Smoke shielding false alarm intelligent identification system and method

Abstract

The embodiment of the invention provides a smoke shielding false alarm intelligent identification system and method, and is applied to the technical field of intelligent fire detection, and the method comprises the steps: synchronously collecting an imaging picture, radiant heat response data and a wind field state parameter of a suspected shielding region; the method comprises the following steps: identifying and identifying an abnormal shielding area in a picture; further analyzing the motion characteristics of the region along with the change of the wind field and the dynamic process of the thermal inertia change of the region; the core is to judge whether a time sequence lag relationship meeting a smoke diffusion physical rule exists between the two, so as to identify whether the shielding object is combustion smoke or not. And if the physical mechanism is not met, the alarm is inhibited, so that real fire smoke and non-fire interferents such as dust and steam are effectively distinguished, and the fire false alarm rate in a complex ventilation environment is remarkably reduced.

Description

Technical Field

[0001] This application relates to the field of intelligent fire detection technology, and in particular to an intelligent identification system and method for false alarms caused by smoke obstruction. Background Technology

[0002] In modern industrial sectors such as iron and steel metallurgy, chemical production, and thermal power generation, large enclosed or semi-enclosed workshops are generally equipped with forced ventilation systems to remove process waste gases, control ambient temperature, and ensure operational safety.

[0003] Existing technologies, whether based on visible light pixel change analysis or fused with thermal imaging or deep learning algorithms, still fundamentally rely on the visual or apparent statistical characteristics of smoke for their discrimination logic. In typical industrial scenarios such as the secondary cooling zone of continuous casting machines, the periphery of heat treatment furnaces, and drying channels, the flow patterns formed by a large number of non-fire aerosols under forced ventilation bear a striking resemblance to early fire smoke. This identification method, based on appearance rather than physical essence, inevitably leads to frequent false alarms in complex dynamic environments, severely disrupting normal production order.

[0004] Standalone photoelectric smoke detectors commonly found on the market are typically designed for civilian or general commercial environments. When directly applied to the aforementioned industrial scenarios, their fixed installation conditions conflict with the complex physical environment; their isolated alarm operation mode makes it difficult to integrate with industrial-grade security systems, failing to meet the management needs for real-time status monitoring and secondary response in critical process areas.

[0005] In industrial settings with forced ventilation, a prominent technical dilemma exists: on the one hand, there is a need for a detection terminal that can be installed as easily as a stand-alone smoke detector and operate with low power consumption; on the other hand, this terminal must have the ability to distinguish physical characteristics to resist complex aerosol interference and intelligently network its sensing data with the system status, achieving a leap from isolated alarms to systematic intelligent judgment. Existing technical solutions have not yet effectively resolved this dilemma. Summary of the Invention

[0006] This application provides an intelligent identification system and method for smoke obstruction false alarms. By examining whether the temporal relationship between motion and thermal changes conforms to the smoke physics model, it fundamentally solves the problem of high false alarm rate of visual fire detectors in complex industrial environments. To achieve the above objectives, this application adopts the following technical solution: A method for intelligently identifying false alarms due to smoke obstruction, applied in industrial sensing scenarios with forced ventilation, the method comprising: The image of the sensing area is acquired, the occlusion abnormal area in the image is identified, the occlusion abnormal area is initially marked as the occlusion area to be verified, and the corresponding area label is generated. Radiation thermal response data is collected for the spatial range corresponding to the area identifier to obtain the heat absorption and release state of the shielded area to be verified over time, and thermal inertial response data of the shielded area to be verified is generated. Obtain wind field state parameters within the sensing area; under the condition that the wind field state parameters change, track the spatial position change of the area marker in the continuous imaging image, and generate wind field response features characterizing the occluded area to be verified as a function of the wind field. Analyze the temporal correlation between the thermal inertial response data and the wind field response characteristics; determine whether there is a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified during wind field changes, which conforms to the physical laws of heat transfer and diffusion; and generate a determination result of the properties of the obstruction based on whether the temporal lag relationship is consistent with the expected behavior of the combustion smoke. If the determination result of the nature of the obstruction does not conform to the mechanism of combustion smoke diffusion, the alarm output is suppressed; if the determination result of the nature of the obstruction conforms to the mechanism of combustion smoke diffusion, the fire alarm is triggered.

[0007] In some possible implementations, generating the thermal inertial response data of the occlusion region to be verified includes: Based on the radiative thermal response acquisition, the apparent temperature sequence of the shielded area to be verified at continuous time points is obtained. Based on the apparent temperature sequence, the temperature change difference between adjacent time points is calculated to capture the dynamic process of heat absorption and release, and the original temperature change sequence is obtained. The original temperature change sequence is subjected to moving average filtering to suppress instantaneous interference, and a smooth thermal inertia change curve that can stably reflect the thermal inertia characteristics of the region is extracted. The smooth thermal inertia change curve is used as thermal inertia response data.

[0008] In some possible implementations, generating wind field response characteristics that characterize the shading region to be verified as a function of the wind field includes: Based on the image and the region marker, the optical flow method is applied to calculate the pixel-level motion vector field. Clustering statistics are performed on the vectors belonging to the occlusion region to be verified in the motion vector field to characterize the macroscopic motion trend of the occlusion region to be verified, and the overall motion speed and direction are obtained. Based on the changing time period indicated by the wind field state parameters, the evolution sequence of motion parameters within the changing time period is extracted from the continuous data of the overall motion speed and direction, and the evolution sequence is used as the wind field response feature.

[0009] In some possible implementations, the analysis of the temporal correlation between the thermal inertial response data and the wind field response characteristics; determining whether there is a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified during wind field changes, conforming to the physical laws of heat transfer and diffusion, includes: The time series of the thermal inertial response data and the time series of the wind field response characteristics are aligned with a unified time reference to obtain a time-synchronized paired sequence. Using time-synchronized paired sequences as input, the normalized cross-correlation function of the two is calculated during the wind field change period to obtain a cross-correlation sequence that reflects their correlation characteristics under different time shifts; Extract the peak value of the correlation coefficient and the corresponding time shift in the cross-correlation sequence, record the time shift as a time delay value, and output the time delay value as a quantitative indicator of the time lag relationship.

[0010] In some possible implementations, generating a result determining the nature of the obstruction based on whether the temporal lag relationship is consistent with the expected behavior of the combustion smoke includes: Call the pre-stored smoke diffusion and heat transfer coupling database to obtain the standard time delay range corresponding to the current ventilation conditions; The time delay value in the quantitative indicator is compared with the standard time delay range; if the time delay value is within the standard time delay range, a judgment result is generated indicating that it conforms to the smoke mechanism. If the delay value is outside the standard delay range, the result is determined to be inconsistent with the smoke mechanism.

[0011] In some possible implementations, the method further includes: Based on the wind field state parameters, a reference wind speed distribution map covering the sensing area is generated through flow field simulation calculations. The overall motion velocity and direction of the shielded area to be verified, characterized by the wind field response features, are vector-compared with the corresponding reference values ​​of the area in the reference wind speed distribution map to calculate the motion fit index. In the determination result of the generated occlusion property, the time delay value obtained by quantifying the time lag relationship is combined with the motion consistency index for joint judgment.

[0012] In some possible implementations, obtaining the wind field state parameters within the sensing area includes: Read the detection data output by the wind speed sensor array on the ventilation duct of the sensing area; Based on the detection data and the corresponding sensor spatial location, a two-dimensional wind speed field distribution map covering the entire sensing area is generated by processing the data using a spatial interpolation algorithm. The two-dimensional wind speed field distribution map and the record of its evolution over time are integrated into the wind field state parameters.

[0013] In some possible implementations, tracking the spatial position change of the region identifier in continuous imaging frames to generate wind field response features characterizing the occluded region to be verified as a function of the wind field includes: The system detects the evolution of the two-dimensional wind speed field distribution map over time. When the change in wind speed or wind direction exceeds a preset threshold, it determines that a wind field disturbance period has begun and marks the starting point. In response to the marker of the starting point, the image tracking sampling frequency for the region identifier is increased from the reference sampling frequency to a tracking sampling frequency higher than the reference sampling frequency; During the wind field disturbance period, image tracking of the area marker is maintained at the tracking sampling frequency to obtain a high temporal resolution location sequence; Based on the high temporal resolution location sequence, wind field response features are generated.

[0014] In some possible implementations, the method further includes: The results of each generation of the determination of the nature of the obstruction, the thermal inertial response data segment, the wind field response feature segment, and the quantitative index of the time lag relationship are associated and encapsulated to form a historical discrimination record. Based on the accumulated historical discrimination records, the discrimination criteria used for judgment are periodically calibrated; the calibration of the discrimination criteria includes: updating and optimizing the standard time delay range or the parameters in the smoke diffusion and heat transfer coupling database according to the time lag relationship quantification index in the historical discrimination records and the final judgment result.

[0015] A smoke obstruction false alarm intelligent identification system is applied in industrial sensing scenarios with forced ventilation conditions. The system includes: The image acquisition and preliminary recognition module is used to acquire continuous imaging images of the sensing area, identify occlusion abnormal areas in the images, initially mark the occlusion abnormal areas as occlusion areas to be verified, and generate corresponding area identifiers. The thermal inertial response acquisition module is used to acquire the radiative thermal response of the spatial range corresponding to the area identifier, obtain the heat absorption and release state of the shielded area to be verified over time, and generate thermal inertial response data of the shielded area to be verified. The wind field response analysis module is used to acquire wind field state parameters within the sensing area; under the condition that the wind field state parameters change, the spatial position change of the area marker in the continuous imaging image is tracked to generate wind field response features characterizing the occluded area to be verified as a function of the wind field. The physical law conformity determination module is used to analyze the temporal correlation between the thermal inertial response data and the wind field response characteristics; determine whether there is a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified during wind field changes that conforms to the physical laws of heat transfer and diffusion; and generate a determination result of the properties of the obstruction based on whether the temporal lag relationship is consistent with the expected behavior of the combustion smoke. The alarm control module is used to suppress alarm output when the determination result of the nature of the obstruction indicates that it does not conform to the mechanism of combustion smoke diffusion; and to trigger a fire alarm when the determination result of the nature of the obstruction indicates that it conforms to the mechanism of combustion smoke diffusion.

[0016] As can be seen from the above technical solution, this application has the following beneficial effects: 1. This invention fundamentally solves the problem of false alarms in high-interference environments by shifting fire detection from visual appearance comparison to physical behavior verification. Traditional methods attempt to distinguish between similar-looking substances, but their basis for judgment is fragile and unstable. This invention utilizes a forced ventilation environment, using wind field changes as a standard excitation, to simultaneously monitor the thermal inertial response and motion response of the target, and extracts the time delay values ​​of key physical quantities. This is used to verify whether it conforms to the convective diffusion heat transfer physical model unique to fire smoke. The temperature of water vapor is dominated by a fixed heat source and its relationship with motion is disordered; dust has extremely low thermal inertia, making it difficult to generate effective time delays. By verifying its temporal correlation law, this invention can deterministically eliminate non-smoke interference in principle, establishing detection reliability on universal physical laws, and achieving a fundamental leap from appearance recognition to mechanism verification.

[0017] 2. This invention also endows the system with high adaptability, strong interpretability, and continuous evolution capabilities. Through a pre-built physical model database, the system dynamically calls upon corresponding standard time delay ranges based on real-time wind fields, enabling the judgment criteria to be scenario-adaptive and ensuring the scientific validity and consistency of judgments under different ventilation conditions. Its alarm output is an evidence chain report containing multiple physical quantities such as time delay values, motion consistency, and thermal change curves, greatly improving the credibility and interpretability of the alarms. Furthermore, the system achieves self-learning through historical judgment records, continuously calibrating parameters and learning new interference patterns, thus transforming from a static tool into a dynamic intelligent system. This significantly reduces long-term operation and maintenance costs and ensures long-term reliable operation in complex industrial environments. Attached Figure Description

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

[0019] Figure 1 The main flowchart for intelligent smoke false alarm recognition provided in the embodiments of this application; Figure 2 A flowchart illustrating the generation of thermal inertial response data provided in this application embodiment; Figure 3 A flowchart for generating wind field response features provided in this application embodiment; Figure 4 A flowchart for calculating the delay value provided in the embodiments of this application; Figure 5 This is a flowchart illustrating the smoke detection logic provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are for distinguishing different objects, not for specifying a particular order.

[0021] In the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0022] Research has found that traditional visual fire detection methods rely solely on image appearance features such as the texture, color, and diffusion pattern of smoke for identification. In industrial environments where visually similar interferences exist, such as water vapor and dust, these methods cannot fundamentally distinguish between different substances, resulting in a high false alarm rate.

[0023] To address the aforementioned issues, this application provides an intelligent recognition system and method for false alarms caused by smoke obstruction: Example 1 To solve the above problems, such as Figures 1-5 As shown, the embodiment of the present invention is applied to a representative complex industrial environment, namely the secondary cooling area of ​​a continuous casting production line in a large steel enterprise.

[0024] Core features of the scene: Forced, organized ventilation: For process heat dissipation and personnel safety, the workshop is equipped with a ventilation system consisting of large fans and controllable vents, which can create a stable and controllable dominant airflow within the plant. This is not a natural breeze, but a wind field with a clear direction and a certain speed.

[0025] A persistent source of strong interference that is visually similar to smoke: High-temperature water vapor: The red-hot steel billet, which is over a thousand degrees Celsius, is cooled by high-pressure water mist spray, and instantly vaporizes to produce a large amount of white, diffuse, and continuously spreading water vapor.

[0026] Industrial dust: The operation of equipment and the conveying of materials will raise dry dust, forming a grayish-brown obscured area.

[0027] Strict fire safety requirements: The area contains electrical equipment, hydraulic systems, and flammable lubricating oils, posing a real fire risk, and early and reliable detection is essential.

[0028] The root of the current technological dilemma lies in the fact that traditional visual fire detectors, including products based on deep learning image recognition, have an extremely high false alarm rate in this scenario because their discrimination criteria remain at the level of visual appearance. Whether it is water vapor, dust, or smoke in the early stages of a fire, they are all semi-transparent obstructions with similar textures, colors, and movement patterns in the camera. The algorithm attempts to find subtle differences from these similar images, like distinguishing different shapes of clouds in dense fog. Its discrimination basis is fragile and unreliable, and is highly susceptible to changes in lighting, viewing angle, and density.

[0029] This invention departs from the traditional approach of distinguishing images at the image level, instead employing an innovative path of identifying the essence of matter through physical behavior. Its core idea is as follows: The root cause of the false alarms is misjudging non-smoke obstructions as smoke. Therefore, the key to solving the problem is not to look more closely, but to verify the underlying cause.

[0030] Utilizing environmental conditions: The inherent forced ventilation in the scene is not merely a disturbance; it can be transformed into a detection tool. This is because the wind field is a common source of excitation that affects all floating objects.

[0031] Establish the principle of identification: Different substances, such as smoke and gas, water vapor droplets, and solid dust particles, have different thermophysical properties, such as specific heat capacity, heat absorption and release rate, and kinetic properties, such as mass, inertia, and ability to follow airflow. When they are excited by the same wind field change, the resulting thermodynamic response, such as how the temperature changes and the kinematic response, how they move, and the correlation between the two will be fundamentally different.

[0032] Construction Method Logic: Therefore, this invention designs a method: First, suspicious targets are visually detected; then, the thermal and motion responses of the target under wind field changes are simultaneously detected; finally, the temporal correlation between these two responses is analyzed to see if it matches the convective-diffusion heat transfer physical model unique to fire smoke. If it matches, an alarm is triggered; otherwise, suppression is initiated.

[0033] Based on the above derivation, the overall implementation process of this embodiment in the continuous casting workshop is as follows: First, an abnormal, diffuse occlusion area was detected in the cooling zone using a visible light camera and marked as a target for investigation. Then, the system simultaneously performs three tasks: 1. Use an infrared thermal imager to monitor the temperature change trend in the area; 2. Read the workshop wind speed network data, and pay special attention to the moments of sudden wind field changes caused by the start-up and shutdown of fans; Third, during sudden changes in the wind field, the system closely tracks the target's movement trajectory using a camera. Next, the system aligns and analyzes the temperature change trend curve and the motion change curve on the time axis, calculating a key indicator: the time delay between the motion change and the temperature change. Finally, this measured time delay value is compared with the theoretical time delay range for smoke obtained from the physical model library based on the current workshop ventilation conditions, thus scientifically determining whether the target is hazardous smoke or harmless water vapor or dust.

[0034] Implementation details: Image Acquisition and Initial Identification: Waterproof and dustproof industrial cameras deployed high in the workshop continuously capture panoramic views of the secondary cooling area. The system's built-in intelligent video analysis module uses an algorithm called visual background extraction to create a dynamic background model for each pixel. When high-temperature steel billets are moved in or large amounts of water vapor are ejected, the pixel brightness and contrast in local areas of the image continuously change, creating a significant difference from the background model. The system extracts these changing areas, forming a binary foreground mask. After filtering out minor noise and merging adjacent areas, any connected region exceeding a preset size (approximately the size of a cabinet door) is considered an occlusion anomaly. The system immediately generates a digital profile for this region, i.e., a region identifier, which refers to the digital representation created by the system for each identified occlusion anomaly region. The data structure should include: a unique ID, a timestamp, and the contour pixel coordinates or bounding rectangle coordinates in the image. Implementation Path: After connected component analysis in computer vision, an ID is assigned to each contour, and its position information is recorded. This step completes the crucial transformation from anomaly images to specific target objects to be verified.

[0035] Acquiring Thermal Inertia Response: An infrared thermal imager is installed at a precisely calibrated location, not far from the visible light camera. Once the area is identified, the system automatically locates the same physical position in the imager's view based on the pre-defined spatial correspondence between the two devices. The thermal imager measures the infrared radiation intensity of the area at a fixed frequency, such as twice per second, and converts it into an apparent temperature value representing the surface's hot or cold state. Continuously recording these temperature values ​​yields the raw temperature sequence. To eliminate random fluctuations from single-point temperature measurements, the system calculates the temperature difference between adjacent time points, obtaining a raw sequence reflecting the rate of heating or cooling. This difference sequence is then processed using a moving average filter, averaging the differences across multiple consecutive time points to represent the trend at that moment. After smoothing, a stable curve clearly reflecting the overall thermal inertia of the area—the degree of sluggishness of temperature change—is generated. This is the thermal inertia response data, representing a smooth curve of the apparent temperature change in the area to be verified, reflecting its thermal inertia. Implementation path: From acquiring the original temperature sequence to calculating the difference between adjacent time points to obtain the original change sequence, to applying the moving average filter to generate a smooth curve, the significance of this step is to extend the visual features of the target to the infrared thermal radiation features and obtain its intrinsic thermophysical property information.

[0036] Wind field response characteristics are collected: Several ultrasonic anemometers installed at key locations within the workshop continuously transmit real-time wind speed and direction data back to the system. The system software uses spatial interpolation algorithms to depict this discrete data into a wind speed distribution overview map covering the entire workshop plane, complete with arrows indicating direction and length—the wind field state parameters. The system continuously monitors changes in this map. When a large fan is activated by the control room operator, or the ventilation window opening is adjusted, the main airflow in the workshop changes significantly within a short period. When the system detects that this change exceeds a threshold—for example, an increase in average wind speed exceeding 0.5 meters per second—it immediately determines that a valuable wind field disturbance observation window has been entered. Simultaneously, the system abruptly increases the visual tracking frequency of the area marker from the usual twice per second to ten times per second. Using a high-performance kernel correlation filter tracking algorithm, the system precisely records the center coordinates of the target area at every moment, much like closely following a moving icon with a mouse. Based on the coordinate changes, the system calculates the magnitude and direction of its movement per second. Arranging all velocity and direction data within the entire observation window in chronological order constitutes the wind field response characteristics of the target. The core of this step is to actively capture and record the dynamic motion performance of the target under known environmental excitation.

[0037] Analysis of temporal correlation and judgment: This is the brain of the entire method. The system places the thermal inertial response data curve and the motion velocity curve on the same high-precision time axis. The core question is: which comes first, the event of the target starting to accelerate or the event of a significant temperature change? How much time is between them? To quantify this relationship, the system uses the mathematical tool of cross-correlation analysis. It slides the two curves relative to each other on the time axis, calculating the degree of similarity under each possible misalignment. When a misalignment time is found that makes the waveforms of the two curves most closely match, this misalignment time is the time delay value we are looking for. For example, the analysis found that when the motion curve is advanced by three seconds, its shape matches the temperature change curve the most. This quantifies the physical fact that motion precedes thermal change by three seconds.

[0038] Cross-correlation analysis is used to accurately quantify the time lag relationship (delay value) between thermal inertial response and motion response.

[0039] Step 1: Time Series Alignment and Preprocessing. This involves aligning and preprocessing the thermal inertial response data sequence. Wind field response characteristics (velocity) sequence Time synchronization is performed, and the DC component is removed to obtain... ; For a specific moment Direct readings of the raw thermal inertial response data collected; All within the time window used for analysis The average value represents the baseline thermal state during that time period. The data represents the preprocessed thermal inertial response, reflecting the fluctuation of the thermal state at time t relative to the average thermal state. For a specific moment Direct readings of the raw wind field response characteristics (velocity); All within the time window used for analysis The average value represents the baseline level of exercise during that time period; The preprocessed wind field response characteristics reflect the fluctuation of the velocity at time t relative to the average velocity.

[0040] Step 2: Calculate the normalized cross-correlation function (NCC). For each possible time shift... (For example, from -10 seconds to +10 seconds, with a step size of 0.1 seconds), calculate the normalized cross-correlation of the two sequences: .

[0041] Step 3: Extract the delay value. Find the value that makes... Get the global maximum value This value is the desired time delay. For example, if the maximum value appears... =3.2 seconds, which means that the change in motion leads the change in heat by 3.2 seconds.

[0042] Judgment: The system internally stores a smoke behavior database based on fluid mechanics and heat transfer principles. Based on the current fan operation status (e.g., the west-side fan is fully open and the south-facing window is half-open), the database can provide the theoretically expected smoke delay value (e.g., two to five seconds) for a typical oil or cable fire under these ventilation conditions. The system compares the calculated measured delay value (e.g., three seconds) with the theoretical range of two to five seconds provided by the database. If the measured value falls within the theoretical range, it proves that the thermal motion coupling behavior of the target conforms to the physical diffusion laws of fire smoke, and the system is judged as a real fire alarm. If the measured value is much smaller or larger than this range (e.g., zero seconds or ten seconds), it indicates that the behavior pattern is abnormal and does not conform to the smoke mechanism, and the system is judged as an interfering object.

[0043] Alarm control: For targets identified as smoke, the system will send the highest level fire alarm signal to the central fire control console without hesitation, and activate emergency broadcasts and evacuation instructions. For targets identified as water vapor or dust, the system will intelligently suppress the alarm triggered by visual detection to avoid false alarms and disturbing residents. At the same time, it will generate a detailed record in the background that includes all analysis data for safety engineers to review and analyze afterward.

[0044] In forced-ventilation industrial settings, the visual similarity between water vapor, dust, and smoke is a source of false alarms. This invention does not delve into the more difficult image differentiation at the source; instead, it introduces a new and more reliable identification dimension: physical behavior response. By simultaneously acquiring the thermal and motion responses of the same target under wind field changes and analyzing their temporal correlation, this invention essentially verifies whether the target conforms to the fluid-heat coupling physical model unique to smoke: "motion driven by convection occurs first, followed by temperature changes due to heat exchange." For water vapor, its temperature may be dominated by the steel billet below, resulting in a disordered relationship with motion; for dust, its thermal inertia is minimal, with almost no lag. Therefore, by verifying the conformity of the physical model, non-smoke obstructions can be definitively excluded from the alarm queue in principle. This solves the inherent false alarm problem of traditional methods relying on superficial features, establishing the reliability of fire detection on a solid physical foundation, rather than on unstable image statistical features.

[0045] Generate thermal inertial response data and extract stable thermal features from noise.

[0046] In the specific setting of a continuous casting workshop, the thermal environment of the target object, whether smoke or water vapor, is extremely harsh: the background is intense radiation from a high-temperature moving steel billet, there are localized low-temperature zones caused by cooling water spray, and the air itself is turbulent. The temperature values ​​directly read by the infrared thermal imager fluctuate wildly and are filled with noise.

[0047] The specific implementation of this step is as follows: After obtaining the original temperature sequence of the target, the system does not use it directly. It first calculates the temperature difference between adjacent time points to obtain an initial temperature change rate sequence. This sequence reflects whether heat is absorbed or released at each instant, but it contains many spikes. Next, the system uses a moving average filtering technique to smooth it. It sets a time window, for example, covering several seconds, calculates the average of all temperature change rates within the window, and uses this average as the representative value at the center of the window. Then the window slides forward one time step, and the calculation is repeated. This process is analogous to using a wide brush to draw a shaky line, ultimately resulting in a smooth, continuous curve, i.e., a smoothed thermal inertia change curve.

[0048] In the billet cooling zone, water vapor plumes, being close to the high-temperature source, may exhibit high-frequency, dramatic temperature fluctuations; while the temperature changes of smoke are relatively more moderate, influenced by the ambient air. Directly using the raw temperature difference data, both can be difficult to distinguish due to noise interference. The specific effect of moving average filtering is that it effectively filters out short-term temperature fluctuations caused by instantaneous local airflow scouring, thermal imager noise, or rapid changes in background radiation. After filtering, the potentially rapidly oscillating thermal characteristics of water vapor, dominated by billet radiation, and the relatively slowly changing thermal characteristics of smoke, which rely more on convective heat transfer, can be more clearly and stably distinguished and presented. This provides high-quality input data reflecting the inherent thermal inertia of the target for subsequent accurate time-series correlation analysis, avoiding the risk of physical laws being obscured by excessive data noise.

[0049] The generation of wind field response characteristics accurately captures macroscopic motion trends.

[0050] When wind fields change, the movement within an obstruction, especially a non-uniform mass like water vapor, may not be entirely uniform; the edges may diffuse faster while the core moves slower. Simply tracking its outer center point may result in lost information or introduce errors.

[0051] The specific implementation of this step is as follows: During periods of wind disturbance, the system employs an algorithm called dense optical flow on continuous visible light images. This algorithm calculates the direction and distance of movement of each pixel in the image between two frames, forming a motion vector field map. Then, the system extracts only the pixel-level motion vector fields belonging to the region identifier. Since these vector directions may not be identical, the system performs clustering statistics: it analyzes the directional distribution of these vectors, identifies the most dominant movement direction intervals, and then calculates only the average velocity and direction of all vectors within this dominant interval. This average value represents the overall, mainstream movement trend of the occluded area.

[0052] In the complex airflow conditions of a workshop, water vapor plumes may be dispersed at their edges and their shapes may constantly change during movement; dust plumes may partially settle. If only a simple center point is tracked, the trajectory may be abrupt and discontinuous, failing to accurately reflect the overall motion. This step employs a combination of optical flow and cluster statistics. Its specific advantage lies in its ability to resist interference from target deformation, splitting, or internal uneven movement. Through cluster analysis, it intelligently focuses on the motion patterns of the main target component, filtering out the influence of secondary movements such as edge dissipation. The resulting overall motion velocity and direction are a more reliable indicator of the target's ability to respond to the wind field as a unified entity. In the context of a continuously casting workshop where target shapes are dynamically changing, this ensures the representativeness of the extracted motion features, providing reliable kinematic data for subsequent analysis.

[0053] Quantitative analysis of time-series correlations transforms physical laws into calculable indicators.

[0054] The time lag relationship is a qualitative concept and must be quantified into a numerical indicator that can be processed and compared by computers.

[0055] The specific implementation of this step is as follows: The system first ensures that the timestamps of the thermal inertia curve and the velocity curve are completely synchronized. Then, it performs a normalized cross-correlation function calculation. This process can be visualized as follows: keeping the thermal inertia curve stationary, the velocity curve is gradually shifted to the left on the time axis, representing an advance in motion; or to the right, representing a lag in motion. For each small step of shift, such as 0.1 seconds, the similarity of the two curves in the overlapping region is calculated. When the velocity curve is shifted to a position where the undulations of the two curves most closely match, the calculated correlation coefficient reaches its maximum value. The amount of time required for the velocity curve to shift at this point is the desired time delay value. The system automatically records this maximum correlation coefficient and its corresponding time delay value.

[0056] The core characteristic of smoke behavior is that motion precedes thermal change, but how much of a lead is effective? Water vapor may exhibit motion and thermal change almost synchronously, with a time delay close to zero, or even, due to a fixed heat source, motion may be driven by thermal change, resulting in a theoretical negative time delay. The cross-correlation calculation method provided in this step offers a precise and objective mathematical tool, transforming the vague physical perception of leading or lagging into a concrete time delay value measured in seconds. This value forms the basis for subsequent comparisons with theoretical models. In continuous casting workshops, extensive analysis may reveal that the time delay values ​​of water vapor often concentrate in a narrow range near zero seconds, while smoke is distributed within a clearly defined positive range. This quantification transforms the judgment from subjective estimation to objective decision-making based on a defined numerical threshold, greatly improving the accuracy and repeatability of the judgment process.

[0057] Database-based judgments allow standards to adapt to specific environments.

[0058] Different ventilation intensities and wind directions will alter the rate of smoke diffusion and cooling, thus affecting the time delay value. A fixed threshold cannot be applied to all operating conditions.

[0059] The specific implementation of this step is as follows: Before the system is put into use, the implementer will pre-construct a database of smoke diffusion and heat transfer coupling using scientific methods. This includes two parts: First, a three-dimensional digital model of the workshop is established in a computer, and computational fluid dynamics software is used to simulate the diffusion of virtual smoke and temperature field changes under different fan combinations and different fire source locations, from which theoretical time delay values ​​are extracted. Second, when the workshop is safely shut down, controlled small-scale standard fire experiments are conducted, such as using a standard oil pan and measuring the time delay values ​​with precision instruments to calibrate the simulation model. Ultimately, the database will generate multiple records. For example, when fan number one is turned on and the south-facing window is open at 60%, the smoke time delay value ranges from 2.8 seconds to 4.2 seconds. During actual operation, the system judges the current ventilation conditions in real time and dynamically retrieves the matching standard time delay range from the database, that is, the theoretical time delay interval between the actual smoke movement and heat change determined based on the physical model and experimental data.

[0060] For example, taking the continuous casting workshop in this embodiment as an example, an exemplary method for constructing a database is as follows: 1) Based on the 3D CAD model of the workshop, the fire source is set to be located at typical risk points such as the hydraulic station and electrical cabinet; 2) Using CFD software such as ANSYS Fluent, the diffusion process and temperature field evolution of the smoke generated by the above-mentioned fire source were simulated under various ventilation conditions, such as the west-side fan being fully open and the south-side window being half open, and the theoretical time delay value distribution was extracted. 3) During the safe shutdown of the workshop, a standard oil pan fire test is conducted at the above-mentioned risk points. The system is used to synchronously collect thermal and motion data and calculate the measured time delay value for calibration and verification of the simulation model in step 2). 4) Associate the ventilation operating parameters with the corresponding time delay confidence interval (e.g., 2.8-4.2 seconds) to form database records.

[0061] The ventilation mode in a continuous casting workshop is not static and may be adjusted according to the season and production stage. Strong exhaust ventilation may be activated during the day and reduced at night. This step introduces a comparison mechanism using a pre-stored database, the specific effect of which is that it makes the judgment criteria adaptive to different scenarios. The system no longer uses a rigid global threshold, but instead selects the most suitable expected range of the physical model as the judgment benchmark based on the actual ventilation conditions at the current moment. This means that regardless of whether the wind is strong or weak, the system's expected behavior regarding smoke is reasonable and consistent with the current physical environment. This solves the problem of fixed thresholds sometimes missing or sometimes false alarms due to changes in operating conditions, ensuring the consistency and scientific nature of the system's judgment accuracy under various operating states.

[0062] By combining a comprehensive assessment of motion fit, a double verification mechanism is added for added security.

[0063] In some extremely complex situations, relying solely on time delay values ​​may still result in a small probability of misjudgment. For example, a particular thermal disturbance in the airflow might coincidentally produce a time delay resembling smoke. To achieve extremely high system reliability, a second layer of physical verification is required.

[0064] The specific implementation of this step is as follows: While determining the time delay value, the system initiates another verification path. First, based on the current rotational speed of each fan and the opening of the windshield, it runs a simplified airflow simulation program to quickly calculate the theoretically expected average airflow direction map within the workshop under the current equipment settings, i.e., the reference wind speed distribution map. Then, after the wind field stabilizes, the system measures the actual average motion direction of the target to be verified and compares it with the theoretical wind direction at the target's location, calculating the angle between the two. Simultaneously, it also compares the ratio of the actual average speed to the theoretical wind speed. Finally, these measures of directional and speed consistency are combined into a single motion consistency index using a weighted formula, i.e., the degree of consistency between the motion trajectory of the area to be verified and the theoretical wind field prediction trajectory.

[0065] Motion fit calculation steps: Motion fit is used to quantify the degree of matching between the motion trajectory of the shading area to be verified and the theoretical wind field prediction. It is an indicator for comprehensively evaluating the consistency between the target motion and the theoretical wind field.

[0066] Step 1: Calculate the theoretical motion vector. Based on the two-dimensional wind speed field distribution map obtained from flow field simulation or sensor interpolation, the system extracts the theoretical wind speed vector at the center point of the region to be verified. ,in That's the theoretical speed. It's a theoretical trend.

[0067] Step 2: Calculate the actual motion vector. During the wind disturbance period, the overall motion vector of the region to be verified is obtained through optical flow and cluster statistics. ; To indicate how fast the occluded region to be verified moves in the image sequence; This indicates the azimuth angle of the movement of the occluded region to be verified in the image sequence.

[0068] Step 3: Calculate the directional alignment. Calculate the cosine of the angle between the actual direction of motion and the theoretical wind direction: Directional alignment = The closer this value is to 1, the more consistent the directions.

[0069] Step 4: Calculate the speed match. Calculate the ratio of the actual speed to the theoretical speed and normalize it to between 0 and 1: Speed ​​match = .

[0070] Step 5: Calculate the comprehensive index. Use a weighted average to combine the direction and speed consistency: Motion fit = Directional fit + Speed ​​matching degree. Among them... and Weighting coefficients (e.g.) ),and .

[0071] In the final determination, the system will comprehensively consider two conditions: whether the latency value is within the standard range and whether the motion matching index is high enough. Only when both conditions are met will it be finally confirmed as smoke.

[0072] Near the continuous casting machine, the high-temperature steel billet generates a strong upward thermal plume, which may cause the nearby water vapor cloud to not move strictly horizontally, but rather with an upward component. Simultaneously, vortex zones may exist in the corners of the workshop. This step introduces a motion matching degree joint judgment, which provides a check on the rationality of the target's motion trajectory. Real smoke, as a passive tracer, should have a macroscopic motion that closely matches the main ventilation airflow field of the workshop. However, water vapor, strongly influenced by local heat sources, may have a motion direction that deviates significantly from the main airflow. By calculating the motion matching degree, the system can effectively identify interference targets that, although their time delay characteristics may resemble smoke, have significantly different motion trajectories and are dominated by local factors. This adds a solid layer of protection to the discrimination, further confirming that smoke must simultaneously satisfy both the correct time delay relationship and a reasonable motion trajectory—two physical constraints—reducing the overall false alarm rate of the system to an extremely low level.

[0073] Acquiring wind field state parameters, from point to area environmental perception.

[0074] The wind field in the workshop is not uniform. The wind speed is high at the fan outlet and low in the corner. The wind speed data from only one or two points cannot represent the true wind conditions at the location of the obstruction.

[0075] The specific implementation of this step is as follows: Multiple ultrasonic anemometers are carefully arranged within the workshop, forming a sensor network. The system's central processing unit periodically collects readings from all sensors, employing an algorithm called inverse distance weighted spatial interpolation. This algorithm divides the workshop floor into a fine grid. For each grid point, it examines data from all surrounding sensors; the closer the sensor is to that point, the greater its data weight. Through weighted calculation, the wind speed and direction at that grid point are estimated. After traversing all grid points, a continuous electronic map of wind speed and direction distribution covering the entire workshop area is generated—a two-dimensional wind speed field distribution map.

[0076] The tall, densely packed equipment in the continuous casting workshop obstructs and diverts airflow, resulting in a complex flow field. A water vapor cloud might be located in a weak wind zone on the leeward side of a particular piece of equipment. This step uses a sensor array and spatial interpolation to acquire wind field state parameters. Its specific effect is that it upgrades the measurement of the workshop's wind field from discrete points to continuous surface sensing. This allows the system to know the local wind conditions at the specific location of the obstruction to be verified, rather than relying on a rough estimate from data from a distant sensor. This is crucial for calculating motion fit, as comparisons are only meaningful when the theoretical reference wind speed is based on an estimate of the target's local wind field. This improves the spatial accuracy and contextual accuracy of the entire physical verification process.

[0077] By finely tracking wind field disturbances, we can capture key analytical windows.

[0078] Continuously performing image tracking and optical flow calculations at the highest frequency places a huge burden on the processor. Furthermore, when the wind field is stable, the slow drift of the target is of little analytical value.

[0079] The specific implementation of this step: The system operates by default at the baseline sampling frequency, which is the normal data acquisition frequency for imaging images and thermal response when the wind field is stable; for example, processing two frames per second, solely for maintaining target lock. Its core is the wind field disturbance detection module, which continuously analyzes the average data transmitted from the wind speed sensor network. Once it detects that the change in wind speed or direction exceeds a preset sensitivity threshold within a short time, such as one second, it immediately sends a trigger signal indicating the start of a wind field disturbance to the tracking module. Upon receiving the signal, the tracking module instantly increases the processing frequency for that specific target to the tracking sampling frequency, i.e., the sampling frequency increased by the system to capture dynamic responses when the wind field changes significantly; for example, ten frames per second, entering high-resolution data acquisition mode. This high-frequency tracking will continue for the entire preset observation period, such as twenty seconds, after which it will return to the baseline frequency.

[0080] The start-up and shutdown of workshop fans are intermittent events. The event-driven adaptive sampling strategy designed in this step enables on-demand allocation and precise deployment of computing resources. The system remains in a low-power monitoring state most of the time, only concentrating its power at the highest temporal resolution to capture the target's most dynamic response process when a key stimulus event revealing the target's physical nature occurs, such as a change in the wind field. This ensures extremely high temporal accuracy of the motion sequence used to calculate the delay value, resulting in a highly precise calculated delay value, with an error within a fraction of a second. Simultaneously, it avoids unnecessary high-load continuous computation, enabling the system to operate stably on cost-effective edge computing devices, making it suitable for long-term deployment in industrial settings.

[0081] The system is self-learning and optimizing, making it an intelligent system with growth capabilities.

[0082] Pre-established physical model databases and judgment thresholds may not be able to fully cover all unknown interferences, or over time, equipment aging and layout fine-tuning may cause the environment to drift slowly.

[0083] The specific implementation of this step: The system automatically packages each complete judgment process into a historical judgment record. This record is like a patient's complete medical record, including everything from the original image, thermal data, and wind field data, to the intermediate generated feature curves, calculated time delay values, motion matching degree, and finally the judgment result. Administrators can review these records periodically, quarterly, and label typical events, such as confirmed false alarms and confirmed fire drills, as real. The system uses these labeled records to initiate a calibration process. For example, if it finds that all events marked as real smoke have time delay values ​​that are generally slightly higher than the original range in the database under a certain ventilation mode, it will automatically make a gentle adjustment to that range, such as shifting a portion towards the mean of the new data. At the same time, it will also cluster and analyze records marked as water vapor, learning their time delay value and motion feature combination patterns. In the future, when encountering targets with similar feature combinations, even if their individual time delay values ​​are close to the smoke range, they can make a more cautious judgment or provide a prompt because their overall pattern matches known interference.

[0084] The production site is dynamic; new cooling processes may be introduced, generating new types of steam, or changes in dust removal equipment may alter dust characteristics. The self-optimization mechanism established in this step empowers the system to continuously learn from operational experience, enabling it to adapt to slow environmental changes and accumulate experience in dealing with unknown disturbances. The system is no longer a static system fixed at deployment, but a dynamic system capable of continuous self-calibration and self-improvement over time. In the complex industrial environment of a continuous casting workshop, this capability is crucial, allowing the system's judgment to keep pace with changes in the production environment and even learn disturbance patterns that have not been clearly summarized by humans. This achieves long-term, stable, and highly reliable operation, significantly reducing subsequent maintenance and calibration costs.

[0085] This invention, through a detailed explanation of a typical scenario in a continuous casting workshop, fully demonstrates a complete intelligent smoke false alarm identification scheme from theory to practice. Centered on physical law compliance verification, it constructs a logically rigorous and highly adaptable intelligent identification system through a series of closely integrated technical means, including multi-source sensor synchronization, spatiotemporal information alignment, feature quantization extraction, dynamic model comparison, dual-verification decision-making, and continuous self-learning optimization. It not only solves the specific false alarm problem in this scenario, but its methodology also provides an innovative solution paradigm for other industrial safety sensing scenarios with similar forced ventilation conditions and complex interference, such as chemical plants, power plants, and large warehouses.

[0086] Example 2: Application of VOCs emission pipelines in chemical enterprises.

[0087] This embodiment aims to further demonstrate the universality and specific technical implementation path of the method of the present invention through an application scenario with significantly different technical challenges, environmental conditions and data characteristics - the volatile organic compound (VOCs) emission pipeline of a chemical enterprise.

[0088] This embodiment selects a main VOCs waste gas treatment pipeline in a chemical industrial park as the implementation scenario. The pipeline has a diameter of 3 meters and is responsible for collecting and transporting pre-treated organic waste gas from multiple production units to the Regenerative Thermal Oxidizer (RTO). Its interior is a typical closed, long-distance, forced-ventilation environment.

[0089] Ventilation characteristics: Ventilation is forced by an induced draft fan, with a unidirectional airflow and a stable and adjustable velocity (typically 8-15 m / s). The flow field tends to be stable laminar in the central region of the pipe cross-section, while turbulence exists in the boundary layer. This contrasts sharply with the open space and complex turbulent wind field in Example 1.

[0090] Core source of interference: High-concentration, non-uniform exhaust gas cloud: The composition, concentration, and temperature of exhaust gas emitted from different production units fluctuate, forming gas clouds with uneven composition and density in the pipeline, causing a smoke-like obstruction and refraction effect on optical observation.

[0091] Condensed water vapor: Changes in exhaust gas temperature may cause water vapor to condense, forming fine droplets suspended in the air.

[0092] Dust adhering to the inner wall of the pipe: Airflow may wash away a small amount of dust adhering to the inner wall.

[0093] Fire risk characteristics: The gas inside the pipeline is flammable. If it comes into contact with an open flame, static electricity or local high temperature (such as a hot spot), it is very easy to cause a deflagration or a fire that spreads along the pipeline, with extremely serious consequences.

[0094] Current technological challenges: Traditional pipeline fire detection mainly relies on point-type or line-type temperature sensing cables, which have slow response times and cannot distinguish between the actual temperature rise of flames / smoke and normal fluctuations in the temperature of exhaust gas components. Visible light or single-band infrared cameras are severely obstructed by exhaust gas, resulting in high false alarm and false negative rates.

[0095] Specific implementation details and supplementary information: Achieving high-precision synchronization and spatial registration of multi-source sensors: To overcome the engineering challenges of data spatiotemporal synchronization, this embodiment adopts the following feasible solution: Hardware Synchronization Trigger: The system's main controller generates a unified hardware trigger pulse signal, which directly drives the visible light high-speed camera, mid-wave infrared thermal imager, and pipeline anemometer via shielded cables. All sensors use this pulse as the starting reference for data acquisition, ensuring hardware-level synchronization of data acquisition with a time deviation of less than 1 millisecond.

[0096] Spatial registration: At the pipeline inspection port (equipped with a pressure-resistant observation window), visible light and infrared cameras are rigidly mounted side-by-side on a calibrated pan-tilt unit. By photographing the same high-low temperature combination calibration plate with a special shape placed at a known location on the pipeline cross-section, the precise perspective transformation matrix between the two camera images is calculated, achieving pixel-level coordinate mapping from abnormal areas in the visible light image to the infrared image.

[0097] Motion feature extraction in low visibility environments: To address the issues of severe obstruction by exhaust gas inside the pipeline and the failure of traditional optical flow methods, this embodiment employs an improved scheme that fuses multispectral information: Data source: A multispectral imager is used to simultaneously output registered images of the visible light channel, short-wave infrared channel (which has a certain penetrability to some organic gases), and mid-wave infrared channel (which directly reflects thermal radiation).

[0098] Feature fusion tracking: The system first identifies anomalous disturbance clusters in the shortwave infrared image that are significantly different from the background exhaust gas concentration distribution using an adaptive background subtraction method. Guided by these cluster regions, dense optical flow based on grayscale gradients is calculated within the corresponding regions of the mid-wave infrared image. This method effectively utilizes the higher contrast of temperature distribution in MWIR images, reducing the dependence on the sharpness of visible light images.

[0099] Motion vector calculation: For the extracted optical flow vector, the median statistical method, which is more robust, is used to calculate the overall motion velocity and direction of the clump, effectively resisting the outlier vector interference caused by local eddies in the pipe.

[0100] Accurate extraction and anti-interference processing of thermal inertial response data: To address the issues of uneven background temperature and periodic disturbances within the pipeline, the following refined treatment process is implemented: Establishment of reference temperature field: During normal operation cycles without alarms, the system continuously learns the temperature statistical model (mean and variance) of each pixel on the mid-wave infrared image to establish a dynamic reference temperature field for the pipe cross section.

[0101] Relative temperature sequence generation: After identifying the target area to be measured, the system calculates the difference between the average temperature of the area and the historical average of the reference temperature at the corresponding location, and obtains the relative temperature change sequence ΔT_rel(t) to eliminate the influence of long-term background temperature drift.

[0102] Feature band filtering: Based on the prior knowledge that the frequency of thermal inertia changes in real fires is low (usually below 1 Hz) while the frequency of interference such as turbulent mixing of exhaust gas is high (2 Hz), the system applies a low-pass digital filter to ΔT_rel(t). For example, a Butterworth filter with a cutoff frequency of 1.5Hz can be used to directly filter out high-frequency noise and obtain a smooth curve that can clearly reflect the thermal inertia nature of the target.

[0103] Basis for determining key parameters and adaptive settings: To clarify the feasibility of the technical solution, the setting logic of key parameters in this embodiment is explained as follows: Wind field change judgment threshold: During the initial commissioning phase, the standard deviation σ of wind speed data is collected when the wind turbines are running stably. The threshold for judging effective wind field disturbance is dynamically set to k*σ (k is usually 3~5). This method is based on the principle of statistical process control, ensuring that high-precision analysis is triggered only when the wind speed change significantly exceeds the normal random fluctuation.

[0104] The time shift search range for time series correlation analysis: The search range for the time shift τ in the normalized cross-correlation calculation is set based on the estimated airflow transmission time T_transit = L / V, calculated using the physical length L of the pipe and the current average wind speed V. For example, it can be preset to [-0.5T_transit, 1.5T_transit]. This allows the analysis to focus on a physically reasonable time delay interval, improving computational efficiency and accuracy.

[0105] Motion consistency index threshold: A comprehensive judgment threshold for the consistency of direction and velocity. The optimal decision boundary is obtained through supervised machine learning training on known real-world interference events and simulated fire alarm experimental data from historical data, for example, using a Support Vector Machine (SVM) classifier. The system can be periodically retrained with new data to achieve adaptive optimization.

[0106] Implementation effect analysis: To verify the effectiveness of the solution described in this embodiment, a systematic test was conducted in a simulated experimental pipeline. The test simulated three scenarios 50 times each: fire smoke (Class A), high-concentration organic waste gas plumes (Class B), and water mist (Class C). The method of this invention achieved a 100% detection rate for Class A scenarios; the false alarm rates for Class B and Class C non-fire interference were only 2% and 0%, respectively. In contrast, an AI video fire detector relying on visual appearance analysis, tested during the same period, achieved an 88% detection rate for Class A, but its false alarm rates for Class B and Class C were as high as 76% and 64%, respectively. While traditional temperature-sensing cables showed no response to non-fire interference, their detection rate for early-stage low-temperature smoke in Class A was only 82%, with a significant delay in warning.

[0107] The above data clearly demonstrates that even in novel scenarios involving enclosed pipes and strong optical interference, this invention, by utilizing the core principle of wind field excitation and analyzing the temporal lag relationship between heat and motion, can achieve an extremely low false alarm rate against visually similar non-fire interference while ensuring extremely high fire sensitivity. This proves that the technical effectiveness of this invention does not depend on specific scenarios or sensor combinations, but rather stems from its innovative methodology of transcending appearances and making essential identification based on physical laws, resulting in unexpected technological advancements.

[0108] From open turbulent flow workshops to closed laminar flow pipelines, the core physical identification principle of this invention remains consistent and performs exceptionally well. This demonstrates that this invention provides a universal, robust, and physically-based innovative solution paradigm for addressing the problem of false alarms in fire detection caused by visual similarity interference in forced ventilation industrial environments.

[0109] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently identifying false alarms due to smoke obstruction, applied in industrial sensing scenarios with forced ventilation, characterized in that, The method includes: The image of the sensing area is acquired, the occlusion abnormal area in the image is identified, the occlusion abnormal area is initially marked as the occlusion area to be verified, and the corresponding area label is generated. Radiation thermal response data is collected for the spatial range corresponding to the area identifier to obtain the heat absorption and release state of the shielded area to be verified over time, and thermal inertial response data of the shielded area to be verified is generated. Obtain wind field state parameters within the sensing area; under the condition that the wind field state parameters change, track the spatial position change of the area marker in the continuous imaging image, and generate wind field response features characterizing the occluded area to be verified as a function of the wind field. Analyze the temporal correlation between the thermal inertial response data and the wind field response characteristics; determine whether there is a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified during wind field changes, which conforms to the physical laws of heat transfer and diffusion; and generate a determination result of the properties of the obstruction based on whether the temporal lag relationship is consistent with the expected behavior of the combustion smoke. If the determination result of the nature of the obstruction does not conform to the mechanism of combustion smoke diffusion, the alarm output is suppressed; if the determination result of the nature of the obstruction conforms to the mechanism of combustion smoke diffusion, the fire alarm is triggered.

2. The method according to claim 1, characterized in that, The generation of thermal inertial response data for the occlusion region to be verified includes: Based on the radiative thermal response acquisition, the apparent temperature sequence of the shielded area to be verified at continuous time points is obtained. Based on the apparent temperature sequence, the temperature change difference between adjacent time points is calculated to capture the dynamic process of heat absorption and release, and the original temperature change sequence is obtained. The original temperature change sequence is subjected to moving average filtering to suppress instantaneous interference, and a smooth thermal inertia change curve that can stably reflect the thermal inertia characteristics of the region is extracted. The smooth thermal inertia change curve is used as thermal inertia response data.

3. The method according to claim 1, characterized in that, The generation of wind field response features characterizing the occlusion region to be verified as a function of wind field includes: Based on the image and the region marker, the optical flow method is applied to calculate the pixel-level motion vector field. Clustering statistics are performed on the vectors belonging to the occlusion region to be verified in the motion vector field to characterize the macroscopic motion trend of the occlusion region to be verified, and the overall motion speed and direction are obtained. Based on the changing time period indicated by the wind field state parameters, the evolution sequence of motion parameters within the changing time period is extracted from the continuous data of the overall motion speed and direction, and the evolution sequence is used as the wind field response feature.

4. The method according to claim 1, characterized in that, The analysis examines the temporal correlation between the thermal inertial response data and the wind field response characteristics; it determines whether, during wind field changes, there exists a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified (the area under test) that conforms to the physical laws of heat transfer and diffusion, including: The time series of the thermal inertial response data and the time series of the wind field response characteristics are aligned with a unified time reference to obtain a time-synchronized paired sequence. Using time-synchronized paired sequences as input, the normalized cross-correlation function of the two is calculated during the wind field change period to obtain a cross-correlation sequence that reflects their correlation characteristics under different time shifts; Extract the peak value of the correlation coefficient and the corresponding time shift in the cross-correlation sequence, record the time shift as a time delay value, and output the time delay value as a quantitative indicator of the time lag relationship.

5. The method according to claim 4, characterized in that, The step of generating a determination result of the nature of the obstruction based on whether the time lag relationship is consistent with the expected behavior of the combustion smoke includes: Call the pre-stored smoke diffusion and heat transfer coupling database to obtain the standard time delay range corresponding to the current ventilation conditions; The time delay value in the quantitative indicator is compared with the standard time delay range; if the time delay value is within the standard time delay range, a judgment result is generated indicating that it conforms to the smoke mechanism. If the delay value is outside the standard delay range, the result is determined to be inconsistent with the smoke mechanism.

6. The method according to claim 1, characterized in that, The method further includes: Based on the wind field state parameters, a reference wind speed distribution map covering the sensing area is generated through flow field simulation calculations. The overall motion velocity and direction of the shielded area to be verified, characterized by the wind field response features, are vector-compared with the corresponding reference values ​​of the area in the reference wind speed distribution map to calculate the motion fit index. In the determination result of the generated occlusion property, the time delay value obtained by quantifying the time lag relationship is combined with the motion consistency index for joint judgment.

7. The method according to claim 1, characterized in that, The acquisition of wind field state parameters within the sensing area includes: Read the detection data output by the wind speed sensor array on the ventilation duct of the sensing area; Based on the detection data and the corresponding sensor spatial location, a two-dimensional wind speed field distribution map covering the entire sensing area is generated by processing the data using a spatial interpolation algorithm. The two-dimensional wind speed field distribution map and the record of its evolution over time are integrated into the wind field state parameters.

8. The method according to claim 7, characterized in that, The step of tracking the spatial position change of the region identifier in continuous imaging images and generating wind field response features characterizing the occluded region to be verified as a function of the wind field includes: The system detects the evolution of the two-dimensional wind speed field distribution map over time. When the change in wind speed or wind direction exceeds a preset threshold, it determines that a wind field disturbance period has begun and marks the starting point. In response to the marker of the starting point, the image tracking sampling frequency for the region identifier is increased from the reference sampling frequency to a tracking sampling frequency higher than the reference sampling frequency; During the wind field disturbance period, image tracking of the area marker is maintained at the tracking sampling frequency to obtain a high temporal resolution location sequence; Based on the high temporal resolution location sequence, wind field response features are generated.

9. The method according to claim 5, characterized in that, The method further includes: The results of each generation of the determination of the nature of the obstruction, the thermal inertial response data segment, the wind field response feature segment, and the quantitative index of the time lag relationship are associated and encapsulated to form a historical discrimination record. Based on the accumulated historical discrimination records, the discrimination criteria used for judgment are periodically calibrated; the calibration of the discrimination criteria includes: updating and optimizing the standard time delay range or the parameters in the smoke diffusion and heat transfer coupling database according to the time lag relationship quantification index in the historical discrimination records and the final judgment result.

10. A smoke obstruction false alarm intelligent recognition system, applied in industrial sensing scenarios with forced ventilation, characterized in that, The system includes: The image acquisition and preliminary recognition module is used to acquire continuous imaging images of the sensing area, identify occlusion abnormal areas in the images, initially mark the occlusion abnormal areas as occlusion areas to be verified, and generate corresponding area identifiers. The thermal inertial response acquisition module is used to acquire the radiative thermal response of the spatial range corresponding to the area identifier, obtain the heat absorption and release state of the shielded area to be verified over time, and generate thermal inertial response data of the shielded area to be verified. The wind field response analysis module is used to acquire wind field state parameters within the sensing area; under the condition that the wind field state parameters change, the spatial position change of the area marker in the continuous imaging image is tracked to generate wind field response features characterizing the occluded area to be verified as a function of the wind field. The physical law conformity determination module is used to analyze the temporal correlation between the thermal inertial response data and the wind field response characteristics; determine whether there is a temporal lag relationship between the thermal inertial response data and the spatial motion response of the area to be verified during wind field changes that conforms to the physical laws of heat transfer and diffusion; and generate a determination result of the properties of the obstruction based on whether the temporal lag relationship is consistent with the expected behavior of the combustion smoke. The alarm control module is used to suppress alarm output when the determination result of the nature of the obstruction indicates that it does not conform to the mechanism of combustion smoke diffusion; and to trigger a fire alarm when the determination result of the nature of the obstruction indicates that it conforms to the mechanism of combustion smoke diffusion.

Images