A building fire safety monitoring and early warning management method and system

Abstract

This application relates to a method and system for building fire safety monitoring and early warning management, applicable to chemical storage sites adjacent to residential areas, used to distinguish between fires involving household goods and fires involving chemicals. The method includes: simultaneously collecting smoke, temperature, VOC, and video data; extracting smoke characteristics, flame temperature, VOC characteristics, and video liquid leakage characteristics; based on serialized fusion judgment, first triggering initial screening through VOC anomalies or high temperature and low smoke to locate spatiotemporal anchor points; then verifying liquid spreading characteristics and their temporal consistency with VOC changes, outputting a conclusion of chemical or mixed fire. The system includes a multi-source sensing and serial discrimination module, which can prevent accidental spraying, prompt for appropriate extinguishing agents, and provide judgment criteria for differentiated linkages such as explosion-proof smoke extraction and power outages, solving the problems of difficulty in identifying fire types in mixed scenarios and secondary disasters caused by accidental application of water-based extinguishing agents.

Description

Technical Field

[0001] This application relates to the field of intelligent building fire protection, and in particular to a method and system for building fire safety monitoring and early warning management. Background Technology

[0002] Existing building fire monitoring systems mostly use point or line detectors for early warning, such as smoke detectors, temperature detectors, and combustible / toxic gas detectors. Upon alarm activation, they trigger sprinkler systems, smoke extraction systems, audible and visual alarms, and fire control equipment. In recent years, some scenarios have also introduced video surveillance and image recognition to identify fire signs such as open flames and smoke obscuring the view, thereby improving alarm timeliness and visualization. However, most of these solutions focus on whether a fire has occurred, emphasizing the determination of fire occurrence and alarm triggering, with limited ability to differentiate the nature of the fire (e.g., the combustion of ordinary combustibles versus the combustion of chemicals / flammable liquids).

[0003] In chemical storage facilities adjacent to residential areas, there may be both ordinary fires caused by household goods (cardboard boxes, wooden packaging, plastic products, etc.) and fires caused by or following leaks of chemicals (such as flammable liquids like ethanol). The response strategies for different fire situations vary significantly: for example, using water-based fire extinguishing or directly activating sprinklers under certain flammable liquid combustion conditions may lead to the spread of flammable liquids, the spread of fire, and even secondary risks. In existing technologies, single sensors (only one of smoke / temperature / gas) are easily affected by factors such as ventilation, dust, water vapor, and light reflection, resulting in false alarms or missed alarms. Even when using multiple sensors, common practices are often simple threshold parallel connection or coarse-grained fusion, lacking a mechanism for stratified screening of "chemical fires," regional localization, and correlation between liquid spread signs in video and changes in gas concentration. This makes it difficult to reliably identify cases of chemical involvement in combustion or leaks at an early stage. Summary of the Invention

[0004] The purpose of this application is to solve the aforementioned problems of difficulty in reliably distinguishing between chemical fires and ordinary fires, and the easy misinterpretation leading to improper handling.

[0005] According to one aspect of this application, a method for monitoring and early warning management of building fire safety is provided, applied to chemical storage sites adjacent to residential areas, for distinguishing between fires involving household goods and fires involving chemical products, comprising: S10. Collect smoke, temperature, VOC concentration and video data in the target area, perform time-series alignment, and generate multi-source data units; S20. Extract a discrimination feature set from the multi-source data unit, including: smoke concentration and growth rate, smoke obstruction rate, flame equivalent temperature, VOC concentration and growth rate, and suspected liquid area features based on video. S30. Based on the discriminative feature set, perform serialization fusion determination: S31. Determine whether the initial screening conditions for chemical products ignition are met: the VOC concentration or its growth rate exceeds the corresponding threshold, or the flame equivalent temperature exceeds the high temperature threshold and the smoke obscuration rate is lower than the low obscuration threshold; if not, output the determination of household products ignition. S32. If so, determine the triggering time and the corresponding suspected fire area; S33. Determine whether the suspected liquid area feature in the suspected fire area exceeds the spreading threshold within the time window before and after the triggering time. S34. If so, determine whether the changes in the characteristics of the suspected liquid area and the changes in the VOC growth rate meet the consistency condition. S35. If so, output a determination that the fire situation includes chemical products on fire.

[0006] Preferably, after S35, the method further includes: if it is determined that there are chemical products on fire in the fire, determining whether there are ordinary combustibles participating in the combustion at the same time, and distinguishing between mixed fire and chemical product fire accordingly; The determination of whether the common combustible material participates in combustion is made within a preset time window centered on the triggering time within the suspected ignition area, and satisfies any of the following determination rules: The smoke obscuration rate exceeds a preset optical obscuration threshold and the smoke color blackness exceeds a preset blackness index threshold, and continues to meet a preset duration threshold; or The smoke concentration growth rate exceeds a preset smoke growth rate threshold, and the smoke obstruction rate remains above the optical obstruction threshold for a preset duration; When the condition that the common combustible material participates in combustion is met, the output is "mixed ignition"; otherwise, the output is "chemical ignition".

[0007] Preferably, the smoke blocking rate is calculated in the following manner: Within the suspected fire area, at least one frame of image within a reference time window prior to the triggering time is used as reference image I. ref And acquire the current image I within a preset time window centered on the trigger time. t Calculate the following within the suspected fire area: The brightness contrast index C is the standard deviation of the brightness channel, the local contrast statistic, or a combination of the two. Edge intensity index E, wherein the edge intensity index is gradient magnitude energy, edge pixel ratio, or a combination of the two; And calculate the contrast attenuation ΔC=1 C(I t ) / C(I ref) and / or edge attenuation ΔE=1 E(I t ) / E(I ref The smoke obstruction rate Occ(t) is obtained by normalizing ΔC and / or ΔE, where Occ(t) is the weighted sum of ΔC and ΔE or the maximum value thereof; the moving average or median filter is applied to Occ(t) to suppress instantaneous jitter.

[0008] Preferably, the blackness of the smoke color is calculated in the following manner: In the suspected fire area, at least one frame of image within a reference time window before the triggering time is selected as a reference image, and the current image is obtained within a preset time window centered on the triggering time. Smoke candidate areas are determined in the suspected fire area, and the bright flame areas are masked. Within the smoke candidate area, a brightness statistic is calculated, which is the mean, quantile, or a combination of the grayscale / brightness channels; and the dark pixel percentage is calculated, which is the percentage of pixels with brightness below a preset brightness threshold. The blackness index is obtained by normalizing the change of the brightness statistic relative to the corresponding brightness statistic of the reference image and the proportion of dark pixels. The blackness index is the weighted sum of the change and the proportion of dark pixels or the maximum value of the sum. The blackness index is then subjected to moving average or median filtering to suppress illumination fluctuations and instantaneous jitter, and the filtered blackness index is used as the blackness feature of the smoke color.

[0009] Preferably, the smoke concentration growth rate and / or VOC growth rate are calculated as follows: Within a preset time window centered on the trigger time, smoke concentration sequence and VOC concentration sequence are acquired respectively, and the growth rate is calculated within a sliding calculation window. The growth rate includes: the difference mean growth rate obtained by performing first-order difference on adjacent sampling points and taking the mean, and / or, the fitting slope growth rate obtained by fitting the data within the sliding calculation window. The growth rate is filtered by moving average or median to suppress instantaneous disturbances, and the filtered growth rate is compared with a preset growth rate threshold.

[0010] Preferably, the liquid spreading characteristic is calculated as follows: Within the suspected fire area, a preset ground / countertop sub-area is determined as the liquid detection area; at least one frame of reference image is selected within a reference time window before the triggering time, and the current image is acquired within a preset time window centered on the triggering time; A liquid candidate change map is calculated within the liquid detection area. The liquid candidate change map includes at least one of the following: a reflection enhancement change obtained based on the brightness difference between the current image and the reference image, and / or a texture change obtained based on the local texture energy change between the current image and the reference image. The liquid candidate change map is segmented by threshold and filtered by connected components to obtain liquid candidate regions, and the area change, expansion rate and / or boundary continuity index of the liquid candidate regions are calculated within the preset time window. The area change, expansion rate and / or boundary continuity indices are normalized and combined to obtain liquid spreading characteristics, and the liquid spreading characteristics are subjected to moving average or median filtering; when the liquid spreading characteristics exceed the liquid spreading threshold, the liquid spreading determination condition in S33 is satisfied.

[0011] Preferably, the gas-liquid consistency is calculated in the following manner: Within a preset time window centered on the trigger moment, a liquid spreading characteristic sequence and a VOC growth rate sequence are acquired, and their consistency is quantified. The consistency quantification includes at least one of the following: Proportion of changes in the same direction: The percentage of sampling points where the trend of changes in liquid spreading characteristics changes in the same direction as the VOC growth rate; Time-series correlation: Calculate the correlation between liquid spreading characteristics and VOC growth rate within the allowable time offset range and take the maximum correlation; The consistency quantification result is compared with the consistency threshold. When the consistency quantification result exceeds the consistency threshold, the consistency determination condition in step S34 is satisfied.

[0012] Preferably, after outputting the fire of a mixture or chemical substance, the system further includes generating and outputting a fire extinguishing and coordinated control strategy that matches the type of fire, specifically including: When the output indicates a chemical fire or a mixed fire, a water-based fire extinguishing disabling warning is issued, and a water system linkage suppression command is sent to the fire alarm control equipment. The water system linkage suppression command includes at least intercepting, delaying, and / or manually confirming the start command for sprinkler and / or fire hydrant water supply. When the output indicates a fire involving chemicals or a mixture of chemicals, an alternative fire extinguishing method prompt and / or a corresponding linkage command will be output. The alternative fire extinguishing method shall include at least one or more of foam fire extinguishing, dry powder fire extinguishing, carbon dioxide fire extinguishing and / or inert gas suppression. When the output indicates a mixed fire, a phased handling prompt is provided. The phased handling prompt includes at least covering and isolating or suppressing the combustion of flammable liquids first, and then handling the remaining embers of ordinary flammable materials.

[0013] Preferably, it also includes implementing conservative linkage control for uncertain branches suspected of chemical ignition, specifically including: Following S33, the following is also included: If the suspected liquid area features do not exceed the spreading threshold, a suspected chemical fire warning is output, and the water system linkage suppression command is generated and issued. Following S34, the following is also included: If the consistency condition is not met, a suspected chemical fire warning will be output, and the water system linkage suppression command will be generated and issued. The water system linkage suppression command includes at least intercepting, delaying, and / or manually confirming the execution of the start command for water supply to sprinklers and / or fire hydrants; Furthermore, after the suspected chemical fire warning is issued, the water system linkage suppression command is maintained until one of the following release conditions is met: manual verification release; or the chemical fire determination conditions are met in subsequent monitoring and the disposal and linkage control strategy is switched to; or the preset chemical risk release conditions are met in subsequent monitoring.

[0014] This application also provides a building fire safety monitoring and early warning management system, which applies the building fire safety monitoring and early warning management method as described in any of the above claims, including: The multi-source sensing module is used to collect smoke data, temperature data, VOC concentration data and video data, and output multi-source raw monitoring streams; The data processing module, whose input end is connected to the output end of the multi-source sensing module, is used to synchronize the multi-source raw monitoring stream in time to generate time-series data units, and to construct a fire discrimination feature set based on the time-series data units; A serial discrimination module, whose input is connected to the output of the data processing module, is used to perform serial fusion judgment based on the fire discrimination feature set to output the fire type; The strategy output module, whose input is connected to the output of the serial discrimination module, is used to output the fire type and generate and issue early warning information and / or linkage control commands that match the fire type.

[0015] This application offers the following advantages: Multi-source synchronous acquisition (smoke / temperature / VOC / video) avoids the susceptibility of single sensors to interference from ventilation, dust, and lighting, preventing false alarms / missed alarms and improving the reliability of fire monitoring. Constructing a discriminative feature set (abnormal VOC, high temperature and low obscuration, suspected liquid features) quantifies the differences between ordinary fires and chemical fires into comparable threshold features, enhancing distinguishability. Serial fusion judgment (initial screening triggers, then locating suspected fire areas, then verifying liquid spread, and then verifying gas-liquid consistency) uses serial gating to converge the evidence chain step by step, reducing false alarms caused by "simple threshold parallelism" and improving the confidence level of chemical fire identification. Calculation within a limited time window and within the suspected fire video area reduces global search false alarms and computational load, facilitating real-time engineering implementation and improving judgment stability. Determining that "the fire includes chemical fires" provides a basis for subsequent selection of differentiated linkage strategies, reducing the risk of secondary risks such as spread caused by treating chemical fires as ordinary fires. Attached Figure Description To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a logic block diagram of the building fire safety monitoring and early warning management method according to one embodiment of this application. Detailed Implementation

[0017] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0019] Please refer to Figure 1 One embodiment of this application provides a building fire safety monitoring and early warning management method, applied to chemical storage sites adjacent to residential areas, for distinguishing between fires involving household goods and fires involving chemicals, including: S10. Collect smoke, temperature, VOC concentration, and video data within the target area, perform time-series alignment, and generate multi-source data units. It should be noted that this embodiment does not simply summarize the data from each sensor, but rather aligns data from different sampling frequencies using a unified time base, enabling the simultaneous acquisition of smoke, temperature, VOC concentration, and corresponding video frames / segments at the same time point. During the alignment process, resampling can be performed using sensor timestamps, and short-term missing data can be interpolated or filled with the most recent value to ensure that subsequent feature extraction and judgment logic uses "same source, same time" information within the same time segment, thereby reducing misjudgments caused by cross-sensor time deviations.

[0020] S20. Extract a discriminative feature set from the multi-source data units, including: smoke concentration and growth rate, smoke obscuration rate, flame equivalent temperature, VOC concentration and growth rate, and suspected liquid area features based on video. In this step, it should be noted that the discriminative features selected in this embodiment are constructed around the risk point of "involvement by chemicals (especially flammable liquids)," emphasizing distinguishability and verifiability.

[0021] Among them, smoke concentration and growth rate are used to reflect the trend of smoke generation and diffusion; smoke obstruction rate is used to reflect the degree of decrease in video visibility, thus characterizing the intensity of smoke obstruction of the field of view; flame equivalent temperature is used to reflect the intensity and change of heat released by combustion; VOC concentration and growth rate are used to reflect abnormal release of volatile organic compounds or gas anomalies accompanying combustion; suspected liquid area features are used to describe the appearance changes, area changes, or expansion trends of locations in the video such as the ground or tabletop that are prone to leakage / spreading. The above features provide the basic input for subsequent serial judgment, and their extraction range can be limited to the target area or key sub-regions to reduce the interference caused by changes in irrelevant areas.

[0022] S30. Based on the discriminant feature set, perform serial fusion judgment. In this step, it should be noted that this embodiment adopts a "trigger first, focus then verify" serial fusion judgment method, rather than a simple parallel connection of multiple features. That is, first, a suspected chemical fire branch is triggered under conditions sensitive to chemical risks; then, the suspected fire area is determined near the trigger time; subsequently, within this area and a limited time window, evidence of liquid spreading and gas-liquid consistency is further verified, forming a progressively converging evidence loop in the judgment chain, thereby improving reliability and reducing false alarms. Specifically, this includes: S31. Determine if the initial screening conditions for chemical ignition are met: VOC concentration or its growth rate exceeds the corresponding threshold, or the flame equivalent temperature exceeds the high-temperature threshold and the smoke obscuration rate is lower than the low-obscuration threshold. If not, output a fire determination for household goods. In this step, it should be noted that the initial screening conditions are set with two trigger paths. One is the VOC anomaly path, used to capture the rapid increase in VOCs accompanying leakage, volatilization, or combustion; the other is the "high temperature and low obscuration" combination path, used to capture the "strong heat release but relatively low smoke obscuration" that may occur when some flammable liquids burn. The thresholds can be preset or adaptively calibrated based on the site's background conditions (such as ventilation conditions, normal VOC levels, and daily operational fluctuation range) to reduce false triggers caused by environmental factors. If neither path is triggered, the fire is classified into the regular determination path for household goods fires to avoid unnecessary chemical risk linkage.

[0023] S32. If yes, determine the trigger time and the corresponding suspected fire area. In this step, it should be noted that: the trigger time can be the earliest time that meets the initial screening criteria, or the time when the triggering degree is most significant within a preset time range; the suspected fire area can be determined by the flame candidate area in the video, the area corresponding to the temperature anomaly, or a combination of both, and can be appropriately expanded to cover the flame boundary and its adjacent area. By determining the trigger time and the suspected fire area, subsequent video analysis and liquid identification can be limited to a smaller spatial range and a shorter time range, reducing false detections caused by full-screen retrieval and lowering computational overhead.

[0024] S33. Determine whether the characteristics of the suspected liquid area in the suspected fire area exceed the spreading threshold within the time window before and after the triggering time. In this step, it should be noted that: this embodiment sets a time window before and after the triggering time to cover the adjacent processes of "leakage / spreading—fire"; and extracts spreading-related characteristics such as area changes, expansion trends, or continuous changes of the suspected liquid area within the suspected fire area to form liquid spreading characteristics. The spreading threshold is used to filter out non-liquid spreading situations such as transient reflections, shadow changes, and ground water stains, ensuring that only when significant spreading signs appear does the next consistency verification step proceed, thereby improving the discrimination constraint on "chemical (especially liquid) involvement".

[0025] S34. If so, determine whether the changes in the characteristics of the suspected liquid area and the changes in the VOC growth rate meet the consistency condition. In this step, it should be noted that the consistency condition is used to establish the correlation between "visible liquid spreading in the video" and "abnormal VOC growth" to achieve cross-verification. Specifically, it can be evaluated from the perspectives of consistent change direction, temporal proximity of the changes, and the degree of synchronization within the allowable sensor response lag range. When enhanced liquid spreading and VOC growth show a mutually supportive relationship, it is considered that gas evidence and visual evidence corroborate each other, thereby significantly reducing the risk of misjudgment caused by relying solely on a single path (VOC only or video only).

[0026] S35. If yes, output a determination that the fire involves ignited chemicals. It should be noted in this step that the output of this embodiment is a determination that "the fire involves ignited chemicals," used to support the selection of subsequent linkage strategies. Because the handling methods for chemical fires (such as the combustion of flammable liquids like ethanol) differ from those for fires involving everyday consumer goods, and water-based fire extinguishing methods may pose a risk of spread and diffusion in some liquid combustion scenarios, this determination result can serve as an important basis for subsequent linkage control and fire extinguishing strategy switching.

[0027] The technical solution implemented in this embodiment can establish a "trigger-focus-verification" evidence chain in chemical storage sites adjacent to residential areas using multi-source information such as smoke, temperature, VOCs, and video, effectively distinguishing between fires of household goods and fires of chemical products. Compared to the single-sensor alarm or multi-threshold parallel judgment commonly used in the background technology, this embodiment constrains the calculation and discrimination to the vicinity of the trigger time and the suspected fire area through serialized fusion judgment, reducing false alarms and missed alarms caused by environmental disturbances (ventilation, dust, water vapor, light reflection, etc.). Furthermore, by introducing suspected liquid spreading characteristics and their consistency verification with VOC growth rate, "visible liquid signs in video" and "gas concentration changes" are cross-verified, improving the reliability and interpretability of chemical product fire judgment. This provides a more accurate basis for subsequent early warning and linkage control strategies matched to the fire type, reducing the secondary risks such as the spread caused by treating chemical product fires as ordinary fires.

[0028] Furthermore, S35 also includes: if it is determined that there are chemicals on fire in the fire, it is necessary to determine whether ordinary combustibles are also involved in the combustion, and to distinguish between mixed fire and chemical fire based on this.

[0029] The determination of whether ordinary combustibles participate in combustion is carried out within a preset time window centered on the triggering time within the suspected ignition area, and meets any of the following determination rules: The smoke obscuration rate exceeds a preset optical obscuration threshold, and the smoke color blackness exceeds a preset blackness index threshold, and continues to meet a preset duration threshold. The smoke concentration growth rate exceeds the preset smoke growth rate threshold, and the smoke obscuration rate remains above the optical obscuration threshold for a preset duration.

[0030] The output indicates mixed ignition when the criteria for combustion of ordinary combustibles are met; otherwise, the output indicates chemical ignition.

[0031] In this embodiment, it should be noted that after determining that "the fire involves the ignition of chemicals," this embodiment does not simply classify the fire as a chemical fire. Instead, it further considers the actual scenario of adjacent residential areas—chemical storage areas are often surrounded by packaging materials, cardboard boxes, plastic products, and other daily necessities or common combustibles. Once a chemical fire ignites surrounding combustibles, the fire will evolve from a single form to a mixed form of "liquid / chemical combustion + solid combustible combustion." This mixed form differs from a simple chemical fire in terms of spread path, smoke characteristics, and handling sequence. Therefore, it is necessary to determine whether common combustibles are also involved in the combustion, provided that a chemical fire is confirmed, in order to output a fire type that more closely reflects the actual situation on-site.

[0032] Specifically, this embodiment limits the determination of whether ordinary combustibles are involved in combustion to a preset time window centered on the triggering time within the suspected ignition area. This is to focus the determination on "accompanying combustion near the chemical ignition source" rather than irrelevant smoke from a distance, thereby reducing misjudgments caused by smoke flow, personnel movement, and local obstruction. The determination logic adopts two types of evidence paths: "smoke obstruction intensity + smoke color and blackness" and "smoke generation rate + continuous obstruction." On the one hand, ordinary combustibles (such as cardboard boxes, plastics, etc.) are more likely to produce higher smoke obscuration and darker smoke color characteristics when they burn. Therefore, when the smoke obscuration rate exceeds the optical obscuration threshold and the smoke color blackness exceeds the blackness index threshold, and the above state continues to reach the preset duration threshold, it can be considered that the probability of ordinary combustibles participating in combustion is significantly increased. On the other hand, if there is enhanced combustion or rapid spread, the smoke concentration will show a continuous upward trend. Therefore, when the smoke concentration growth rate exceeds the smoke growth rate threshold, and the smoke obstruction rate remains above the optical obstruction threshold for a preset duration, it can also be used as another determination path for ordinary combustibles to participate in combustion. By "continuously satisfying" this time constraint, occasional false triggering caused by instantaneous smoke plumes, short-term obstruction, or lens shake can be suppressed.

[0033] When either of the above paths is valid, this embodiment outputs a mixed fire to indicate that on-site handling should take into account the parallel risks of chemical combustion and residual fire of ordinary combustibles; when neither path is valid, it outputs a chemical fire, indicating that the fire is more likely to be dominated by chemical combustion, and subsequent linkage strategies can prioritize the safe handling of chemical fires.

[0034] The technical solution implemented in this embodiment can further identify whether ordinary combustibles are ignited based on the confirmed chemical fire, thereby subdividing the fire from a single chemical fire into two categories: "chemical fire" and "mixed fire," which more closely reflects the actual risk evolution process of adjacent residential areas. By limiting the judgment scope to the time window near the suspected fire area and the triggering time, and introducing a dual-path judgment method with continuous constraints of "masking + blackness" and "growth rate + continuous masking," it can reduce misjudgments caused by smoke disturbance, short-term smoke plumes, and changes in light, and improve the stability of mixed fire identification. This provides a more accurate basis for subsequent linkage control and on-site handling, avoiding the problem of mistreating mixed fires as a single chemical fire or a single ordinary fire, which could lead to improper handling order and increased risk of spread.

[0035] Furthermore, the smoke obscuration rate is calculated as follows: Within the suspected fire area, at least one frame of image within a baseline time window prior to the triggering time is used as the reference image Iref, and the current image It is acquired within a preset time window centered on the triggering time. The following calculations are performed within the suspected fire area: The brightness contrast ratio C is the standard deviation of the brightness channel, the local contrast statistic, or a combination of the two.

[0036] Edge strength index E is the gradient magnitude energy, edge pixel ratio, or a combination of the two.

[0037] And calculate the contrast attenuation ΔC=1 C(It) / C(Iref) and / or edge attenuation ΔE=1 E(It) / E(Iref) is used to normalize ΔC and / or ΔE to obtain the smoke obstruction rate Occ(t), where Occ(t) is the weighted sum of ΔC and ΔE or the maximum value of the sum. Moving mean or median filtering is applied to Occ(t) to suppress transient jitter.

[0038] In this embodiment, it should be noted that the "smoke obscuration rate" is not simply measured by changes in image grayscale or a single threshold. Instead, it is a dual-index obscuration metric constructed to address the typical visual effects caused by fire smoke—a decrease in overall contrast and the smoothing of edge details—thereby maintaining better robustness under different lighting conditions, different background textures, and different camera parameters.

[0039] Specifically, this embodiment first selects at least one frame of image within a reference time window prior to the triggering time in the suspected fire area as a reference image. The purpose is to obtain the baseline visual state when "not significantly affected by smoke". Subsequently, the current image is continuously acquired within a preset time window centered on the triggering time to characterize the visibility changes caused by smoke generation and diffusion. To avoid interference across the entire screen, this embodiment limits the calculation to the suspected fire area, making the occlusion rate closer to the actual smoke obstruction near the fire point, and unaffected by distant smoke drift or personnel obstruction.

[0040] In terms of metric selection, the brightness contrast metric is used to reflect the "overall graying and weakened tonal gradation" caused by smoke. For example, the standard deviation can be calculated on the brightness channel as a global contrast measure, or a local contrast statistic can be calculated to reflect the intensity of brightness fluctuations in textured areas; in some scenarios, the two can be combined to accommodate both uniform and complex backgrounds. The edge strength metric is used to reflect the "blurred details and lost outlines" caused by smoke. For example, the strength of edge response within a region can be measured by gradient amplitude energy, or the proportion of edge pixels can be measured by the proportion of detectable edges; similarly, the two can be combined to adapt to different background structures.

[0041] After obtaining the contrast and edge intensity indices of the reference and current images, this embodiment uses "relative attenuation" rather than "absolute value" to characterize occlusion. This reduces the impact of camera exposure changes and background material differences on absolute indices. The contrast attenuation reflects the degree of decrease in current contrast relative to the reference contrast, and the edge attenuation reflects the degree of decrease in current edge intensity relative to the reference edge intensity. The attenuation is then normalized to obtain the smoke occlusion rate, ensuring it falls within a range suitable for threshold comparison. The occlusion rate can be fused using a weighted sum to utilize both types of evidence simultaneously, or by maximizing the value so that a significant attenuation of one type triggers a higher occlusion judgment, thus adapting to different smoke behaviors such as "contrast reduction first" or "edge disappearance first." Finally, to suppress jitter caused by video noise, automatic exposure fine-tuning, or instantaneous occlusion, this embodiment applies a sliding mean or median filter to the smoke occlusion rate, making it better reflect the continuous smoke occlusion trend and providing stable input for subsequent judgment of common combustibles and the "low occlusion" condition in the initial screening of chemicals.

[0042] The technical solution implemented in this embodiment can jointly characterize the impact of smoke on video visibility through the dual indicators of contrast attenuation and edge attenuation, obtaining an occlusion rate that more closely approximates the actual smoke obscuration level at the ignition point within the suspected fire area and the neighborhood of the triggering time. Compared to approaches that rely solely on brightness changes or single visual features, this embodiment can reduce the interference of illumination fluctuations, automatic camera exposure, background texture differences, and short-term occlusion on occlusion determination, thereby improving the stability and transferability of smoke occlusion rate features. This provides a more reliable quantitative basis for subsequent fire type determination (such as the identification of ordinary combustibles participating in combustion, and the "high temperature and low occlusion" criterion in the initial screening of chemical fires), thereby reducing misjudgments and improving the credibility of early warning linkage decisions.

[0043] Furthermore, the blackness of the smoke color is calculated as follows: Select at least one frame of image within a reference time window before the triggering time in the suspected fire area as a reference image, and acquire the current image within a preset time window centered on the triggering time. Determine the smoke candidate area in the suspected fire area and mask the bright flame area.

[0044] Within the candidate smoke region, calculate the brightness statistics, which are the mean, quantiles, or a combination of both for the grayscale / luminance channels. Also calculate the dark pixel percentage, which is the proportion of pixels with brightness below a preset brightness threshold.

[0045] The blackness index is obtained by normalizing the change in luminance statistics relative to the corresponding luminance statistics in the reference image and the proportion of dark pixels. The blackness index is the weighted sum of the change and the proportion of dark pixels, or the maximum value of the sum. The blackness index is then filtered using a moving average or median to suppress illumination fluctuations and instantaneous jitter, and the filtered blackness index is used as the blackness feature of the smoke color.

[0046] In this embodiment, it should be noted that the "smoke color blackness" is not a simple threshold judgment based on the color value of the smoke area. Instead, it is based on the application characteristics of chemical storage sites. Considering that ordinary combustibles (such as cardboard boxes, plastics, wooden pallets, etc.) are more likely to produce "black" and "thick" smoke when burning, a blackness quantification method that is less sensitive to changes in light and strong flame interference is constructed. This method is used to subsequently determine whether ordinary combustibles participate in combustion, thereby distinguishing between mixed fires and fires involving only chemicals.

[0047] Specifically, this embodiment still adopts the relative comparison approach of "reference image + current image": at least one frame of image within a reference time window before the triggering time is selected in the suspected fire area as a reference image to characterize the background brightness level when not significantly affected by smoke; and the current image is obtained within a preset time window centered on the triggering time to reflect the brightness changes during the smoke development process. Since the flame itself will produce bright saturated areas, directly including them in the statistics would cause the blackness to be diluted by the "bright flame" or even have a reverse effect. Therefore, this embodiment first determines the smoke candidate area in the suspected fire area and masks the bright flame area, so that the blackness statistics mainly reflect the influence of the smoke itself on the brightness and color of the image, rather than the interference of the flame brightness.

[0048] In determining the candidate smoke region, candidate masks can be generated by comprehensively utilizing smoke features such as brightness changes, contrast attenuation, and edge blurring. Obvious static background areas or short-term occlusion areas can be excluded to improve the purity of the candidate region. Subsequently, brightness statistics are calculated within the candidate smoke region. The brightness statistics can be the mean of the grayscale / luminance channels to reflect the overall darkening degree; alternatively, quantiles (e.g., low quantiles or medians) can be used to reduce the influence of abnormal bright spots or localized reflections. When both stability and sensitivity need to be considered, the mean and quantiles can be combined as the brightness statistics. In addition to brightness statistics, this embodiment also calculates the proportion of dark pixels to reflect the proportion of "sufficiently dark pixels" in the smoke region; that is, the darker and thicker the smoke, the higher the proportion of dark pixels is usually.

[0049] To enhance adaptability to different backgrounds and exposure conditions, this embodiment compares the brightness statistics in the current image with the corresponding brightness statistics in the reference image to obtain the relative change, and normalizes it together with the proportion of dark pixels to obtain a blackness index. The blackness index can be fused using a weighted sum to simultaneously consider both "overall darkening" and "dark pixel enrichment" evidence; alternatively, the maximum value of both can be taken to accommodate different scenarios where smoke first shows localized blackening or an overall decrease in brightness. Finally, to suppress short-term fluctuations caused by automatic camera exposure, ambient lighting flicker, or momentary occlusion, this embodiment applies a moving average or median filter to the blackness index, outputting smoother smoke color blackness features for subsequent fire type determination and mixed fire identification.

[0050] The technical solution implemented in this embodiment can obtain smoke color and blackness features that are less sensitive to strong flame interference and ambient light fluctuations within the vicinity of the suspected fire area and the triggering time by combining "smoke candidate area extraction + flame high-brightness shielding + relative brightness change + dark pixel ratio". Compared with the approach of directly using color values ​​or a single brightness threshold, this embodiment can reduce the impact of high-brightness flame areas, local reflections, and automatic camera exposure on blackness judgment, and improve the recognition stability of dark smoke features produced by the combustion of ordinary combustibles. This provides more reliable evidence to distinguish between "chemical fire" and "mixed fire", reduces misjudgment, and improves the credibility of linkage control decisions.

[0051] Furthermore, the smoke concentration growth rate and / or VOC growth rate are calculated as follows: Within a preset time window centered on the trigger time, smoke concentration sequence and VOC concentration sequence are acquired respectively, and the growth rate is calculated within a sliding calculation window. The growth rate includes: the difference mean growth rate obtained by performing first-order difference on adjacent sampling points and taking the mean, and / or, the fitting slope growth rate obtained by fitting the data within the sliding calculation window.

[0052] The growth rate is filtered by moving average or median to suppress instantaneous disturbances, and the filtered growth rate is compared with a preset growth rate threshold.

[0053] In this embodiment, it should be noted that the "growth rate" does not simply use the instantaneous difference value at a single moment, but rather quantifies the trend of the smoke concentration sequence and VOC concentration sequence within a preset time window around the trigger moment, in order to more stably reflect the dynamic characteristic of "continuous increase" that is closely related to the evolution of the fire. Since the ignition or leakage of chemical products often manifests as a rapid increase in VOCs, while the combustion of ordinary combustibles often manifests as a continuous increase in smoke concentration, this embodiment uses the calculation and filtering of the growth rate to ensure that this dynamic characteristic can be reliably used in key stages such as the initial screening of chemical products and the determination of mixed ignition.

[0054] Specifically, in this embodiment, smoke concentration sequences and VOC concentration sequences are acquired within a preset time window centered on the trigger time, and a sliding calculation window is set within this time window. The sliding calculation window is used to repeatedly calculate the growth rate within a continuous time period, so that the growth rate is updated over time, thereby capturing the change process of "from slow rise to rapid jump" or "short fluctuation followed by continuous climb". The growth rate can be calculated using two complementary methods: one is the differential mean growth rate, which involves performing a first-order difference on the concentration changes of adjacent sampling points within the sliding calculation window and averaging them to obtain a growth rate estimate that is not sensitive to local fluctuations but is more sensitive to the overall upward trend; the other is the fitted slope growth rate, which involves fitting the trend of concentration changes over time within the sliding calculation window, and the slope obtained by fitting is used to characterize the upward intensity within the window, so that a relatively stable trend judgment can still be obtained even when there is a lot of noise or uneven sampling intervals. Either of the above two types of growth rates can be used, or they can be calculated simultaneously and the result that better matches the characteristics of the field data can be selected in subsequent judgments.

[0055] To further suppress transient sensor disturbances, sampling jitter, and spikes / dips caused by local ventilation, this embodiment applies a moving average filter or a median filter to the obtained growth rate sequence. Moving average filtering helps smooth high-frequency noise and highlight the overall trend; median filtering helps suppress occasional spikes that could cause false triggering in the judgment. The filtered growth rate is then compared with a preset growth rate threshold to support initial screening conditions for chemical ignition (e.g., whether the VOC growth rate abnormally increases) and the determination of combustion involving common combustibles (e.g., whether the smoke concentration growth rate continuously exceeds the threshold). Through the combined process of "time window selection—moving calculation—filtering and noise suppression—threshold comparison," this embodiment enables the growth rate to possess both sensitivity and sufficient stability, making it suitable for long-term operation in practical engineering environments.

[0056] The technical solution implemented in this embodiment can stably quantify the upward trend of smoke concentration and VOC concentration in the neighborhood at the trigger time. By using two growth rate estimation methods—sliding calculation window and differential mean / fitting slope—it improves the ability to capture continuous growth characteristics. Furthermore, filtering reduces the impact of sensor noise and instantaneous spikes caused by ventilation disturbances on the judgment. As a result, the growth rate characteristics can be used more reliably for the initial screening of chemical fires and the identification of mixed fires, reducing false triggers and missed triggers, improving the stability and reliability of fire type determination, and providing more accurate input basis for subsequent early warning and linkage control strategies.

[0057] Furthermore, the liquid spreading characteristics are calculated as follows: Within the suspected fire area, a pre-defined sub-area of ​​the ground / countertop is designated as the liquid detection area. At least one reference image is selected within a baseline time window prior to the trigger moment, and the current image is acquired within a pre-defined time window centered on the trigger moment.

[0058] A liquid candidate change map is calculated within the liquid detection region. The liquid candidate change map includes at least one of the following: a reflection enhancement change obtained based on the brightness difference between the current image and the reference image, and / or a texture change obtained based on the local texture energy change between the current image and the reference image.

[0059] The candidate liquid regions are obtained by threshold segmentation and connected component screening of the candidate liquid change map, and the area change, expansion rate and / or boundary continuity index of the candidate liquid regions are calculated within a preset time window.

[0060] The liquid spreading characteristics are obtained by normalizing and combining the indicators of area change, expansion rate, and / or boundary continuity, and then applying a moving average or median filter to the liquid spreading characteristics. When the liquid spreading characteristics exceed the liquid spreading threshold, the liquid spreading determination condition in S33 is met.

[0061] In this embodiment, it should be noted that the "liquid spreading feature" is used to characterize the leakage, flow, and spreading phenomena that may occur in chemicals (especially flammable liquids such as alcohol) before and after ignition. Considering that liquid spreading usually occurs first on surfaces such as floors and countertops, and will appear in video as enhanced reflection, changes in texture after wetting, and expansion of the area outline over time, this embodiment uses a method of "limiting the detection area - constructing candidate changes - extracting candidate areas - quantifying the spreading process" to form the liquid spreading feature, as important visual evidence of chemical involvement, and for cross-validation with VOC changes.

[0062] Specifically, this embodiment first identifies a pre-defined sub-area of ​​the ground / countertop within the suspected fire area as the liquid detection area. This setting is to avoid including areas such as walls, shelf facades, and personnel passageways, which are not prone to stable spreading, in liquid recognition, and to reduce false detections caused by non-liquid factors such as mirror reflections and shadow swaying. The liquid detection area can be pre-defined during the deployment phase through scene calibration, or it can be automatically determined based on the image structure (e.g., horizontal plane, ground segmentation results). Subsequently, at least one reference image is selected within a baseline time window before the triggering moment to characterize the background state before spreading occurs, and the current image is continuously acquired within a pre-defined time window centered on the triggering moment to capture the appearance evolution during the spreading process.

[0063] In constructing candidate variation maps, this embodiment employs at least one of the following variation representations: First, reflection enhancement variation, which uses the difference in brightness between the current image and the reference image to depict phenomena such as enhanced reflection and bright spot diffusion that may occur after liquid wetting; second, texture variation, which uses the difference in local texture energy between the current image and the reference image to depict phenomena such as reduced texture contrast, "smoothing" of details, or changes in texture distribution caused by liquid coverage. These variation maps can be used individually or in combination to adapt to differences in liquid performance under different ground materials and lighting conditions.

[0064] After obtaining the candidate liquid change map, this embodiment performs threshold segmentation on the candidate change map to extract significant change areas as preliminary liquid candidate regions. Furthermore, it filters out noisy regions with excessively small areas, abnormal shapes, or unreasonable locations through connected component screening, retaining connected regions that better conform to the "liquid spreading" pattern. Subsequently, within a preset time window, the liquid candidate regions are quantified, calculating at least one or a combination of the following: region area change (reflecting the expansion of the spreading range), expansion speed (reflecting the spreading growth rate), and boundary continuity index (reflecting whether the spreading boundary shows continuous outward expansion rather than discrete jumps). Through these quantification indicators, this embodiment further elevates "seeing liquid sample changes" to dynamic evidence of "liquid spreading," avoiding misjudgments caused by relying solely on single-frame reflections or localized stain changes.

[0065] Finally, this embodiment normalizes and combines the area change, expansion rate, and / or boundary continuity indices to obtain liquid spreading characteristics, ensuring they fall within a range suitable for threshold comparison. The liquid spreading characteristics are then filtered using a moving average or median to suppress jitter caused by automatic camera exposure fine-tuning, short-term human occlusion, and instantaneous reflection disturbances. When the liquid spreading characteristics exceed the liquid spreading threshold, significant evidence of liquid spreading is considered to exist in the neighborhood at the trigger time, thus satisfying the liquid spreading determination condition in step S33 and providing reliable input for subsequent gas-liquid consistency verification.

[0066] The technical solution implemented in this embodiment can reduce false detections caused by reflections from non-contact surfaces and scene interference by defining a sub-region of the ground / tabletop as the liquid detection area; and by constructing a liquid candidate change map with enhanced reflection and / or texture changes, combined with threshold segmentation and connected component screening to extract candidate regions that better match the liquid morphology, and then dynamically quantifying the spreading process by area change, expansion speed, and boundary continuity, thereby obtaining liquid spreading characteristics that are more robust to material differences and lighting changes; at the same time, filtering and smoothing suppress instantaneous disturbances, making the liquid spreading determination more stable and reliable, thus providing key visual evidence for chemical fire identification, reducing the risk of misjudging ordinary reflections or short-term changes as liquid leakage spreading, and improving the accuracy and reliability of fire type determination.

[0067] Furthermore, the gas-liquid consistency is calculated as follows: Within a preset time window centered on the trigger moment, acquire the liquid spreading characteristic sequence and the VOC growth rate sequence, and perform consistency quantification on the two. The consistency quantification includes at least one of the following: Proportion of changes in the same direction: The percentage of sampling points where the trend of changes in the liquid spreading characteristics changes in the same direction as the VOC growth rate.

[0068] Time-series correlation: Calculate the correlation between liquid spreading characteristics and VOC growth rate within the allowable time offset range and take the maximum correlation.

[0069] The consistency quantification result is compared with the consistency threshold. When the consistency quantification result exceeds the consistency threshold, the consistency judgment condition in step S34 is met.

[0070] In this embodiment, it should be noted that the purpose of introducing "gas-liquid consistency" is to further determine whether the spreading process corroborates the abnormal rise in VOCs in time, based on the established "suspected liquid spreading evidence." This establishes a correlation between the "changes in the appearance of the liquid seen in the video" and the "changes in volatile organic compounds detected by the gas sensor," forming a more complete chain of evidence. In chemical storage areas, ground reflections, cleaning marks, water stains, and shifting shadows can all create visual illusions similar to "changes in liquid areas." Simultaneously, VOCs can fluctuate due to short-term ventilation disturbances, door openings, and handling. Therefore, relying on a single path can easily lead to misjudgment. This embodiment uses consistency quantification, employing the "synchronicity" and "correlation" of two signals as key constraints to improve the reliability of determining the combustion / leakage combustion of chemicals.

[0071] Specifically, this embodiment acquires both the liquid spreading characteristic sequence and the VOC growth rate sequence within a preset time window centered on the trigger moment. This time window is designed to cover adjacent processes that may occur between "liquid leakage / spreading—enhanced volatilization—ignition / combustion triggering," allowing for comparative analysis of the two sequences on the same time scale. Considering the response inertia of VOC sensors, differences in sampling periods, and propagation delays in gas diffusion, this embodiment introduces an "allowable time offset range" in the consistency analysis. This allows the VOC growth rate to lag or advance relative to the liquid spreading characteristic within a certain time offset, accommodating temporal differences under different sampling distances and ventilation conditions, thereby avoiding missed detections due to strict simultaneous comparisons.

[0072] Consistency quantification can be achieved in at least one way: One method is the proportion of changes in the same direction, which involves statistically analyzing whether the changing trends of liquid spreading characteristics and VOC growth rate are in the same direction within a preset time window, and calculating the proportion of sampling points where the same direction occurs. This method focuses more on "consistent trends" rather than absolute numerical values, and is suitable for scenarios where liquid spreading gradually increases and VOC gradually increases. When the proportion of changes in the same direction is high, it indicates that the increase in liquid spreading and the abnormal rise in VOC support each other in the overall trend.

[0073] The second method is temporal correlation, which calculates the correlation between liquid spreading characteristics and VOC growth rate within an allowable time offset range, and takes the offset with the highest correlation as the final temporal correlation result. This method focuses more on the correlation between "morphological similarity and time alignment", and is suitable for scenarios with significant sensor lag or a fixed delay between VOC changes and liquid spreading. By taking the maximum correlation, the most likely time alignment relationship can be automatically matched within a certain range, improving adaptability to actual working conditions.

[0074] After obtaining the consistency quantification result, this embodiment compares the result with a preset consistency threshold: when the consistency quantification result exceeds the consistency threshold, it is considered that the change in liquid spreading and the change in VOC growth rate meet the consistency conditions, thereby meeting the consistency judgment requirements in step S34 and supporting the conclusion of "chemical products involved"; if the consistency does not reach the threshold, it indicates that there is a mismatch between the two chains of evidence, which may be a visual fake liquid or a non-fire-related fluctuation in VOC. A more conservative linkage strategy or manual review path should be adopted to reduce the risk of improper linkage caused by misjudgment.

[0075] The technical solution implemented in this embodiment can further introduce temporal consistency verification of gas evidence after the evidence of liquid spreading is established. By cross-verifying the "liquid spreading - VOC anomaly" through the temporal correlation of the same direction change ratio and / or the allowable time offset, the misjudgment caused by relying solely on changes in video reflection or solely on instantaneous fluctuations in VOC can be reduced. At the same time, by setting the allowable time offset range, it can adapt to actual engineering conditions such as sensor response lag, differences in deployment distance, and ventilation diffusion delay, reducing missed judgments. This improves the reliability and stability of chemical fire identification, provides a more credible trigger basis for subsequent linkage control strategies (such as water ban, alternative fire extinguishing method prompts, etc.), and reduces secondary risks such as fire spread caused by improper handling.

[0076] Furthermore, after an output of a mixed fire or a chemical fire, it also includes generating and outputting a fire extinguishing and coordinated control strategy that matches the type of fire, specifically including: When the output indicates a chemical fire or a mixed fire, a water-based fire extinguishing disabling warning will be issued, and a water system linkage suppression command will be sent to the fire alarm control equipment. The water system linkage suppression command shall at least include interception, delayed execution, and / or manual confirmation of the start command for sprinkler and / or fire hydrant water supply.

[0077] When the output indicates a fire involving chemicals or a mixture of chemicals, an alternative extinguishing method prompt and / or a corresponding linkage command will be output. The alternative extinguishing method shall include at least one or more of the following: foam extinguishing, dry powder extinguishing, carbon dioxide extinguishing, and / or inert gas suppression.

[0078] When the output indicates a mixed fire, a phased response prompt is provided. The phased response prompt includes at least covering and isolating or suppressing the combustion of flammable liquids first, and then dealing with the remaining embers of ordinary flammable materials.

[0079] In this embodiment, it should be noted that after outputting the fire type, the fire type is directly mapped to a matching fire extinguishing and linkage control strategy to address the risk of improper handling caused by the "only alarm without type classification and single linkage strategy" in the background technology. Especially in chemical storage areas adjacent to residential areas, if a chemical product (such as flammable liquids like ethanol) catches fire and is mistakenly treated as a normal fire by directly activating water sprinklers or using water-based fire extinguishing, the flammable liquid may be washed away and spread, the fire may expand, or even cause secondary hazards. Therefore, this embodiment uses "water prohibition prompts and water system linkage suppression" as the default safety strategy for chemical product fires and mixed fires, and outputs alternative fire extinguishing method prompts or linkage commands based on this. For mixed fires, it further outputs phased handling prompts to guide on-site handling in a more reasonable order.

[0080] Specifically, when the fire type output is a chemical fire or a mixed fire, this embodiment generates a water-based fire extinguishing disabling prompt, which is used to clearly indicate "Do not use water directly to extinguish the fire / Caution when starting the sprinkler system" on the audible and visual alarm interface, duty terminal, or mobile terminal. Simultaneously, this embodiment outputs a water system linkage suppression command to the fire linkage control equipment. This suppression command is not a simple "shutdown of sprinklers," but rather controls the activation action in the linkage link. For example, it intercepts the activation command of the sprinkler system and / or fire hydrant water supply, preventing it from being automatically executed; or it sets a delay window for the execution of the activation command to allow time for manual confirmation and on-site verification; or it switches the activation command to a manual confirmation mode, allowing execution only after confirmation by the duty officer. By suppressing this critical node of the "activation command," this embodiment can achieve risk control of improper water system linkage without disrupting the original fire linkage system architecture.

[0081] Regarding alternative fire extinguishing methods, when the fire is caused by a chemical fire or a mixed fire, this embodiment outputs an alternative fire extinguishing method prompt and / or a corresponding linkage command. Alternative fire extinguishing methods include at least one or more of foam extinguishing, dry powder extinguishing, carbon dioxide extinguishing, and inert gas suppression, which can be selected and combined according to the capabilities of the fire extinguishing system already configured at the location. For example, if the location is equipped with a foam extinguishing system, foam coverage will be prioritized to isolate flammable liquids; if a gas extinguishing system is configured, inert gas / carbon dioxide suppression will be prompted or linked to reduce combustion conditions; if only portable equipment is configured, a dry powder / carbon dioxide fire extinguisher disposal prompt will be output. These prompts or linkage commands ensure that the disposal measures are matched to the nature of the fire, thereby reducing the risk of spread due to the selection of the wrong extinguishing medium.

[0082] Furthermore, when the output indicates a mixed fire, this embodiment provides a phased response prompt. The core principle is that a mixed fire simultaneously involves the combustion of flammable liquids and ordinary combustibles. If the response order is not differentiated, spraying or dousing ordinary combustibles may promote the spread of the liquid, potentially exacerbating the fire. Therefore, this embodiment at least prompts that the flammable liquid combustion should first be covered, isolated, or suppressed to control the combustion and reduce the possibility of further spread. After the liquid combustion is suppressed, the remaining embers of the ordinary combustibles should then be dealt with to reduce the risk of reignition and spread. This phased prompt can be used to guide manual response or to prioritize coordinated strategies (e.g., triggering foam covering / gas suppression first, then allowing subsequent treatment of solid embers).

[0083] The technical solution implemented in this embodiment can automatically output a response and linkage strategy matching the fire type after identifying a chemical fire or a mixed fire. It reduces the secondary risks of flammable liquid spread and fire propagation caused by improper water spraying by using water-based fire suppression warnings and water system linkage suppression commands. Simultaneously, through alternative fire suppression method warnings and / or linkage commands, the system can guide or trigger more suitable fire suppression methods for chemical fires, improving the targeting and safety of the response. For mixed fires, phased response warnings guide the system to control liquids first, then solids, reducing the risk of escalation due to incorrect response sequence, thereby improving the safety boundaries and emergency response efficiency of fire-fighting linkage in chemical storage sites.

[0084] Furthermore, this also includes implementing conservative, coordinated control measures for uncertain branches where chemical fires are suspected, specifically including: Following S33 are: If the suspected liquid area features do not exceed the spreading threshold, a suspected chemical fire warning will be output, and a water system linkage suppression command will be generated and issued.

[0085] Following S34 are: If the consistency condition is not met, a suspected chemical fire warning will be output, and a water system linkage suppression command will be generated and issued.

[0086] Among them, the water system linkage suppression command includes at least intercepting, delaying the execution of the start command for sprinkler and / or fire hydrant water supply, and / or manually confirming the execution.

[0087] Furthermore, after issuing a suspected chemical fire alert, the water-based fire suppression command will continue to be executed until one of the following conditions is met for deactivation: manual verification; or the chemical fire determination conditions are met in subsequent monitoring, and the response and control strategy is activated; or the preset chemical risk deactivation conditions are met in subsequent monitoring.

[0088] In this embodiment, it should be noted that a conservative linkage control mechanism with uncertain branches is set up for the situation of "suspected chemical fire but incomplete evidence" to address the high-risk problem of "accidental activation of water systems leading to the spread of flammable liquids and the expansion of the fire" in chemical storage sites. Since the chain of determining chemical fire includes initial screening triggering, evidence of liquid spreading, and evidence of gas-liquid consistency, the following situations may occur at the engineering site: On the one hand, the initial screening conditions for chemicals have been triggered, but liquid spreading characteristics are not consistently detected in the video (e.g., the liquid has burned out, the spreading area is obscured, or the ground material makes the spreading appearance unclear); on the other hand, liquid spreading characteristics exist but are not sufficiently consistent with VOC growth changes (e.g., the VOC sensor is far away, ventilation dilution causes a response lag, or the local gas diffusion path is complex). If the sprinkler or water supply is automatically activated as in a normal fire situation, there is a possibility of amplifying the risk. Therefore, in this embodiment, if S33 or S34 is not met, the chemical risk is not directly denied, but the conservative linkage path for "suspected chemical fire" is initiated.

[0089] Specifically, after step S33, this embodiment sets up a first type of uncertain branch: when the characteristics of the suspected liquid area do not exceed the spreading threshold, the system outputs a suspected chemical fire warning and generates and issues a water-based linkage suppression command. The meaning of this branch is that, given that a preliminary chemical screening has occurred and the suspected fire area and trigger time have been located, even if the evidence of liquid spreading is insufficient, the possibility of chemical fire (such as alcohol) cannot be ruled out, especially in the early stages of liquid combustion or when the liquid is obscured, where "insufficient visual evidence" is more likely to occur. Therefore, by using a warning + water-based suppression linkage, time is gained for manual review or subsequent evidence supplementation, and secondary hazards caused by automatic water spraying are avoided.

[0090] Following step S34, this embodiment sets a second type of uncertain branch: when the consistency condition is not met, the system also outputs a suspected chemical fire warning and generates and issues a water system linkage suppression command. This branch means that when evidence of liquid spreading exists but gas-liquid consistency is insufficient, it indicates that the two types of evidence have not yet mutually corroborated each other. This may be due to sensor lag, diffusion delay, or noise disturbance causing the consistency to fall short of the threshold. In this case, hastily releasing the water system linkage could also pose a risk; therefore, a conservative strategy is adopted, and further confirmation from subsequent monitoring data or manual verification is required.

[0091] At the execution level of water system linkage suppression commands, this embodiment uses the "control start command" as the key node for suppression, including at least: intercepting the start command of the sprinkler system and / or fire hydrant water supply to prevent it from being executed automatically; or setting a delay for the execution of the start command, allowing the system to continuously acquire subsequent evidence within the delay window and prompting on-duty personnel for verification; or switching the start link to a manual confirmation execution mode, allowing execution only after confirmation by on-duty personnel. Through the above methods, this embodiment can achieve safety constraints on high-risk water system linkages without changing the existing fire linkage hardware structure of the site.

[0092] Furthermore, to avoid other risks caused by prolonged conservative suppression, this embodiment also includes a suppression release mechanism: after outputting a suspected chemical fire warning, the water-based linkage suppression command is maintained until one of the following release conditions is met: First, manual verification release, i.e., the suppression is released after the on-duty personnel or on-site personnel confirm that there is no risk of chemical fire; Second, subsequent monitoring meets the chemical fire determination conditions, the system upgrades from "suspected" to "confirmed," and switches to a handling and linkage control strategy matching chemical fire or mixed fire; Third, subsequent monitoring meets preset chemical risk release conditions, such as VOC concentration returning to the background stable range and no longer showing an upward trend, suspected liquid spreading characteristics remaining below the threshold, and no continuous flame / temperature anomalies in the suspected fire area, etc., thereby determining that the chemical risk has decreased and the suppression is released. The above release conditions ensure that the conservative strategy can effectively control risks while also being withdrawable, avoiding excessive suppression from affecting normal fire response.

[0093] The technical solution implemented in this embodiment can reduce the secondary risk of flammable liquid spreading and fire spread due to the misoperation of the automatic water spraying function when the evidence chain of chemical fire is not completely closed. This is achieved by adopting a conservative linkage control of "suspected warning + water system linkage suppression" for uncertain branches. At the same time, by setting corresponding trigger points for the two uncertain situations of "insufficient evidence of liquid spreading" and "insufficient gas-liquid consistency", the conservative strategy is tightly coupled with the judgment chain, reducing the window for false linkage. Furthermore, through three exit mechanisms—manual review, subsequent evidence supplementation and upgrading, and chemical risk removal conditions—the system maintains good engineering availability and response efficiency while ensuring safety boundaries, thereby improving the reliability of early warning linkage and the safety of disposal in chemical storage sites.

[0094] The present invention also provides a building fire safety monitoring and early warning management system, wherein the building fire safety monitoring and early warning management method applying any of the above includes: The multi-source sensing module is used to collect smoke data, temperature data, VOC concentration data and video data, and output multi-source raw monitoring streams.

[0095] The data processing module, whose input end is connected to the output end of the multi-source sensing module, is used to synchronize the original monitoring streams from multiple sources in time to generate time-series data units, and to construct a fire discrimination feature set based on the time-series data units.

[0096] The serial discrimination module, whose input is connected to the output of the data processing module, is used to perform serial fusion judgment based on the fire discrimination feature set to output the fire type.

[0097] The strategy output module, whose input is connected to the output of the serial discrimination module, is used to output the fire type and generate and issue early warning information and / or linkage control commands that match the fire type.

[0098] In this embodiment, it should be noted that the building fire safety monitoring and early warning management system provided is not simply a stacking and connection of several sensors and alarms. Instead, it is a system architecture with a clear data flow and decision-making chain, built around the goal of "distinguishing between chemical fires and household fires and outputting matching linkage strategies." The system uses a multi-source sensing module as the entry point, which processes the data to form a comparable and fusionable unified time-series input. Then, the serial discrimination module outputs the fire type according to the serialized fusion logic of "first trigger, then focus, then verify." Finally, the strategy output module maps the fire type into early warning prompts and linkage control commands, realizing a closed loop from perception to decision-making to execution.

[0099] Specifically, the multi-source sensing module is used to simultaneously acquire smoke data, temperature data, VOC concentration data, and video data within the target area, and output a multi-source raw monitoring stream. The smoke data can come from smoke detectors or optical smoke sensors, the temperature data from temperature detectors or temperature arrays, the VOC concentration data from VOC gas sensors, and the video data from surveillance cameras or smart camera terminals facing the storage area. The output of the multi-source sensing module can be a timestamped data stream, enabling unified management and alignment of data from different sources in subsequent stages.

[0100] The input of the data processing module is connected to the output of the multi-source sensing module. It synchronizes the original monitoring streams from multiple sources to generate time-series data units, and then constructs a fire discrimination feature set on these units. Time synchronization can include resampling, aligning, and padding data from different sampling periods, allowing smoke, temperature, VOC, and video information at the same time point to be used in the determination. The construction of the fire discrimination feature set transforms the original data into feature representations for fire differentiation. Examples include extracting smoke concentration and its growth trend, obtaining smoke obscuration and smoke blackness from video, combining temperature and flame detection to obtain the equivalent flame temperature, obtaining concentration anomalies and their growth trends from VOC sequences, and extracting the spreading features of suspected liquid areas from video. Through this module's processing, the subsequent discrimination module obtains more stable and discriminative feature inputs, rather than the noisy original signal.

[0101] The input of the serial discrimination module is connected to the output of the data processing module. It is used to perform serial fusion judgment based on the fire discrimination feature set to output the fire type. This module embodies the core decision-making logic of this application: instead of simply parallelizing multiple features, it employs a serial gating and step-by-step verification mechanism. For example, it first triggers a preliminary screening for chemical fires based on conditions such as abnormal VOCs or high temperature and low shading; after triggering, it determines the trigger time and suspected fire area, allowing subsequent calculations to focus on key areas and key time windows; then, it confirms suspected liquid spreading signs within the suspected fire area, and further uses the consistency between changes in liquid spreading and changes in VOC growth as a cross-verification condition; thus, it outputs "fire includes chemical fire," "chemical fire," "mixed fire," or related branch results. Through this serial discrimination structure, the system can reduce misjudgments in complex scenarios and provide more reliable type conclusions for strategy output.

[0102] The input of the strategy output module is connected to the output of the serial discrimination module. It outputs the fire type and generates and issues warning information and / or linkage control commands matching the fire type. Warning information may include categorized prompts for duty terminals, audible and visual alarms, or mobile devices, such as "suspected / confirmed chemical fire" or "water-based fire extinguishing prohibited." Linkage control commands may include suppression, delay, or manual confirmation of water-based fire suppression, as well as prompts / controls for alternative fire extinguishing methods or related equipment. Through the strategy output module, the system transforms the identification results into actionable recommendations or control actions, thus avoiding the risk of inappropriate handling caused by traditional systems that indiscriminately activate water sprinklers when the fire type is unclear.

[0103] The technical solution implemented in this embodiment can unify the organization of smoke, temperature, VOC, and video information and form a progressively converging judgment chain through a modular data flow structure of "multi-source perception - temporal alignment and feature construction - serial fusion discrimination - strategy output," thereby improving the ability to distinguish between chemical fires and household fires and enhancing the stability of the judgment. At the same time, it can directly map the fire type to a matching early warning prompt and linkage control command. Especially in the scenario of chemical fires or suspected chemical fires, by suppressing, delaying, or manually confirming water system linkage, it can reduce the secondary risk of flammable liquid spread and fire spread caused by automatic water spraying. This improves the safety boundary and engineering availability of fire monitoring, early warning, and linkage response for chemical storage sites adjacent to living areas.

[0104] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A building fire safety monitoring and early warning management method applied to a chemical storage site adjacent to a living area, for distinguishing between a living supply fire and a chemical supply fire, characterized in that, include: S10. Collect smoke, temperature, VOC concentration and video data in the target area, perform time-series alignment, and generate multi-source data units; S20. Extract a discrimination feature set from the multi-source data unit, including: smoke concentration and growth rate, smoke obstruction rate, flame equivalent temperature, VOC concentration and growth rate, and suspected liquid area features based on video. S30. Based on the discriminative feature set, perform serialization fusion determination: S31. Determine whether the initial screening conditions for chemical products ignition are met: the VOC concentration or its growth rate exceeds the corresponding threshold, or the flame equivalent temperature exceeds the high temperature threshold and the smoke obscuration rate is lower than the low obscuration threshold; if not, output the determination of household products ignition. S32. If so, determine the triggering time and the corresponding suspected fire area; S33. Determine whether the suspected liquid area feature in the suspected fire area exceeds the spreading threshold within the time window before and after the triggering time. S34. If so, determine whether the changes in the characteristics of the suspected liquid area and the changes in the VOC growth rate meet the consistency condition. S35. If so, output a determination that the fire situation includes chemical products on fire.

2. The building fire safety monitoring and early warning management method according to claim 1, characterized in that, S35 and above further includes: under the premise that it is determined that there are chemical products on fire in the fire, determining whether there are ordinary combustibles participating in the combustion at the same time, and distinguishing between mixed fire and chemical product fire accordingly; The determination of whether the common combustible material participates in combustion is made within a preset time window centered on the triggering time within the suspected ignition area, and satisfies any of the following determination rules: The smoke obscuration rate exceeds a preset optical obscuration threshold and the smoke color blackness exceeds a preset blackness index threshold, and continues to meet a preset duration threshold; or The smoke concentration growth rate exceeds a preset smoke growth rate threshold, and the smoke obstruction rate remains above the optical obstruction threshold for a preset duration; When the condition that the common combustible material participates in combustion is met, the output is "mixed ignition"; otherwise, the output is "chemical ignition".

3. The building fire safety monitoring and early warning management method according to claim 2, characterized in that, The smoke obstruction rate is calculated as follows: In the suspected fire area, at least one frame image in a reference time window before the triggering time is taken as a reference image I ref , and a current image I is acquired in a preset time window centered on the triggering time t , respectively in the suspected fire area The brightness contrast index C is the standard deviation of the brightness channel, the local contrast statistic, or a combination of the two. Edge intensity index E, wherein the edge intensity index is gradient magnitude energy, edge pixel ratio, or a combination of the two; and the contrast attenuation AC = 1 C(I t ) / C(I ref ) and / or edge attenuation ΔE = 1 E(I t ) / E(I ref ), the ΔC and / or ΔE are normalized to obtain a smoke obscuration ratio Occ(t), where Occ(t) is a weighted sum or a maximum of ΔC and ΔE; the Occ(t) is sliding mean or median filtered to suppress transient jitter.

4. The building fire safety monitoring and early warning management method according to claim 3, characterized in that, The darkness of the smoke color is calculated as follows: In the suspected fire area, at least one frame of image within a reference time window before the triggering time is selected as a reference image, and the current image is obtained within a preset time window centered on the triggering time. Smoke candidate areas are determined in the suspected fire area, and the bright flame areas are masked. Calculate a brightness statistic within the smoke candidate region, wherein the brightness statistic is the mean, quantile, or a combination of the grayscale / brightness channels; And calculate the dark pixel ratio, which is the percentage of pixels whose brightness is lower than a preset brightness threshold; The blackness index is obtained by normalizing the change of the brightness statistic relative to the corresponding brightness statistic of the reference image and the proportion of dark pixels. The blackness index is the weighted sum of the change and the proportion of dark pixels or the maximum value of the sum. The blackness index is then subjected to moving average or median filtering to suppress illumination fluctuations and instantaneous jitter, and the filtered blackness index is used as the blackness feature of the smoke color.

5. The building fire safety monitoring and early warning management method according to claim 4, characterized in that, The smoke concentration growth rate and / or VOC growth rate are calculated as follows: Within a preset time window centered on the trigger time, smoke concentration sequence and VOC concentration sequence are acquired respectively, and the growth rate is calculated within a sliding calculation window. The growth rate includes: the difference mean growth rate obtained by performing first-order difference on adjacent sampling points and taking the mean, and / or, the fitting slope growth rate obtained by fitting the data within the sliding calculation window. The growth rate is filtered by moving average or median to suppress instantaneous disturbances, and the filtered growth rate is compared with a preset growth rate threshold.

6. The building fire safety monitoring and early warning management method according to claim 5, characterized in that, The liquid spreading characteristics are calculated as follows: Within the suspected fire area, a preset ground / countertop sub-area is determined as the liquid detection area; at least one frame of reference image is selected within a reference time window before the triggering time, and the current image is acquired within a preset time window centered on the triggering time; A liquid candidate change map is calculated within the liquid detection area. The liquid candidate change map includes at least one of the following: a reflection enhancement change obtained based on the brightness difference between the current image and the reference image, and / or a texture change obtained based on the local texture energy change between the current image and the reference image. The liquid candidate change map is segmented by threshold and filtered by connected components to obtain liquid candidate regions, and the area change, expansion rate and / or boundary continuity index of the liquid candidate regions are calculated within the preset time window. The area change, expansion rate and / or boundary continuity indices are normalized and combined to obtain liquid spreading characteristics, and the liquid spreading characteristics are subjected to moving average or median filtering; when the liquid spreading characteristics exceed the liquid spreading threshold, the liquid spreading determination condition in S33 is satisfied.

7. The building fire safety monitoring and early warning management method according to claim 6, characterized in that, The gas-liquid consistency is calculated as follows: Within a preset time window centered on the trigger moment, a liquid spreading characteristic sequence and a VOC growth rate sequence are acquired, and their consistency is quantified. The consistency quantification includes at least one of the following: Proportion of changes in the same direction: The percentage of sampling points where the trend of changes in liquid spreading characteristics changes in the same direction as the VOC growth rate; Time-series correlation: Calculate the correlation between liquid spreading characteristics and VOC growth rate within the allowable time offset range and take the maximum correlation; The consistency quantification result is compared with the consistency threshold. When the consistency quantification result exceeds the consistency threshold, the consistency determination condition in step S34 is satisfied.

8. The building fire safety monitoring and early warning management method according to claim 2, characterized in that, After outputting the fire type (mixed fire or chemical fire), the system also includes generating and outputting fire extinguishing and coordinated control strategies that match the fire type, specifically including: When the output indicates a chemical fire or a mixed fire, a water-based fire extinguishing disabling warning is issued, and a water system linkage suppression command is sent to the fire alarm control equipment. The water system linkage suppression command includes at least intercepting, delaying, and / or manually confirming the start command for sprinkler and / or fire hydrant water supply. When the output indicates a fire involving chemicals or a mixture of chemicals, an alternative fire extinguishing method prompt and / or a corresponding linkage command will be output. The alternative fire extinguishing method shall include at least one or more of foam fire extinguishing, dry powder fire extinguishing, carbon dioxide fire extinguishing and / or inert gas suppression. When the output indicates a mixed fire, a phased handling prompt is provided. The phased handling prompt includes at least covering and isolating or suppressing the combustion of flammable liquids first, and then handling the remaining embers of ordinary flammable materials.

9. The building fire safety monitoring and early warning management method according to claim 8, characterized in that, This also includes implementing conservative linkage controls for uncertain branches where chemical fires are suspected, specifically including: Following S33, the following is also included: If the suspected liquid area features do not exceed the spreading threshold, a suspected chemical fire warning is output, and the water system linkage suppression command is generated and issued. Following S34, the following is also included: If the consistency condition is not met, a suspected chemical fire warning will be output, and the water system linkage suppression command will be generated and issued. The water system linkage suppression command includes at least intercepting, delaying, and / or manually confirming the execution of the start command for water supply to sprinklers and / or fire hydrants; Furthermore, after the suspected chemical fire warning is issued, the water system linkage suppression command is maintained until one of the following release conditions is met: manual verification release; or the chemical fire determination conditions are met in subsequent monitoring and the disposal and linkage control strategy is switched to; or the preset chemical risk release conditions are met in subsequent monitoring.

10. A building fire safety monitoring and early warning management system, employing the building fire safety monitoring and early warning management method as described in any one of claims 1-9, characterized in that, include: The multi-source sensing module is used to collect smoke data, temperature data, VOC concentration data and video data, and output multi-source raw monitoring streams; The data processing module, whose input end is connected to the output end of the multi-source sensing module, is used to synchronize the multi-source raw monitoring stream in time to generate time-series data units, and to construct a fire discrimination feature set based on the time-series data units; A serial discrimination module, whose input is connected to the output of the data processing module, is used to perform serial fusion judgment based on the fire discrimination feature set to output the fire type. The strategy output module, whose input is connected to the output of the serial discrimination module, is used to output the fire type and generate and issue early warning information and / or linkage control commands that match the fire type.

Images