Intelligent fire-fighting online monitoring and alarming platform and method

Abstract

The application discloses a smart fire-fighting online monitoring alarm platform and method, and relates to the technical field of online monitoring; the method comprises the following steps: acquiring image data and physical sensing signals of a monitoring area in real time, wherein the physical sensing signals comprise smoke concentration values and temperature values; performing pixel brightness analysis on the image data to obtain image visual state results of the monitoring area; the image visual state results comprise visual effective state and visual invalid state; if the monitoring area is in the visual effective state, performing smoke visual identification on the image data to obtain first smoke state results; if the monitoring area is in the smoke visual invalid state, performing time sequence analysis on the physical sensing signals to obtain second smoke state results; and performing early warning processing on the first smoke state results and the second smoke state results respectively to obtain early warning results; the application realizes automatic discrimination of image quality of the monitoring area, automatically switches to a time sequence analysis channel of the physical sensing signals, and improves the accuracy of the fire-fighting monitoring platform.

Description

Technical Field

[0001] This invention belongs to the field of online monitoring technology, specifically relating to a smart fire protection online monitoring and alarm platform and method. Background Technology

[0002] As urban population density continues to increase, the risk of fire also rises. Traditional fire monitoring methods mainly rely on independently deployed physical sensing devices such as smoke detectors and temperature sensors, or visually determine the location and spread of smoke through image content.

[0003] Existing fire monitoring technologies typically deploy smoke detectors, temperature sensors, and other sensors in a dispersed manner within the monitoring area when processing multi-source monitoring data. By collecting physical signals such as smoke concentration and temperature values, and setting fixed thresholds for judgment, local audible and visual alarms are triggered or alarm information is uploaded to the management platform when the monitored values ​​exceed the preset alarm thresholds. At the same time, network cameras are deployed in the monitoring area to transmit real-time image data to the monitoring center for identification and to issue early warnings.

[0004] However, in actual fire monitoring, firstly, smoke detectors may detect water mist, dust, etc., leading to false alarms; secondly, monitoring smoke in image data may result in image quality degradation (nighttime, backlight, strong light, or low illumination, etc.), making it impossible to detect anomalies in a timely manner, thus causing alarm delays; the above are all independently deployed monitoring systems, lacking an automatic switching mechanism, which reduces the accuracy of the fire monitoring platform. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that fire monitoring cannot perform automatic switching mechanisms, thereby reducing the accuracy of the fire monitoring platform, and to propose a smart fire online monitoring and alarm platform and method.

[0006] In a first aspect of this invention, a smart fire protection online monitoring and alarm method is first proposed, the method comprising: Real-time acquisition of image data and physical sensing signals of the monitoring area, wherein the physical sensing signals include smoke concentration values ​​and temperature values; Pixel brightness analysis is performed on the image data to obtain the image visual state results of the monitored area; the image visual state results include visually effective state and visually ineffective state; If the monitored area is in a visually effective state, the first smoke state result is obtained by performing smoke visual recognition on the image data. If the monitored area is in a state of visual failure due to smoke, a second smoke state result is obtained by performing time-series analysis based on the physical sensing signals. After performing early warning processing on the first smoke state result and the second smoke state result respectively, the early warning result is obtained; The warning results are transmitted to a remote backend for coordinated processing.

[0007] Optionally, the step of performing pixel brightness analysis on the image data to obtain the image visual state result of the monitored area includes: Obtain the luminance component values ​​of each pixel in the image data, and generate a luminance histogram of the monitored area; Calculate the average luminance value and luminance distribution variance based on the luminance histogram; Based on the average brightness value and the brightness distribution variance, combined with the preset effective brightness range, the image data is visually evaluated to obtain a smoke visual quality score. The smoke visual quality score is compared with a preset visual quality threshold. If the smoke visual quality score is greater than or equal to the preset visual quality threshold, the monitoring area is determined to be in a visually valid state; if the smoke visual quality score is less than the preset visual quality threshold, it is determined to be in a visually invalid state.

[0008] By introducing objective quantitative indicators such as brightness histograms, average brightness, and brightness distribution variance, the traditional image quality assessment, which relies on subjective experience, is transformed into a standardized data-driven process, significantly improving the consistency and reproducibility of the judgment. Simultaneously, by establishing a clear visual quality scoring and threshold comparison mechanism, a reliable basis is provided for the platform to automatically switch between visual recognition channels and physical signal channels, ensuring the validity of subsequent smoke detection input data from the source. This effectively avoids missed alarms and alarm delays caused by poor image quality in complex lighting environments such as nighttime and backlighting.

[0009] Optionally, the step of performing smoke visual recognition on the image data to obtain the first smoke state result includes: The image data is subjected to foreground and background separation processing to obtain the color features, texture features and edge blurring features of the smoke, and a multidimensional smoke feature vector is constructed. The multidimensional smoke feature vector is input into a preset smoke visual recognition model, which outputs a first smoke state result for the monitored area. The model is pre-trained based on a convolutional neural network. The first smoke state result is a visual judgment result indicating whether smoke exists, and a visual estimate of the smoke concentration when smoke is present.

[0010] A combination of piecewise functions and weighted summation is used to calculate the visual quality score of smoke. The piecewise function accurately depicts the nonlinear decay relationship when the average brightness deviates from the ideal range, making the score more sensitive to extreme lighting conditions and providing a more reasonable response. The weighted summation comprehensively considers both brightness neutrality and brightness distribution uniformity. This quantitative evaluation method provides a refined numerical basis for determining the effective / ineffective state of vision, avoiding the coarseness of single threshold judgments, and further improving the smoothness and accuracy of the platform's state switching in scenes with gradual changes in lighting.

[0011] Optionally, the second smoke state result obtained by performing time-series analysis based on the physical sensing signals includes: Obtain the time-series sampling sequence of the smoke concentration value and the temperature value within the current time window; The time-series sampling sequence is filtered and denoised to obtain an effective smoke concentration sequence and an effective temperature sequence; Calculate the rate of change of the effective smoke concentration sequence and the rate of change of the effective temperature sequence, respectively. The concentration change rate is compared with a preset concentration abrupt change threshold, and the temperature change rate is compared with a preset temperature abrupt change threshold; If the rate of change of concentration exceeds the preset concentration mutation threshold, or the rate of change of temperature exceeds the preset temperature mutation threshold, then the monitoring area is determined to have a smoke risk, and a smoke risk level index is obtained by weighting the rate of change of concentration and the rate of change of temperature, which is used as the second smoke state result.

[0012] By separating the foreground and background and extracting multi-dimensional complementary features such as color, texture, and edge blurring of smoke, the model can fully characterize the essential visual attributes of smoke, such as translucency and blurred edges, enhancing its robustness to changes in lighting, background interference, and the diversity of smoke morphology. Simultaneously, relying on a pre-trained convolutional neural network for end-to-end inference, the model possesses strong generalization ability, adapting to differences in smoke appearance under different scenarios. This effectively reduces the false alarm rate caused by water mist, dust, or shadows, and simultaneously outputs the presence of smoke and its concentration estimate, achieving integrated processing of discrimination and estimation.

[0013] Optionally, transmitting the warning results to a remote backend for coordinated processing includes: The alert results are reported to the unified receiving interface of the remote backend via the MQTT protocol; After receiving the warning result, the unified receiving interface parses and obtains the message type, and determines whether the message type belongs to the preset alarm type set. If the alarm belongs to the preset alarm type set, the alarm result is converted into structured alarm data, and the alarm result is processed synchronously according to the preset alarm type set to generate alarm event records; the structured alarm data and the corresponding alarm event records are written into the database for storage. The remote backend responds to the frontend's active pull request and transmits the structured alarm data and corresponding alarm event records to the frontend via an HTTP interface.

[0014] Employing the MQTT protocol for reporting ensures stable early warning transmission capabilities for front-end devices even in weak network environments, guaranteeing information is not lost due to network fluctuations. A unified receiving interface enables centralized aggregation and protocol standardization of multi-source early warning data, reducing access complexity. In the data delivery phase, message type validation and matching with preset alarm type sets effectively filter non-alarm messages, reducing the consumption of invalid data on platform resources. Early warning results are converted into structured data and alarm event records are generated simultaneously, ensuring data standardization and traceability. Furthermore, a proactive front-end retrieval mechanism replaces passive push, allowing the front-end to obtain the latest alarm data on demand without maintaining a constant connection, reducing power consumption and bandwidth usage of front-end devices and improving overall platform scalability and resource utilization efficiency.

[0015] In a second aspect of this invention, a smart fire protection online monitoring and alarm platform is proposed, comprising: Data acquisition module: acquires image data and physical sensor signals of the monitoring area in real time, including smoke concentration and temperature values; Visual status module: performs pixel brightness analysis on image data to obtain the visual status results of the monitored area; the visual status results include visually valid status and visually invalid status; Smoke visual recognition module: If the monitored area is in a visually valid state, the smoke visual recognition is performed on the image data to obtain the first smoke state result; Physical signal recognition module: If the monitored area is in a state of visual failure due to smoke, the second smoke state result is obtained by performing time-series analysis based on the physical sensing signal; Early warning result module: After performing early warning processing on the first smoke state result and the second smoke state result respectively, the early warning result is obtained; Transmission and linkage module: Transmits the early warning results to the remote backend and performs linkage processing.

[0016] Optionally, the visual state module includes: a brightness histogram module, a calculation module, a visual quality assessment module, and a judgment module. The brightness histogram module is used to obtain the brightness component values ​​of each pixel in the image data and generate a brightness histogram of the monitoring area. The calculation module is used to calculate the average brightness value and the brightness distribution variance based on the brightness histogram; The visual quality assessment module is used to perform visual quality assessment on the image data based on the average brightness value and the brightness distribution variance, combined with a preset effective brightness range, to obtain a smoke visual quality score. The judgment module is used to compare the smoke visual quality score with a preset visual quality threshold. If the smoke visual quality score is greater than or equal to the preset visual quality threshold, the monitoring area is determined to be in a visually valid state; if the smoke visual quality score is less than the preset visual quality threshold, it is determined to be in a visually invalid state.

[0017] Optionally, the smoke visual recognition module includes: a separation processing module and a smoke status module. The separation processing module is used to perform foreground and background separation processing on the image data to obtain the color features, texture features and edge blurring features of the smoke, and construct a multidimensional smoke feature vector; The smoke state module is used to input the multidimensional smoke feature vector into a preset smoke visual recognition model and output a first smoke state result for the monitored area; the model is pre-trained based on a convolutional neural network. The first smoke state result is a visual judgment result indicating whether smoke exists, and a visual estimate of the smoke concentration when smoke exists.

[0018] Optionally, the physical signal recognition module includes: a time-series sampling sequence module, a filtering and denoising module, a data calculation module, a comparison module, and a discrimination module. The time-series sampling sequence module is used to obtain the time-series sampling sequence of the smoke concentration value and the temperature value within the current time window; The filtering and denoising module is used to perform filtering and denoising processing on the time-series sampling sequence to obtain an effective smoke concentration sequence and an effective temperature sequence; The data calculation module is used to calculate the concentration change rate of the effective smoke concentration sequence and the temperature change rate of the effective temperature sequence, respectively. The comparison module is used to compare the concentration change rate with a preset concentration mutation threshold and the temperature change rate with a preset temperature mutation threshold. The discrimination module is used to determine that there is a smoke risk in the monitoring area if the rate of change of concentration exceeds the preset concentration mutation threshold or the rate of change of temperature exceeds the preset temperature mutation threshold, and to calculate a smoke risk level index by weighting the rate of change of concentration and the rate of change of temperature, as the second smoke state result.

[0019] Optionally, the transmission linkage module is further used for: The alert results are reported to the unified receiving interface of the remote backend via the MQTT protocol; After receiving the warning result, the unified receiving interface parses and obtains the message type, and determines whether the message type belongs to the preset alarm type set. If the alarm belongs to the preset alarm type set, the alarm result is converted into structured alarm data, and the alarm result is processed synchronously according to the preset alarm type set to generate alarm event records; the structured alarm data and the corresponding alarm event records are written into the database for storage. The remote backend responds to the frontend's active pull request and transmits the structured alarm data and corresponding alarm event records to the frontend via an HTTP interface.

[0020] The beneficial effects of this invention are: By introducing an image visual state assessment mechanism, the system automatically judges the image quality of the monitored area. Under conditions of image quality degradation such as nighttime, backlight, strong light, or low illumination, it automatically switches to the time-series analysis channel of physical sensor signals, avoiding missed smoke alarms or delayed alarms due to image failure. This improves the platform's adaptability in complex environments and the accuracy of alarm response. Simultaneously, a convolutional neural network is used to perform smoke visual recognition on valid visual images, extracting multi-dimensional features such as color, texture, and edge blurring to determine the presence and concentration of smoke. Furthermore, the physical sensor signals are filtered, denoised, and analyzed for rate of change, effectively distinguishing smoke from interference sources such as water mist and dust, reducing false alarm rates, and improving the accuracy of monitoring results. In addition, through dual-channel discrimination and early warning result fusion processing, combined with MQTT reporting, structured storage, and HTTP active retrieval mechanisms, efficient transmission and coordinated handling of early warning information are achieved, enhancing the platform's backend data integration capabilities and frontend information acquisition convenience. Attached Figure Description

[0021] The invention will now be further described with reference to the accompanying drawings.

[0022] Figure 1 A flowchart of the intelligent fire protection online monitoring and alarm method provided in the embodiments of the present invention; Figure 2 A framework diagram of the intelligent fire protection online monitoring and alarm platform provided in an embodiment of the present invention; Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] It should be noted that all formula calculations in the scheme are purely numerical calculations.

[0025] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention provides a method for intelligent online fire monitoring and alarm. See also... Figure 1 The method includes the following steps: S101: Real-time acquisition of image data and physical sensor signals of the monitored area, including smoke concentration and temperature values; S102: Perform pixel brightness analysis on the image data to obtain the image visual status results of the monitored area; the image visual status results include visually effective status and visually ineffective status; S103: If the monitored area is in a visually effective state, then perform smoke visual recognition on the image data to obtain the first smoke state result; S104: If the monitored area is in a state of visual failure due to smoke, the second smoke state result is obtained by performing time-series analysis based on the physical sensor signal; S105: After performing early warning processing on the first smoke state result and the second smoke state result respectively, the early warning result is obtained; S106: Transmit the warning results to the remote backend and perform linkage processing.

[0027] In one implementation, pixel brightness analysis of image data is performed to obtain the visual state results of the monitored area, including: Obtain the brightness component values ​​of each pixel in the image data and generate a brightness histogram of the monitored area; Calculate the average luminance value and luminance distribution variance based on the luminance histogram; Based on the average brightness value and the variance of brightness distribution, combined with the preset effective brightness range, the visual quality of the image data is evaluated to obtain the smoke visual quality score. The smoke visual quality score is compared with a preset visual quality threshold. If the smoke visual quality score is greater than or equal to the preset visual quality threshold, the monitoring area is determined to be in a visually effective state; if the smoke visual quality score is less than the preset visual quality threshold, it is determined to be in a visually ineffective state.

[0028] One implementation addresses the lack of objective quantitative indicators for image quality assessment in traditional fire monitoring, transforming visual state judgment from experience-based to data-driven, thus improving the consistency and reproducibility of the judgment. Furthermore, the assessment result, as the core of the subsequent processing chain, directly determines whether the platform should activate the smoke visual recognition channel based on convolutional neural networks or switch to the time-series analysis channel of physical sensor signals. This effectively avoids the problems of missed alarms and alarm delays caused by unreliable visual model outputs in scenarios with degraded image quality, such as low illumination, overexposure, or drastic fluctuations in lighting. It ensures the validity of subsequent smoke judgment input data from the source, significantly improving the adaptability of the fire monitoring platform in complex lighting environments and the determinism of alarm response.

[0029] In one implementation, the visual quality assessment of image data to obtain a smoke visual quality score specifically includes: Obtain the average brightness value and preset the effective brightness range as [L]. min ,L max ] Calculate the center brightness value of this interval using the following formula: ,in, L is the center brightness value of this range. min and L max These are the minimum and maximum brightness values ​​within that range; The luminance neutrality score is calculated using a piecewise function; the formula for calculating the luminance neutrality score is:

[0030] in, It is a neutral brightness rating. This is the average brightness value. ( ) represents the maximum value function; specifically, At that time, the brightness is determined to be within the ideal visual range; The visual quality score of smoke is obtained by weighted summation of the luminance neutral score and the luminance distribution variance.

[0031] Specifically, the preset visual quality threshold and the preset effective brightness range are reasonable values ​​set by staff based on historical data; In one implementation, obtaining a first smoke state result by performing smoke visual recognition on image data includes: Foreground and background separation processing is performed on image data to obtain the color features, texture features and edge blurring features of smoke, and a multidimensional smoke feature vector is constructed. The multidimensional smoke feature vector is input into a pre-defined smoke visual recognition model, which outputs the first smoke state result of the monitored area. The model is pre-trained based on a convolutional neural network. The first smoke state result is the visual judgment result of whether smoke exists, and the visual estimate of the smoke concentration when smoke exists.

[0032] In one embodiment, a visual recognition chain is constructed, consisting of foreground / background separation, multi-dimensional smoke feature extraction, and inference using a pre-trained smoke visual recognition model (such as a CNN model), achieving integrated processing for smoke presence determination and concentration estimation. First, the image is separated into foreground and background, accurately extracting three complementary features of the smoke: color, texture, and edge blurriness. A multi-dimensional feature vector is constructed to fully characterize the essential visual attributes of the smoke, such as its translucency, blurred edges, and specific color distribution. Then, an end-to-end inference judgment is performed using a pre-trained convolutional neural network model, outputting a binary result indicating the presence of smoke and its concentration estimate. The smoke concentration visual estimate refers to the continuous numerical quantitative index output by the pre-trained smoke visual recognition model (based on a pre-trained convolutional neural network) after inference on the image data of the monitored area, used to characterize the relative level of smoke concentration in the image.

[0033] One implementation significantly enhances the model's robustness to changes in illumination, background interference, and the diversity of smoke morphology, effectively reducing the false alarm rate caused by water mist, dust, or shadows. At the same time, the CNN model, trained with a large number of labeled samples, has strong generalization ability and can adapt to the differences in smoke appearance under different scenarios, avoiding the problem of insufficient adaptability caused by manually setting fixed rules.

[0034] In one implementation, the second smoke state result is obtained by performing time-series analysis based on physical sensor signals, including: Obtain the time-series sampling sequence of smoke concentration and temperature values ​​within the current time window; The time-series sampling sequence is filtered and denoised to obtain the effective smoke concentration sequence and the effective temperature sequence. Calculate the rate of change of concentration for the effective smoke concentration sequence and the rate of change of temperature for the effective temperature sequence, respectively. The rate of concentration change is compared with a preset concentration abrupt change threshold, and the rate of temperature change is compared with a preset temperature abrupt change threshold; If the rate of change of concentration exceeds the preset concentration mutation threshold, or the rate of change of temperature exceeds the preset temperature mutation threshold, then the monitoring area is determined to have a smoke risk. The smoke risk level index is calculated by weighting the rate of change of concentration and the rate of change of temperature, and is used as the second smoke state result.

[0035] Specifically, it should be noted that the smoke risk level index is a continuous numerical indicator used to quantify the severity of the current smoke risk in the monitoring area; the second smoke status result refers to the final judgment result output by the physical sensor signal time sequence analysis channel after the image visual state is judged to be visually ineffective, which is used to characterize the smoke risk status of the monitoring area (specifically including: smoke risk presence indicator, smoke risk level index size, the higher the level, the higher the risk, etc.).

[0036] In one implementation, by analyzing the temporal sampling and rate of change of smoke concentration and temperature, the static threshold comparison is upgraded to dynamic trend discrimination, which can keenly capture the transient change characteristics in the early stage of a fire, fundamentally avoiding false triggers caused by slow sensor drift or environmental fluctuations. The dual physical quantity joint discrimination and weighted fusion mechanism can effectively distinguish between fire smoke and non-fire interference sources such as water vapor and dust. The latter usually only causes a change in a single physical quantity, thus greatly reducing the false alarm rate. Thirdly, the output smoke risk level index is a continuous quantitative value, which enables the subsequent early warning module to implement graded responses according to the severity of the risk, replacing the single handling method of traditional binary alarms, and improving the precision of platform decision-making and the rationality of linkage response.

[0037] In one implementation, the specific process of S105 includes: Obtain the first smoke state result, which includes the visual judgment result of whether smoke exists and the visual estimate of smoke concentration when smoke exists; The results of the first smoke state are analyzed to extract the smoke presence markers and visual estimates of smoke concentration. If smoke is present, it is marked as no smoke, and a first channel no-warning status record is generated; If smoke is detected, the visually estimated smoke concentration is assessed based on preset concentration grading thresholds: when the visually estimated smoke concentration is below the first concentration threshold, it is classified as a first warning level; when the visually estimated smoke concentration is between the first and second concentration thresholds, it is classified as a second warning level; when the visually estimated smoke concentration is above the second concentration threshold, it is classified as a third warning level; wherein, the first concentration threshold is less than the second concentration threshold. Based on the determined warning level, a first-channel warning information is generated. The first-channel warning information includes a warning type identifier, a warning level, a visual estimate of smoke concentration, and a timestamp. The first-channel warning information is then encapsulated into a first-structured warning result. Obtain the second smoke state result, which includes a smoke risk presence indicator and a smoke risk level index; The results of the second smoke state were analyzed to extract the smoke risk presence marker and the smoke risk level index. If the smoke risk exists but is marked as no smoke risk, a second channel no-warning status record is generated; If a smoke risk is identified as present, the smoke risk level index is determined according to a preset risk level threshold range: when the smoke risk level index is lower than the first risk threshold, it is determined to be the first warning level; when the smoke risk level index is between the first risk threshold and the second risk threshold, it is determined to be the second warning level; when the smoke risk level index is higher than the second risk threshold, it is determined to be the third warning level; wherein, the first risk threshold is lower than the second risk threshold. Based on the determined warning level, a second-channel warning information is generated. The second-channel warning information includes a warning type identifier, warning level, smoke risk level index, and timestamp. The second-channel warning information is then encapsulated into a second structured warning result. The first structured early warning result or the first channel no-early-warning-state record, together with the second structured early warning result or the second channel no-early-warning-state record, are output as early warning results to the transmission linkage module; wherein, the early warning processing of the first channel and the second channel is executed independently.

[0038] In one implementation, transmitting the early warning result to a remote backend for coordinated processing includes: The alert results are reported to the unified receiving interface of the remote backend via the MQTT protocol; After receiving the warning result through the unified receiving interface, the message type is parsed and determined to see if the message type belongs to the preset alarm type set. If the alarm belongs to a preset alarm type set, the alarm result is converted into structured alarm data, and the alarm result is processed synchronously according to the preset alarm type set to generate alarm event records; the structured alarm data and the corresponding alarm event records are written into the database for storage. The remote backend responds to the frontend's proactive pull request and transmits structured alarm data and corresponding alarm event records to the frontend via an HTTP interface.

[0039] Specifically, the preset alarm type set includes temperature alarm, infrared monitoring alarm, gas concentration exceeding standard alarm, etc.; the alarm event record refers to a "work order" created in the database for each "equipment alarm to be processed", which records the equipment type, time, alarm type, etc.

[0040] In one implementation, the MQTT protocol is used for reporting. Its lightweight, low bandwidth consumption, and publish / subscribe characteristics enable front-end monitoring devices to maintain stable early warning transmission capabilities even in weak network environments, ensuring that early warning information is not lost due to network fluctuations. A unified receiving interface is set up in the remote backend, realizing centralized aggregation and protocol unification of multi-source early warning data, avoiding access complexity caused by heterogeneous interfaces. In the data landing stage, non-alarm messages are effectively filtered by message type verification and matching with preset alarm type sets, reducing the occupation of platform resources by invalid data. The early warning results are converted into structured alarm data and alarm event records are generated simultaneously, ensuring the standardization and traceability of the data, providing complete data support for subsequent auditing, statistics, and responsibility determination. At the same time, a front-end active pull mechanism is adopted instead of the traditional passive push, so that the front end can obtain the latest alarm data on demand without maintaining a constant connection, reducing the power consumption and bandwidth occupation of the front-end devices. This is particularly suitable for scenarios with a large number of edge devices connected in fire monitoring, and the overall scalability and resource utilization efficiency of the platform are improved.

[0041] This invention also provides an intelligent fire protection online monitoring and alarm platform. See [link / reference] Figure 2 The platform includes the following modules: Data acquisition module (201): acquires image data and physical sensor signals of the monitored area in real time. The physical sensor signals include smoke concentration value and temperature value. Visual status module (202): performs pixel brightness analysis on image data to obtain the image visual status results of the monitored area; the image visual status results include visually valid status and visually invalid status; Smoke visual recognition module (203): If the monitoring area is in a visually valid state, the smoke visual recognition is performed on the image data to obtain the first smoke state result; Physical signal recognition module (204): If the monitored area is in a smoke visual failure state, the second smoke state result is obtained by performing time-series analysis based on the physical sensing signal; Early warning result module (205): After performing early warning processing on the first smoke state result and the second smoke state result respectively, the early warning result is obtained; Transmission and linkage module (206): Transmits the warning results to the remote backend and performs linkage processing.

[0042] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A smart fire protection online monitoring and alarm method, characterized in that, The method includes: Real-time acquisition of image data and physical sensing signals of the monitoring area, wherein the physical sensing signals include smoke concentration values ​​and temperature values; Pixel brightness analysis is performed on the image data to obtain the image visual state results of the monitored area; the image visual state results include visually effective state and visually ineffective state; If the monitored area is in a visually effective state, the first smoke state result is obtained by performing smoke visual recognition on the image data. If the monitored area is in a state of visual failure due to smoke, a second smoke state result is obtained by performing time-series analysis based on the physical sensing signals. After performing early warning processing on the first smoke state result and the second smoke state result respectively, the early warning result is obtained; The warning results are transmitted to a remote backend for coordinated processing.

2. The intelligent fire protection online monitoring and alarm method according to claim 1, characterized in that, The step of performing pixel brightness analysis on the image data to obtain the image visual state results of the monitored area includes: Obtain the luminance component values ​​of each pixel in the image data, and generate a luminance histogram of the monitored area; Calculate the average luminance value and luminance distribution variance based on the luminance histogram; Based on the average brightness value and the brightness distribution variance, combined with the preset effective brightness range, the image data is visually evaluated to obtain a smoke visual quality score. The smoke visual quality score is compared with a preset visual quality threshold. If the smoke visual quality score is greater than or equal to the preset visual quality threshold, the monitoring area is determined to be in a visually valid state; if the smoke visual quality score is less than the preset visual quality threshold, it is determined to be in a visually invalid state.

3. The intelligent fire protection online monitoring and alarm method according to claim 1, characterized in that, The process of obtaining the first smoke state result by performing smoke visual recognition on the image data includes: The image data is subjected to foreground and background separation processing to obtain the color features, texture features and edge blurring features of the smoke, and a multidimensional smoke feature vector is constructed. The multidimensional smoke feature vector is input into a preset smoke visual recognition model, and the first smoke state result of the monitoring area is output; the model is obtained by pre-training a convolutional neural network.

4. The intelligent fire protection online monitoring and alarm method according to claim 1, characterized in that, The second smoke state result obtained by performing time-series analysis based on the physical sensor signals includes: Obtain the time-series sampling sequence of the smoke concentration value and the temperature value within the current time window; The time-series sampling sequence is filtered and denoised to obtain an effective smoke concentration sequence and an effective temperature sequence; Calculate the rate of change of the effective smoke concentration sequence and the rate of change of the effective temperature sequence, respectively. The concentration change rate is compared with a preset concentration abrupt change threshold, and the temperature change rate is compared with a preset temperature abrupt change threshold; If the rate of change of concentration exceeds the preset concentration mutation threshold, or the rate of change of temperature exceeds the preset temperature mutation threshold, then the monitoring area is determined to have a smoke risk, and a smoke risk level index is obtained by weighting the rate of change of concentration and the rate of change of temperature, which is used as the second smoke state result.

5. The intelligent fire protection online monitoring and alarm method according to claim 1, characterized in that, Transmitting early warning results to a remote backend and performing coordinated processing includes: The alert results are reported to the unified receiving interface of the remote backend via the MQTT protocol; After receiving the warning result, the unified receiving interface parses and obtains the message type, and determines whether the message type belongs to the preset alarm type set. If the alarm belongs to the preset alarm type set, the alarm result is converted into structured alarm data, and the alarm result is processed synchronously according to the preset alarm type set to generate alarm event records; the structured alarm data and the corresponding alarm event records are written into the database for storage. The remote backend responds to the frontend's active pull request and transmits the structured alarm data and corresponding alarm event records to the frontend via an HTTP interface.

6. A smart fire protection online monitoring and alarm platform, characterized in that, The device includes: Data acquisition module: acquires image data and physical sensor signals of the monitoring area in real time, including smoke concentration and temperature values; Visual status module: performs pixel brightness analysis on image data to obtain the visual status results of the monitored area; the visual status results include visually valid status and visually invalid status; Smoke visual recognition module: If the monitored area is in a visually valid state, the smoke visual recognition is performed on the image data to obtain the first smoke state result; Physical signal recognition module: If the monitored area is in a state of visual failure due to smoke, the second smoke state result is obtained by performing time-series analysis based on the physical sensing signal; Early warning result module: After processing the first smoke state result and the second smoke state result separately, the early warning result is obtained; Transmission and linkage module: Transmits the early warning results to the remote backend and performs linkage processing.

7. The intelligent fire protection online monitoring and alarm platform according to claim 6, characterized in that, The visual state module includes: a brightness histogram module, a calculation module, a visual quality assessment module, and a judgment module. The brightness histogram module is used to obtain the brightness component values ​​of each pixel in the image data and generate a brightness histogram of the monitoring area. The calculation module is used to calculate the average brightness value and the brightness distribution variance based on the brightness histogram; The visual quality assessment module is used to perform visual quality assessment on the image data based on the average brightness value and the brightness distribution variance, combined with a preset effective brightness range, to obtain a smoke visual quality score. The judgment module is used to compare the smoke visual quality score with a preset visual quality threshold. If the smoke visual quality score is greater than or equal to the preset visual quality threshold, the monitoring area is determined to be in a visually valid state; if the smoke visual quality score is less than the preset visual quality threshold, it is determined to be in a visually invalid state.

8. The intelligent fire protection online monitoring and alarm platform according to claim 6, characterized in that, The smoke visual recognition module includes: a separation processing module and a smoke status module. The separation processing module is used to perform foreground and background separation processing on the image data to obtain the color features, texture features and edge blurring features of the smoke, and construct a multidimensional smoke feature vector; The smoke state module is used to input the multidimensional smoke feature vector into a preset smoke visual recognition model and output the first smoke state result of the monitoring area; the model is obtained by pre-training a convolutional neural network.

9. The intelligent fire protection online monitoring and alarm platform according to claim 6, characterized in that, The physical signal recognition module includes: a time-series sampling sequence module, a filtering and denoising module, a data calculation module, a comparison module, and a discrimination module. The time-series sampling sequence module is used to obtain the time-series sampling sequence of the smoke concentration value and the temperature value within the current time window; The filtering and denoising module is used to perform filtering and denoising processing on the time-series sampling sequence to obtain an effective smoke concentration sequence and an effective temperature sequence; The data calculation module is used to calculate the concentration change rate of the effective smoke concentration sequence and the temperature change rate of the effective temperature sequence, respectively. The comparison module is used to compare the concentration change rate with a preset concentration mutation threshold and the temperature change rate with a preset temperature mutation threshold. The discrimination module is used to determine that there is a smoke risk in the monitoring area if the rate of change of concentration exceeds the preset concentration mutation threshold or the rate of change of temperature exceeds the preset temperature mutation threshold, and to calculate a smoke risk level index by weighting the rate of change of concentration and the rate of change of temperature, as the second smoke state result.

10. The intelligent fire protection online monitoring and alarm platform according to claim 6, characterized in that, The transmission linkage module is also used for: The alert results are reported to the unified receiving interface of the remote backend via the MQTT protocol; After receiving the warning result, the unified receiving interface parses and obtains the message type, and determines whether the message type belongs to the preset alarm type set. If it belongs to the preset alarm type set, the warning result is converted into structured alarm data, and the warning result is processed synchronously according to the preset alarm type set to generate an alarm event record; Structured alarm data and corresponding alarm event records are written into the database for storage; The remote backend responds to the frontend's active pull request and transmits the structured alarm data and corresponding alarm event records to the frontend via an HTTP interface.

Images