

[0034] Global displacement integral and coordinate mapping unit, using physical displacement increments obtained by converting displacement pulse sequences. Used to calculate the current cumulative travel position of the conveyor line. Furthermore, a warning coordinate mapping benchmark is established between image pixel coordinates and conveyor line physical coordinates, and the cumulative travel position is calculated. The calculation satisfies the formula: ,

[0035] in, The physical position corresponding to the initial encoder reading. From the start time to the current frame Cumulative variables; warning coordinates Satisfying the formula:

[0036] ,

[0037] In the formula, These are the early warning coordinates in the physical coordinate system of the conveyor line. The coordinates are the longitudinal coordinates along the conveying direction. The horizontal coordinate is perpendicular to the conveying direction. and represents the physical resolution coefficients of image pixels in the vertical and horizontal directions;

[0038] Material fragment identifier dynamic generation unit, referencing cumulative travel position Image pixel physical resolution coefficients, key frame sequences, and load data acquired by the weighing sensor. Used to generate the material fragment identifier corresponding to the current keyframe. Material fragment identification The generation logic satisfies the string concatenation formula: ,

[0039] in, For string concatenation operators, This is the fixed number for the current conveyor line. and The timestamps are the start and end values ​​corresponding to the current keyframe. The width of the visible light image matrix is ​​in pixels. The physical length of the material covered by the current keyframe. The material fragment identifier is used as a unique index key value for the encapsulation and traceability of subsequent warning packages, based on the weight level coding according to the load data.

[0040] Preferably, the environmental baseline adaptive adjustment module includes:

[0041] The flicker differential and interference index calculation unit is used in a preset dark field reference area. The average brightness values ​​of the frames with and without supplemental lighting are extracted, and the interference index is calculated. The computational logic satisfies the formula: ,in, For the strobe cycle index, For the first Even-numbered frames within a period In the dark field reference area The average grayscale value of the pixels within the range. For the first Odd-numbered frames within a period In the dark field reference area The average pixel grayscale value within the range is used to prevent the sensor noise floor constant from having a denominator of zero. The interference sensitivity gain coefficient is a formula that quantifies the backscattering intensity of near-infrared active illumination by environmental suspended particles. The interference index... Used to characterize the degree of visibility impairment in the current monitoring space;

[0042] The thermal imaging background baseline iteration unit is used to update the statistical thermal imaging background temperature baseline using a weighted moving average method. The update logic satisfies the formula:

[0043] ,

[0044] in, The background learning rate coefficient. This is the background temperature baseline matrix of the previous frame. When updating, this unit iterates over pixels that are not marked as suspected high-temperature targets to establish baseline data that reflects the normal heat distribution of the current environment.

[0045] The adaptive threshold generation unit is used to generate dynamic hotspot discrimination thresholds, and the generation logic satisfies the formula: ,

[0046] in, The preset static temperature rise alarm threshold, The threshold correction coefficient is positively correlated. As a safety margin constant, this formula automatically raises the alarm threshold when the interference index increases to suppress false alarms of thermal image noise caused by dust-scattered infrared radiation. The hotspot discrimination threshold... and thermal imaging background temperature baseline It will be transmitted as a judgment parameter to the heterogeneous feature fusion judgment module.

[0047] Preferably, the heterogeneous feature fusion determination module includes:

[0048] The polarization difference and reflection suppression unit is used to calculate the grayscale difference between the current keyframe and the previous keyframe at the same pixel coordinates, and generate a polarization difference map. According to the polarization difference diagram Generate a reflective suppression mask Reflection suppression mask The generation logic satisfies the formula:

[0049] ,

[0050] in, Preset reflection detection threshold; reflection suppression mask This is used to mark metallic reflective areas and high-brightness reflective areas as reflective interference areas, and to remove reflective interference areas from the candidate judgment area to obtain non-reflective effective areas;

[0051] A cross-modal hotspot verification unit is used in reflectivity suppression masks. Within the corresponding non-reflective effective area, combined with the thermal imaging background temperature baseline and dynamic hot spot discrimination threshold Perform temperature anomaly detection and generate a candidate set of potential fire points. Suspected Fire Point Candidate Set Conditions met: ,

[0052] The cross-modal hotspot verification unit is used to filter the set of pixels that simultaneously meet the non-reflective characteristics and have a temperature rise exceeding the dynamic hotspot discrimination threshold, and to extract a candidate set of suspected fire points. The highest temperature values ​​of the inner pixels are arranged into a highest temperature sequence according to the time order of the keyframe sequence.

[0053] The spatiotemporal trajectory consistency determination unit is used to calculate the candidate set of suspected fire points. The deviation of the centroid of the candidate region in the motion trajectory of the keyframe sequence Movement trajectory deviation The calculation satisfies the formula: ,

[0054] in, This refers to the centroid position of the candidate region in the current keyframe on the vertical image coordinate system. For the front The centroid position of the candidate region in the frame on the vertical image coordinate system. The vertical physical resolution coefficient is used when the candidate region is located within a non-reflective effective region, the temperature rise exceeds the dynamic hot spot discrimination threshold, and the highest temperature sequence meets the preset persistence or heating trend conditions. Less than the preset trajectory tolerance At that time, the heterogeneous feature fusion judgment module outputs an early warning event.

[0055] Preferably, the early warning package generation and closed-loop processing module includes:

[0056] The structured encapsulation and hash digest solidification unit is used to construct an early warning package containing material fragment identifiers, early warning coordinates, maximum temperature sequences, and keyframe images. Early warning package The construction logic satisfies the vector combination formula: ,in, For material fragment identification, As early warning coordinates, This is the highest temperature sequence. Keyframe images marked with a reflection suppression mask;

[0057] The structured encapsulation and hash digest persistence unit is further used to process the warning packet using a secure hash algorithm. Complete content calculation digital summary Digital summary The generation satisfies the formula:

[0058] ,in, for Bit-safe hash function, For XOR operation, Preset timestamp salt value; digital digest Used for early warning packages Perform anti-tampering verification and event tracing;

[0059] Risk quantification and strategy mapping unit, referencing early warning package The highest temperature sequence in Used to calculate fire hazard rating According to the fire hazard rating Generate safe operation instructions Fire risk rating Satisfies the weighted formula:

[0060] ,in, This is the absolute temperature rise weighting coefficient. This is the static temperature rise alarm threshold. The heating rate is the weighting factor. The first derivative of the highest temperature sequence;

[0061] Safe operation instructions Satisfies a piecewise function:

[0062] ,in, and The thresholds for low-risk and high-risk assessment are defined as follows: It is a directional spray mode. Emergency shutdown mode;

[0063] The dynamic delay compensation and execution unit is used to calculate the warning coordinates. The average running speed of the conveyor belt obtained by converting displacement pulse sequence The fixed physical coordinates of the downstream directional spraying device relative to the starting point of the conveyor line. and the mechanical response time of the solenoid valve Calculate the lag compensation time for the start-up of the directional sprinkler system. It also issues a safe operation command to the field controller, with a delay in compensation time. The calculation satisfies the formula: ,in, For early warning coordinates Longitudinal coordinates along the conveying direction;

[0064] The dynamic delay compensation and execution unit is used to ensure that the spray medium covers the target area corresponding to the material fragment identification.

[0065] The multi-dimensional physical quantity collaborative sensing module integrates visible light polarization imaging, thermal imaging, near-infrared strobe lighting, and rotary encoder devices to simultaneously acquire optical polarization texture, thermal radiation distribution, environmental medium concentration, and motion displacement information of the target area. This multi-dimensional physical quantity acquisition method breaks through the perception limitations of a single visual sensor and provides comprehensive and time-series-aligned underlying data support for eliminating reflective interference, penetrating dust obstructions, and tracking dynamic targets in complex industrial environments.

[0066] The spatiotemporal reference unified alignment module uses homography transformation and motion integral algorithms to establish a rigid mapping relationship between the pixel coordinates of heterogeneous images and the physical absolute coordinates of the conveyor line. This transforms discrete visual signals into spatial information with physical scale. At the same time, the generated material fragment identifiers containing spatiotemporal attributes can achieve digital identity binding for each batch of materials, ensuring continuous tracking accuracy of the fire source location under high-speed conveyor belt operation, and providing a precise spatiotemporal reference for fixed-point spraying and historical source tracing.

[0067] The environmental baseline adaptive adjustment module uses near-infrared stroboscopic differential technology to quantify the interference of on-site dust and steam on imaging in real time, and dynamically adjusts the sensitivity threshold for fire detection accordingly. At the same time, it continuously updates the thermal imaging background baseline through statistical methods. This adaptive mechanism can effectively offset the noise false alarms introduced by visibility fluctuations and changes in ambient temperature at the solid waste treatment site, ensuring that the system always maintains a high signal-to-noise ratio capture capability and operational stability for real fires under harsh working conditions.

[0068] The heterogeneous feature fusion judgment module physically filters out specular reflection spots from metals and bright materials through polarization difference operations, locks in the real high-temperature area by combining cross-modal temperature verification logic, and eliminates static heat source interference in the background by using a spatiotemporal trajectory consistency judgment algorithm. This comprehensive judgment strategy that integrates optical polarization characteristics, infrared thermal radiation characteristics and kinematic laws completely solves the problem of false alarms caused by light and shadow fluctuations and false heat sources in complex photoelectric environments, and significantly improves the accuracy of identifying hidden smoldering fire sources.

[0069] The early warning package generation and closed-loop handling module uses hash digest technology to encrypt and solidify early warning data containing spatiotemporal location and image evidence, constructing an immutable chain of evidence for accident tracing. At the same time, based on a dynamic time delay compensation algorithm, it sends precise timing pulses to the execution mechanism to ensure that the spraying medium can accurately cover the fire point area that has moved to the handling station without stopping the conveyor belt or during deceleration. This achieves millisecond-level automatic closed-loop control from hazard discovery to physical handling, minimizing the risk of fire spread and asset loss.

[0070] In a second aspect, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method of executing the various modules in the first aspect.

[0071] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for executing the various modules in the first aspect.

[0072] The present invention has the following beneficial effects:

[0073] 1. This invention employs near-infrared stroboscopic differential calculation and dynamic correction technology for environmental interference index to achieve the technical effect of quantifying the backscattering intensity of dust and water vapor, and realizes the adaptive adjustment of the judgment threshold according to the degree of visibility impairment. This solves the shortcomings of traditional equipment, which cannot effectively distinguish between environmental suspended particles and real smoke, resulting in high false alarm rates and increased downtime costs.

[0074] 2. This invention employs multi-angle polarization imaging and differential suppression technology to achieve the technical effect of physically filtering out high-brightness noise by utilizing polarization characteristics. It enables precise removal and masking of false light spots caused by metal reflections and vehicle headlight illumination, thus solving the problem of reduced accuracy in fire light recognition due to dynamic light and shadow fluctuations in complex optoelectronic environments.

[0075] 3. This invention employs pixel-level spatiotemporal registration and cross-modal verification technology of infrared temperature field and visible light image to achieve the technical effect of visual and thermal data fusion perception, enabling penetrating capture of weak smoldering and early heating trends inside the reactor body, and solving the shortcomings of a single visible light scheme that cannot detect fire signs inside the shielded area in time during the stage without open flame.

[0076] 4. This invention adopts conveyor belt displacement pulse integration and dynamic coordinate mapping technology to achieve the technical effect of converting image pixels into physical absolute coordinates in real time, realizing full-process dynamic tracking of fire point location and accurate timing compensation of handling instructions, solving the shortcomings of alarm positioning deviation and difficulty in verification and handling caused by rapid displacement of fire source by material during continuous conveying. Attached Figure Description

[0077] Figure 1 This is a framework diagram of an industrial fire early warning device based on image recognition according to the present invention. Detailed Implementation

[0078] To enable those skilled in the art to understand the present invention, 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 some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0079] The present invention will now be described in detail with reference to the accompanying drawings:

[0080] Example 1: Verification of high reflectivity and dust resistance in a municipal solid waste recycling sorting line

[0081] Please see the appendix Figure 1 This embodiment is applied to the sorting workshop of a large-scale municipal solid waste recycling center. The conveyor belt in this workshop carries a large quantity of mixed metal cans, glass bottles, plastic packaging, and stained paper. Due to the operation of the robotic sorting arms, there is a high concentration of dust, and the workshop ceiling has skylights, causing sunlight and ambient lighting to create strong speckled reflections on the surfaces of the metal and glass waste.

[0082] After startup, the multi-dimensional physical quantity collaborative sensing module starts working; the near-infrared supplementary lighting component performs high-frequency strobe according to the preset timing sequence, and the thermal imaging component synchronously collects temperature field data. At this time, the environmental baseline adaptive adjustment module identifies the existence of continuous dust interference in the current environment by comparing the brightness difference in the dark field area between the supplementary lighting on frame and the off frame, and automatically calculates the interference index.

[0083] Based on the above interference index, the hot spot discrimination threshold is automatically increased, and the current background temperature baseline is established to prevent noise caused by dust particles scattering infrared light from being misjudged as fire points, effectively avoiding false alarms caused by environmental media.

[0084] During transport, a pile of compressed aluminum cans produced bright spots under strong light, appearing as if they were open flames in ordinary visible light. At this point, the timing trigger unit of the early fire warning system drove the electrically controlled corner mechanism to rapidly switch the angle of the linear polarizer; the visible light imaging component continuously acquired images under different polarization states. The heterogeneous feature fusion judgment module performed differential calculations on the acquired images under different polarization states and found that the bright area showed significant grayscale changes under different polarization angles, consistent with the physical characteristics of metal reflection. Therefore, a reflection suppression mask was generated, marking the area as "non-fire source reflection interference" and directly removing it.

[0085] Meanwhile, a piece of oil-soaked cotton yarn mixed in with the conveyor line smoldered due to heat accumulation, showing a slight burning without obvious open flame but with an abnormal temperature. Because cotton yarn is a diffuse reflective material, its image characteristics remained stable at different polarization angles and were not masked. Simultaneously, the thermal imaging component detected that the temperature in this area was significantly higher than the background baseline.

[0086] The heterogeneous feature fusion judgment module combines the dual evidence of "non-reflective features" and "abnormal infrared temperature rise" to determine that the location is a real hidden danger and outputs an early warning event.

[0087] This embodiment demonstrates that the invention can effectively utilize polarized optical properties to distinguish between real firelight and metal / glass reflections, and quantifies and cancels dust interference through near-infrared stroboscopic technology. In complex lighting and poor air quality environments, it successfully filters out false alarms, accurately pinpoints early smoldering fire sources in diffuse reflective materials, and verifies its excellent anti-interference performance.

[0088] Example 2: Verification of high-speed dynamic tracking and closed-loop treatment applied to industrial solid waste crushing processes

[0089] Please see the appendix Figure 1 This embodiment is applied to the crushing and feeding section of an industrial hazardous waste treatment plant. The conveyor belt runs at a relatively high speed, feeding hazardous waste such as waste lithium batteries and aerosol cans into the crusher. These materials are highly susceptible to internal thermal runaway, and the source of ignition is often hidden inside or at the bottom of the material pile. Because the material moves quickly by the conveyor belt, it is not easily detected by surface observation and is difficult to locate.

[0090] During operation, the rotary encoder collects the rotation pulses of the conveyor rollers in real time, and the continuous weighing sensor records the weight of the material. The spatiotemporal reference unified alignment module uses a homography transformation matrix to precisely align the thermal imaging pixels with the visible light pixels, and maps the pixel coordinates to the physical absolute coordinates of the conveyor line;

[0091] A batch of used batteries on the conveyor belt experienced a short circuit and overheating, with the heat gradually being conducted to the surface. No open flame was observed, but the thermal imaging unit detected an abnormal temperature rise. The cross-modal hotspot verification unit extracted the highest temperature sequence of the suspected area. Subsequently, the spatiotemporal trajectory consistency determination unit began working: calculating the motion trajectory of the hotspot in multiple consecutive frames of images and comparing it with the theoretical physical displacement of the conveyor belt calculated using encoder pulses. The comparison results showed that the hotspot's movement speed and direction were completely consistent with the conveyor belt, eliminating interference from stationary heat sources in the background and confirming it as a genuine potential hazard caused by material movement.

[0092] The early warning package generation module immediately generates a unique material fragment identifier ID, encapsulates key frame images and temperature rise data, performs hash digest solidification, and stores it in an unchangeable log to ensure traceability of evidence.

[0093] Given that the conveyor belt is running at high speed and the location of the fire point is constantly changing, the dynamic delay compensation and execution unit calculates the precise lag time for the spray to start based on the current physical longitudinal coordinates of the fire point, the fixed position of the downstream directional spray device, and the real-time speed of the conveyor belt.

[0094] When the abnormal material moves from the conveyor belt to directly beneath the spray system, the countdown ends, and the on-site controller precisely triggers the solenoid valve to open, accurately covering the fire area with the spray medium. Simultaneously, the system sends a command to slow down and eventually stop the conveyor line, preventing flammable material from entering the downstream crusher and causing an explosion.

[0095] This embodiment demonstrates that the invention solves the problem of difficult dynamic target localization through deep fusion of physical displacement integrals and image features. The system can detect hidden internal heat sources and achieve closed-loop control of "detection and locking, arrival and disposal" during high-speed material movement, verifying its extremely high feasibility in preventing fire spread and precise coordinated response.

[0096] Those skilled in the art should understand that the embodiments of the present invention can be implemented using a pure hardware architecture, a pure software architecture, or an integrated hardware and software architecture. The present invention can be prepared as a computer program product, which can be stored in various non-volatile computer-readable storage media, including but not limited to solid-state drives, flash memory chips, mobile storage devices, optical discs, cloud storage servers, and other standardized storage media, and is not limited to traditional storage media.

[0097] Embodiments of the present invention have been presented and described. It will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An industrial fire early warning device based on image recognition, characterized in that, include: The multi-dimensional physical quantity collaborative sensing module is used to acquire visible light images, keyframe images, infrared temperature field images, displacement pulse sequences, and load data containing different polarization angles. The spatiotemporal reference unification and alignment module is used to perform spatial mapping on the visible light image and the infrared temperature field image, and to generate material fragment identifiers based on the displacement pulse sequence and load data. An environmental baseline adaptive adjustment module is used to generate an interference index based on the brightness difference of the fill light switch frame and to generate a dynamic hot spot discrimination threshold by combining it with the thermal imaging background temperature baseline. The heterogeneous feature fusion judgment module is used to generate a reflection suppression mask based on polarization difference. In the non-reflective effective area, it combines the thermal imaging background temperature baseline, dynamic hot spot discrimination threshold and the highest temperature sequence to generate early warning events. The early warning events include suspected fire point areas, event time and highest temperature sequence. The early warning package generation and closed-loop processing module is used to respond to early warning events. It generates early warning coordinates based on the early warning event and spatial mapping results, and encapsulates material fragment identifiers, early warning coordinates, maximum temperature sequence and key frame images to form an early warning package. It performs hash digest solidification on the early warning package and issues safe operation instructions.

2. The industrial fire early warning device according to claim 1, characterized in that, The material segment identifier is generated by splicing together the conveyor line number, the start and end values ​​of the warning time window, the displacement range of the material in the conveying direction, and the weight range of the material.

3. The industrial fire early warning device according to claim 1, characterized in that, When performing hash digest solidification, the early warning packet generation and closed-loop processing module uses a hash algorithm to calculate a digest value for the complete content of the early warning packet and writes the digest value into a local unchangeable log.

4. The industrial fire early warning device according to claim 3, characterized in that, The early warning packet generation and closed-loop processing module is used for: Sending control pulses to the field controller to slow down the conveyor line and execute a shutdown action after a preset delay, or activating the on-site directional spraying device and atomizing dust suppression device to identify the material segments. The corresponding areas were physically treated.

5. The industrial fire early warning device according to claim 1, characterized in that, The multi-dimensional physical quantity collaborative sensing module includes: The timing-triggered and polarization imaging unit is used to generate a global synchronization clock signal and a frame index. According to the frame index The drive-controlled rotating mechanism switches the linear polarizer to the target polarization angle. And when the polarization angle is stable, the visible light imaging component is triggered to expose to obtain a resolution of Visible light image matrix ,in, and Pixel coordinates, target polarization angle The control logic satisfies the formula: ， in, The first polarization angle reference value, This is the angular difference between the first polarization angle and the second polarization angle; Near-infrared strobe and temperature field capture unit, referencing the frame index Used to control the radiation intensity of near-infrared supplementary lighting components And simultaneously trigger the thermal imaging component to acquire the infrared temperature field image matrix. ,in, and For thermal imaging pixel coordinates, radiation intensity The control logic satisfies the formula: ， in, This refers to the high-level radiated power when the supplementary light is turned on. The ambient light noise floor power when the supplementary lighting is off is given by this formula, which causes the supplementary lighting state to form an alternating bright and dark strobe sequence between odd and even frames, and the infrared temperature field image matrix... Each pixel value in the image represents the radiation temperature of the target object. The motion increment and load synchronization unit references the global synchronization clock signal for frame indexing. The cumulative pulse value of the rotary encoder is read at the corresponding exposure time. and the instantaneous weight value collected by the weighing sensor. Instantaneous weight value As load data, the physical displacement increment of the transport line in the current sampling period is calculated based on the pulse difference between the current frame and the previous frame. Physical displacement increment The calculation satisfies the formula: ， in, The physical diameter of the drive roller of the conveyor line. This represents the total number of pulses output by the rotary encoder in one revolution. This represents the cumulative number of pulses at the current frame time. The physical displacement increment is the cumulative pulse count of the previous frame. With the instantaneous weight value Packed as a matrix corresponding to visible light images The motion attribute metadata.

6. The industrial fire early warning device according to claim 1, characterized in that, The spatiotemporal reference unification and alignment module includes: Heterogeneous image pixel-level registration unit, used to establish pixel coordinates of visible light images. and infrared temperature field image pixel coordinates The homography mapping relationship between them, and the generated registered temperature field matrix with the same resolution as the visible light image. The registration logic satisfies the following formula: , in, for The homography transformation matrix describes the rotation, translation, and perspective projection relationships of the thermal imaging component relative to the visible light imaging component. The heterogeneous image pixel-level registration unit applies the homography mapping relationship to the infrared temperature field image matrix. Perform backsampling interpolation to obtain the pixel coordinates corresponding to the visible light image. The radiation temperature value is used to achieve pixel-level alignment between visible light images and infrared temperature field images in the spatial dimension; Global displacement integral and coordinate mapping unit, using physical displacement increments obtained by converting displacement pulse sequences. Used to calculate the current cumulative travel position of the conveyor line. Furthermore, a warning coordinate mapping benchmark is established between image pixel coordinates and conveyor line physical coordinates, and the cumulative travel position is calculated. The calculation satisfies the formula: , in, The physical position corresponding to the initial encoder reading. From the start time to the current frame Cumulative variables; warning coordinates Satisfying the formula: ， In the formula, These are the early warning coordinates in the physical coordinate system of the conveyor line. The coordinates are the longitudinal coordinates along the conveying direction. The horizontal coordinate is perpendicular to the conveying direction. and represents the physical resolution coefficients of image pixels in the vertical and horizontal directions; Material fragment identifier dynamic generation unit, referencing cumulative travel position Image pixel physical resolution coefficients, key frame sequences, and load data acquired by the weighing sensor. Used to generate the material fragment identifier corresponding to the current keyframe. Material fragment identification The generation logic satisfies the string concatenation formula: , in, For string concatenation operators, This is a fixed number for the current conveyor line. and The timestamps are the start and end values ​​corresponding to the current keyframe. The width of the visible light image matrix is ​​in pixels. The physical length of the material covered by the current keyframe. The material fragment identifier is used as a unique index key value for the encapsulation and traceability of subsequent warning packages, based on the weight level coding according to the load data.

7. The industrial fire early warning device according to claim 1, characterized in that, The environmental baseline adaptive adjustment module includes: The flicker differential and interference index calculation unit is used in a preset dark field reference area. The average brightness values ​​of frames with and without supplemental lighting are extracted, and the interference index is calculated. The computational logic satisfies the formula: ,in, For the strobe cycle index, For the first Even-numbered frames within a period In the dark field reference area The average grayscale value of the pixels within the range. For the first Odd-numbered frames within a period In the dark field reference area The average pixel grayscale value within the range is used to prevent the sensor noise floor constant from having a denominator of zero. The interference sensitivity gain coefficient is a formula that quantifies the backscattering intensity of near-infrared active illumination by environmental suspended particles. The interference index... Used to characterize the degree of visibility impairment in the current monitoring space; The thermal imaging background baseline iteration unit is used to update the statistical thermal imaging background temperature baseline using a weighted moving average method. The update logic satisfies the formula: ， in, The background learning rate coefficient. This is the background temperature baseline matrix of the previous frame. When updating, this unit iterates over pixels that are not marked as suspected high-temperature targets to establish baseline data that reflects the normal heat distribution of the current environment. The adaptive threshold generation unit is used to generate dynamic hotspot discrimination thresholds, and the generation logic satisfies the formula: , in, The preset static temperature rise alarm threshold, The threshold correction coefficient is positively correlated. As a safety margin constant, this formula automatically raises the alarm threshold when the interference index increases to suppress false alarms of thermal image noise caused by dust-scattered infrared radiation. The hotspot discrimination threshold... and thermal imaging background temperature baseline It will be transmitted as a judgment parameter to the heterogeneous feature fusion judgment module.

8. The industrial fire early warning device according to claim 1, characterized in that, The heterogeneous feature fusion determination module includes: The polarization difference and reflection suppression unit is used to calculate the grayscale difference between the current keyframe and the previous keyframe at the same pixel coordinates, and generate a polarization difference map. According to the polarization difference diagram Generate a reflective suppression mask Reflection suppression mask The generation logic satisfies the formula: ， in, Preset reflection detection threshold; reflection suppression mask This is used to mark metallic reflective areas and high-brightness reflective areas as reflective interference areas, and to remove reflective interference areas from the candidate judgment area to obtain non-reflective effective areas; A cross-modal hotspot verification unit is used in reflectivity suppression masks. Within the corresponding non-reflective effective area, combined with the thermal imaging background temperature baseline and dynamic hot spot discrimination threshold Perform temperature anomaly detection and generate a candidate set of potential fire points. Suspected Fire Point Candidate Set Conditions met: , The cross-modal hotspot verification unit is used to filter the set of pixels that simultaneously meet the non-reflective characteristics and have a temperature rise exceeding the dynamic hotspot discrimination threshold, and to extract a candidate set of suspected fire points. The highest temperature values ​​of the inner pixels are arranged into a highest temperature sequence according to the time order of the keyframe sequence. The spatiotemporal trajectory consistency determination unit is used to calculate the candidate set of suspected fire points. The deviation of the centroid of the candidate region in the motion trajectory of the keyframe sequence deviation in motion trajectory The calculation satisfies the formula: , in, This refers to the centroid position of the candidate region in the current keyframe on the vertical image coordinate system. For the front The centroid position of the candidate region in the frame on the vertical image coordinate system. The vertical physical resolution coefficient is used when the candidate region is located within a non-reflective effective region, the temperature rise exceeds the dynamic hot spot discrimination threshold, and the highest temperature sequence meets the preset persistence or heating trend conditions. Less than the preset trajectory tolerance At that time, the heterogeneous feature fusion judgment module outputs an early warning event.

9. The industrial fire early warning device according to claim 1, characterized in that, The early warning package generation and closed-loop processing module includes: The structured encapsulation and hash digest solidification unit is used to construct an early warning package containing material fragment identifiers, early warning coordinates, maximum temperature sequences, and keyframe images. Early warning package The construction logic satisfies the vector combination formula: ,in, For material fragment identification, As early warning coordinates, This is the highest temperature sequence. Keyframe images marked with a reflection suppression mask; The structured encapsulation and hash digest persistence unit is further used to apply a secure hash algorithm to the warning packet. Calculate the complete content of the digital summary Digital summary The generation satisfies the formula: ,in, for Bit-safe hash function, For XOR operation, Preset timestamp salt value; digital digest Used for early warning packages Perform anti-tampering verification and event tracing; Risk quantification and strategy mapping unit, referencing early warning package The highest temperature sequence in Used to calculate fire hazard rating According to the fire hazard rating Generate safe operation instructions Fire risk rating Satisfies the weighted formula: ,in, This is the absolute temperature rise weighting coefficient. This is the static temperature rise alarm threshold. The heating rate is the weighting factor. The first derivative of the highest temperature sequence; Safe operation instructions Satisfies a piecewise function: ,in, and The thresholds for low-risk and high-risk assessment are defined as follows: It is a directional spray mode. Emergency shutdown mode; The dynamic delay compensation and execution unit is used to calculate the warning coordinates. The average running speed of the conveyor belt obtained by converting displacement pulse sequence The fixed physical coordinates of the downstream directional spraying device relative to the starting point of the conveyor line. and the mechanical response time of the solenoid valve Calculate the lag compensation time for the start-up of the directional sprinkler system. Furthermore, it issues a safe operation command to the field controller, with a delay in compensation time. The calculation satisfies the formula: ,in, Early warning coordinates Longitudinal coordinates along the conveying direction; The dynamic delay compensation and execution unit is used to ensure that the spray medium covers the target area corresponding to the material fragment identification.

Images