Multi-point temperature sampling array fire judgment system and control method thereof

Abstract

The invention relates to the technical field of fire safety, in particular to a multi-point temperature sampling array fire judgment system and a control method, and the system comprises a temperature sampling array unit which is used for synchronously collecting temperature data of sampling points and outputting a temperature sampling matrix; the optical focusing assembly is used for expanding a detection field angle or enhancing energy convergence of a specific sensitive area; the processing and control unit is used for receiving the temperature sampling matrix, operating a fire judgment algorithm and generating a judgment signal according to a fire judgment result; the output interface unit is used for converting the judgment signal into a corresponding control signal and outputting the control signal to an external execution device; the communication module is used for uploading the system state and the fire judgment data to a monitoring platform or a cloud server; and the power supply and self-checking module is used for supplying power to the system and realizing automatic calibration and health state monitoring. Therefore, the problems of response speed, structural complexity, cost, privacy protection defects and the like are solved.

Description

Technical Field

[0001] This application relates to the field of fire safety technology, and in particular to a multi-point temperature sampling array fire detection system and its control method. Background Technology

[0002] Fire detection and early warning technologies play a crucial role in building safety and industrial protection, serving as a key line of defense for safeguarding lives and property and maintaining social stability. With accelerating urbanization and continuous industrial development, the harm caused by fire accidents is becoming increasingly severe; therefore, efficient and reliable fire monitoring systems have become an urgent need. However, while existing fire monitoring technologies have played a role to some extent, many problems still need to be addressed.

[0003] Currently, common fire monitoring systems mainly rely on smoke detectors, single-point temperature detectors, and infrared imaging-based fire identification devices. Smoke detectors determine the presence of a fire by detecting changes in the concentration of smoke particles in the air; when the smoke particle concentration exceeds a set threshold, an alarm signal is triggered. These detectors can respond to smoke during a fire, thus providing a certain degree of fire early warning. Single-point temperature detectors directly detect changes in the surrounding environment's temperature; when the temperature reaches or exceeds a preset alarm temperature, an alarm is issued. They can visually reflect local temperature conditions and promptly detect abnormal temperature increases. Infrared imaging-based fire identification devices utilize infrared thermal imaging technology to convert the infrared radiation emitted by objects into visible images. By analyzing the temperature distribution and changes in the images, the location and size of the fire source can be identified. This type of device is used in scenarios where precise fire source location is crucial.

[0004] These traditional fire monitoring technologies have some significant drawbacks. Smoke detectors rely on the diffusion of smoke in the air; in well-ventilated environments or those with significant airflow disturbances, smoke diffusion is affected, leading to a noticeable response delay. For example, in places with complex ventilation conditions, such as large shopping malls and factory workshops, smoke detectors may fail to detect a fire in time, thus delaying optimal firefighting opportunities. Single-point temperature detectors only reflect localized information; when the fire source is far from the sensor, there are blind spots. In some large buildings or industrial facilities, single-point temperature detectors cannot fully cover the entire space, failing to detect distant fire sources in a timely manner, thus reducing the reliability of fire monitoring. To improve spatial coverage, some solutions employ infrared scanning or thermal imaging. For example, the infrared scanning-based spray fire suppression system disclosed in British Patent GB2471317A (Plumis Ltd, 2011) uses a motor-driven infrared sensor rotation to locate the fire source. While this approach improves detection coverage to some extent, the internal rotating mechanism suffers from mechanical wear, noise, and response delays. During long-term operation, components of the rotating mechanism are prone to failure due to wear, increasing system maintenance costs and downtime, making it unsuitable for long-term operation or civilian scenarios requiring high stability and reliability. Furthermore, the fire detection methods based on thermal imaging arrays disclosed in US Patent US20190345678A1 and Chinese Patent CN110987654A, while capable of fire source identification, require high-cost infrared imaging components, increasing system procurement and deployment costs. Additionally, in office and residential environments, thermal imaging technology may raise privacy concerns, limiting its widespread application. These systems rely on complex image processing algorithms and high-bandwidth communication, which place high demands on hardware performance and network conditions, making them difficult to implement in low-power devices and further limiting their application scope.

[0005] In fields such as energy storage devices, power distribution cabinets, rail transit, and smart buildings, users have placed higher demands on fire detection systems. They urgently need a fire detection system that requires no rotating parts, is low-cost, can respond quickly, and has intelligent judgment capabilities. To fill this market gap, this invention proposes a non-rotating, multi-point temperature sampling array fire judgment system and control method. Through static multi-point temperature array sampling and a multi-parameter fusion algorithm, this system can achieve real-time analysis and intelligent judgment of fire conditions, overcoming the shortcomings of existing technologies in terms of mechanical reliability, response delay, and privacy protection. Without the need for rotating parts, the system can achieve rapid, accurate, and low-power fire detection, providing more reliable protection for fire safety in various fields. Summary of the Invention

[0006] This application provides a multi-point temperature sampling array fire detection system and its control method to solve problems such as deficiencies in response speed, structural complexity, cost and privacy protection.

[0007] The first aspect of this application provides a multi-point temperature sampling array fire determination system, including: a temperature sampling array unit, fixedly installed in a monitoring area, used to synchronously collect temperature data from multiple discrete spatial sampling points in the monitoring area and output a temperature sampling matrix; An optical focusing component is disposed at the sensing front end of the temperature sampling array unit to expand the detection field of view or enhance the energy convergence of a specific sensitive area; A processing and control unit, electrically connected to the temperature sampling array unit, is used to receive the temperature sampling matrix and run a fire situation determination algorithm, and generate a determination signal based on the fire situation determination result. The processing and control unit includes: The data preprocessing module is used to perform noise suppression, background temperature compensation, and outlier correction on the temperature sampling matrix. The feature extraction module is used to extract time-domain and spatial-domain temperature feature parameters from the preprocessed data; The fire situation determination logic module is used to generate a fire situation determination signal based on the fusion analysis results of the time domain and spatial domain temperature characteristic parameters; An output interface unit, connected to the processing and control unit, is used to receive the determination signal and convert the determination signal into a corresponding control signal for output to an external execution device; The communication module is used to upload system status and fire situation assessment data to the monitoring platform or cloud server; The power supply and self-test module is used to power the system and to achieve automatic calibration and health status monitoring. The system is a static non-scanning architecture, without mechanical scanning or imaging components. The temperature sampling array unit is a non-imaging multi-point temperature sensor array. The fire determination is based on multi-parameter fusion analysis in the time and spatial domains.

[0008] Optionally, the feature extraction module further includes a temperature change trend analysis submodule and an energy density calculation submodule, wherein, The temperature change trend analysis submodule is used to analyze the change pattern of the temperature characteristic parameters over time; The energy density calculation submodule is used to calculate the energy density distribution within the monitoring area based on the spatial distribution of the temperature sampling matrix.

[0009] Optionally, the temperature characteristic parameters include any one or more of the following: temperature rise rate, energy density, average temperature, maximum temperature, and spatial continuity index, wherein the spatial continuity index is obtained by performing image connected component analysis on the binarized high-temperature region.

[0010] Optionally, the processing and control unit further includes a dynamic threshold and learning module, used to adaptively adjust the judgment threshold and fusion feature weights for fire determination based on changes in ambient temperature and historical fire data.

[0011] Optionally, the dynamic threshold and learning module includes: Input feature set, used to cache historical temperature feature parameters; The feature normalization unit is used to normalize the temperature feature parameters; The online learning algorithm unit is used to output optimized judgment thresholds and feature weights based on normalized feature parameters and feedback signals through online machine learning algorithms. The parameter update unit is used to smoothly update the optimized parameters to the fire determination logic module. Local storage units are used to store system parameters and historical data in a non-volatile manner.

[0012] Optionally, the output interface unit is connected to an external actuator, which includes an audible and visual alarm, an automatic fire extinguishing device, a power cut-off device, and other safety linkage equipment.

[0013] Optionally, the communication module supports wired interfaces or wireless communication protocols, including at least one of RS485, CAN, NB-IoT, Wi-Fi, and LoRa communication protocols.

[0014] Optionally, the power supply and self-test module integrates non-uniformity correction function, background temperature adaptive update function, contamination detection function, and temperature drift compensation function.

[0015] Optionally, the judgment signal output by the processing and control unit is a graded fire judgment signal, including a warning signal, a fire confirmation signal, and a fire persistence signal.

[0016] Optionally, the fire situation determination adopts a dual-time-window comparison mechanism, which determines transient heat sources and continuous fire situations by the trend difference between short-time windows and long-time windows.

[0017] Optionally, the system is applicable to at least one of the following applications: building fire protection, industrial safety, energy storage equipment protection, rail transit environmental monitoring, and smart home security.

[0018] A second aspect of this application provides a fire detection method using a multi-point temperature sampling array, the method being executed by a processing and control unit, and including the following steps: Temperature data from multiple discrete spatial sampling points within the monitoring area are simultaneously collected by a temperature sampling array unit to generate a temperature sampling matrix. The temperature matrix is ​​preprocessed and spatially mapped to establish the correspondence between sampling points and physical monitoring areas; Extract and calculate multiple temperature feature parameters, including time-domain and spatial-domain feature parameters; Fire status is determined based on the fusion and change trends of the time-domain and spatial-domain feature parameters. If the fire conditions are determined to continue, a control signal is generated and output to the external actuator.

[0019] Optionally, the fire situation determination adopts a dual-timescale comparison strategy that combines short-term and long-term windows, and makes a decision by comprehensively considering the sudden changes within the short-term window and the trends within the long-term window.

[0020] Optionally, the method further includes dynamically updating the fire determination threshold and feature weights through an online learning mechanism, and achieving adaptive optimization based on historical operating data and real-time environmental parameters.

[0021] Optionally, the online learning mechanism uses machine learning algorithms to incrementally train the input feature set and outputs optimized judgment thresholds and multi-feature fusion weights.

[0022] A third aspect of this application provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being configured to perform a fire determination method using a multi-point temperature sampling array as described in the above embodiments.

[0023] Therefore, the present invention has at least the following beneficial effects: This application employs a non-rotating structure, enabling multi-point parallel sampling without the need for mechanical moving parts. This design eliminates the risk of failure caused by rotational wear and positional misalignment, significantly improving the system's lifespan and stability. During long-term operation, performance degradation or failures due to mechanical component wear will not occur, reducing maintenance costs and downtime. This makes it suitable for applications with extremely high reliability requirements. Data acquisition focuses solely on temperature data, without involving images or visual information. This non-imaging approach avoids privacy issues that may arise from image acquisition, while effectively monitoring fires and ensuring the safety of personnel and property. By combining multi-dimensional parameters such as temperature rise rate, energy density, maximum temperature, and spatial continuity, and employing a comprehensive threshold and time-duration logic, the system significantly reduces... By comprehensively analyzing multiple parameters, the system can more accurately determine the fire situation, avoiding misjudgments that may result from relying on a single parameter. The system automatically adjusts the judgment threshold and weights based on environmental changes and historical data, achieving algorithmic adaptation. This not only ensures stable judgment performance over long-term operation but also continuously optimizes the judgment parameters, improving the ability to judge fire situations and reducing false alarms and missed alarms. The system's modular hardware design, including communication, power supply and self-test modules, and dynamic threshold and learning modules, provides excellent scalability and adaptability to different scenarios. The combination of multi-point parallel sampling and embedded computing reduces the response time to less than 1-2 seconds, enabling early fire identification and automatic linkage control, thus improving the efficiency and accuracy of fire response.

[0024] This addresses issues related to response speed, structural complexity, cost, and privacy protection.

[0025] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0026] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a schematic diagram of a multi-point temperature sampling array fire detection system provided according to an embodiment of this application; Figure 2 This is a flowchart of a fire situation determination method using a multi-point temperature sampling array according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0027] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0028] The following description, with reference to the accompanying drawings, describes a multi-point temperature sampling array fire detection system and its control method according to embodiments of this application. Addressing the shortcomings mentioned in the background art regarding response speed, structural complexity, cost, and privacy protection, this application provides a multi-point temperature sampling array fire detection system. This system employs a non-rotating structure, achieving multi-point parallel sampling without mechanical moving parts. This design eliminates the risk of failure caused by rotational wear and positional offset, significantly improving the system's lifespan and stability. During long-term operation, performance degradation or failures due to mechanical component wear will not occur, reducing maintenance costs and downtime. It is suitable for applications with extremely high reliability requirements. In data acquisition, only temperature data is collected, without involving images or visual information. This non-imaging scheme avoids privacy leaks that may arise from image acquisition, while effectively monitoring fire conditions and ensuring the safety of personnel and property. The system combines temperature rise rate, energy density, and maximum temperature... By incorporating multi-dimensional parameters such as spatial continuity and employing a comprehensive threshold and time-based logic, the system significantly reduces false alarms and missed alarms. Through comprehensive analysis of multiple parameters, it can more accurately determine the fire situation, avoiding misjudgments that may arise from relying on a single parameter. The system automatically adjusts the judgment threshold and weights based on environmental changes and historical data, achieving algorithmic adaptation. This not only ensures stable judgment performance over long-term operation but also continuously optimizes the judgment parameters, improving the ability to judge fire situations and reducing false alarms and missed alarms. The system expands upon communication, power supply and self-testing, and dynamic threshold and learning modules. These modular hardware designs provide excellent scalability and adaptability to different scenarios. The combination of multi-point parallel sampling and embedded computing reduces the response time to less than 1-2 seconds, enabling early fire identification and automatic linkage control, thus improving the efficiency and accuracy of fire response. This addresses shortcomings in response speed, structural complexity, cost, and privacy protection.

[0029] Specifically, Figure 1 This is a schematic diagram of the structure of the multi-point temperature sampling array fire detection system provided in the embodiments of this application.

[0030] like Figure 1 As shown in the diagram, the multi-point temperature sampling array fire determination system 10 includes: a temperature sampling array unit 101, an optical focusing component 102, a processing and control unit 103, an output interface unit 104, a communication module 105, a power supply and self-test module 106, and an external execution device 107.

[0031] Specifically, the temperature sampling array unit 101 is fixedly installed within the monitoring area to synchronously collect temperature data from multiple discrete spatial sampling points within the monitoring area and output a temperature sampling matrix; the optical focusing component 102 is installed at the sensing front end of the temperature sampling array unit 101 to expand the detection field of view or enhance energy convergence in specific sensitive areas; the processing and control unit 103 is electrically connected to the temperature sampling array unit 101 to receive the temperature sampling matrix, run the fire determination algorithm, and generate a determination signal based on the fire determination result; the output interface unit 104 is connected to the processing and control unit to receive the determination signal and convert it into a corresponding control signal for output to an external actuator; the communication module 105 is used to upload system status and fire determination data to a monitoring platform or cloud server; the power supply and self-test module 106 is used to power the system and achieve automatic calibration and health status monitoring; wherein, the system is a static non-scanning architecture, without mechanical scanning or imaging components, the temperature sampling array unit 101 is a non-imaging multi-point temperature sensor array, and the fire determination is based on multi-parameter fusion analysis in the time and spatial domains.

[0032] Optionally, in some embodiments, the processing and control unit 103 includes: a data preprocessing module for noise suppression, background temperature compensation, and outlier correction of the temperature sampling matrix; a feature extraction module for extracting time-domain and spatial-domain temperature feature parameters from the preprocessed data; and a fire determination logic module for generating a fire determination signal based on the fusion analysis results of the time-domain and spatial-domain temperature feature parameters.

[0033] It should be noted that this system is not only applicable to common fields such as building fire protection, industrial safety, energy storage equipment protection, rail transit, and smart homes, but also has broad scalability, and can be used for electrical overload monitoring, equipment thermal runaway detection, energy leakage monitoring, and other temperature anomaly identification scenarios. In electrical overload monitoring, it can promptly detect abnormal temperature increases in electrical equipment caused by overload, preventing equipment damage and fires. Regarding equipment thermal runaway detection, it can monitor the thermal runaway status of equipment such as energy storage batteries in real time, providing early warnings to ensure equipment and personnel safety. In energy leakage monitoring, by detecting temperature changes, it can promptly identify energy leakage points, reducing energy waste and safety hazards.

[0034] The following will describe in detail the various modules of the multi-point temperature sampling array fire determination system 10 of this application embodiment.

[0035] Temperature sampling array unit 101 is fixedly installed within the monitoring area to synchronously collect temperature data from multiple discrete spatial sampling points within the monitoring area and output a temperature sampling matrix.

[0036] The temperature sampling array unit 101 undertakes the crucial task of collecting temperature data, enabling it to simultaneously acquire temperature data from multiple spatial sampling points within the monitoring area, forming a temperature sampling matrix. The temperature sampling array unit 101 can employ an integrated two-dimensional infrared temperature array sensor, which boasts high integration and can quickly and accurately acquire temperature information; it can also consist of multiple independent sampling modules, allowing for flexible deployment based on actual needs; or it can function as a distributed temperature acquisition node, achieving comprehensive coverage of large areas. The sampling resolution and frame rate can be flexibly configured according to different application scenarios to meet monitoring requirements in various complex environments. The temperature sampling matrix represents the absolute temperature values ​​of all sampling points within the entire field of view at a given moment.

[0037] Understandably, all sampling points are sampled synchronously at the same time to ensure data consistency in the spatial dimension, accurately capturing instantaneous temperature changes during a fire and significantly improving system response speed. The non-imaging multi-point temperature sensor array design only outputs temperature numerical signals, avoiding privacy risks and significantly reducing data transmission volume and processing load, thus reducing hardware resource consumption. The system as a whole has a static architecture with no mechanical moving parts, fundamentally extending the service life of the equipment and eliminating the need for regular maintenance due to component wear, thereby reducing operation and maintenance costs.

[0038] An optical focusing component 102, located at the sensing front end of the temperature sampling array unit 101, is used to expand the detection field of view or enhance energy convergence in specific sensitive areas. The pixel distribution of the temperature sensing array 101 determines the sampling accuracy and coverage. The optical focusing component 102, mounted at the front end of the temperature sensing array 101, can expand the sampling angle or focus on specific areas according to actual needs. For example, in large-area monitoring scenarios, a wide-angle optical focusing component can be used to expand the field of view; when performing precise monitoring of specific equipment, a short-focal-length or better-focusing component can be selected. The supporting components of the optical focusing component 102 are used to fix and protect the temperature sensing array and the optical focusing component, ensuring their stability during operation. The signal output interface 205 of the optical focusing component 102 transmits the data collected by the temperature sensing array to the processing and control unit 103 for subsequent data processing and analysis. The temperature sensing array 101 and the optical focusing component 102 can be integrated into a single package to improve structural compactness and stability; alternatively, a multi-module distributed layout can be used to enhance system flexibility and scalability to adapt to different application scenarios.

[0039] Understandably, integrating the optical focusing component 102 into the front end of the temperature sensing array 101 enables large-scale monitoring within the coverage area of ​​a single device, eliminating the need for multi-device splicing deployment. This reduces hardware costs, simplifies on-site installation, and adapts to complex spatial layout requirements. The focusing capability of the optical focusing component can effectively gather infrared energy from the target area, reduce environmental interference signals, significantly improve the system signal-to-noise ratio, and enable the device to accurately detect distant heat sources or weak temperature anomalies, greatly improving the sensitivity of early fire identification.

[0040] The following section focuses on the processing and control unit 103. The data preprocessing module within the processing and control unit 103 is used to perform noise suppression, background temperature compensation, and outlier correction on the temperature sampling matrix.

[0041] Considering that the raw temperature matrix output by the temperature sampling array unit 101 is susceptible to noise interference, outliers, and ambient temperature fluctuations, it needs to be preprocessed to generate a clean, stable, and reliable temperature matrix. This ensures that the data accurately reflects the actual temperature distribution characteristics of the monitored area, providing a precise data foundation for subsequent fire assessment. The core function of noise suppression is to eliminate random and irregular micro-fluctuations in the data—these fluctuations mainly originate from the sensor's own electronic noise and external electromagnetic interference. The ultimate goal is to obtain a smooth data curve while preserving the true temperature change trend, avoiding interference signals from misleading the assessment results. In some optional embodiments, noise suppression can be achieved through median filtering: a 3×3 window is constructed with a single sampling point as the core, incorporating the temperature values ​​of that point and eight adjacent sampling points. After sorting, the median value is taken as the updated temperature value for that sampling point. This method is highly effective in suppressing impulse noise (also known as salt-and-pepper noise, which manifests as isolated outliers that differ significantly from surrounding data) and effectively protects edge information in the temperature field, preventing the smoothing and blurring of real, drastic temperature changes and ensuring the accuracy of identifying local abnormal heat sources. In some alternative embodiments, noise suppression can employ a Kalman filter scheme: this algorithm not only references the current temperature measurement value but also predicts the theoretical temperature value at the current moment based on the system's historical operating status. It then fuses the predicted and measured values ​​through a weighted average to output the optimal temperature estimate. Compared to simple averaging filtering, Kalman filtering offers superior smoothing for random noise and more accurately captures temperature change trends, making it particularly suitable for physical quantities like temperature that change relatively slowly, further enhancing data stability and reliability.

[0042] Furthermore, the core value of background compensation lies in eliminating the interference of ambient temperature changes on monitoring data and accurately highlighting the true abnormal temperature rise signal, which is a key technical step in reducing the false alarm rate of the system. Essentially, it eliminates the influence of "ambient baseline fluctuations," making fire-related temperature anomalies clearer and providing a more accurate temperature difference basis for subsequent judgments. In some optional embodiments, background compensation is implemented using the moving minimum method: for each sampling pixel or independent monitoring area, a historical temperature data queue of fixed duration (e.g., the past 10 minutes) is maintained, and the minimum temperature value in the queue is calculated in real time, and this value is used as the current background temperature estimate; finally, the compensated temperature difference data is obtained by calculating "current measured temperature - background temperature estimate". The advantage of this method is its simple implementation logic, low computational resource consumption, and ability to effectively track the slow drift of ambient temperature (e.g., diurnal temperature variation); however, it has limitations—if there is a long-term stable heat source (e.g., a constantly running device) in the monitoring area, its temperature may be gradually "included" in the background estimate, making it impossible to effectively identify further temperature rises based on that heat source. In some alternative embodiments, background compensation is achieved through time series prediction methods (such as exponential smoothing algorithms): based on the time series characteristics of historical temperature data, the background temperature at the current moment is predicted as an estimate, and then the temperature difference data is obtained by subtracting the estimated background temperature from the measured current temperature. The core logic of this method is similar to a low-pass filter, allowing only slowly changing signals such as ambient temperature to pass through, while filtering out rapid temperature rise signals related to fire. Its advantage is that the background estimation result is smooth and insensitive to instantaneous noise, but its disadvantage is that if the fire temperature rise rate is slow (such as in the early stage of smoldering), some temperature changes may be misjudged as environmental drift and "integrated" into the background, leading to a decrease in sensitivity. In some other alternative embodiments, background compensation adopts the non-interest area method: a "non-interest area" is pre-defined within the monitoring range—that is, an area where no fire or abnormal heat source will occur (such as static structural surfaces such as walls, ceilings, and floors). The average temperature of all sampling points in this area is calculated in real time and used as the globally unified background temperature estimate for the system; subsequent compensation is completed by subtracting the estimated global background temperature from the measured current temperature. This method is characterized by its intuitive concept, lack of complex algorithms, and avoidance of misclassifying permanent heat sources as background. However, it is highly dependent on the accuracy of region division—if a non-interested region experiences temperature anomalies due to external factors (such as direct sunlight on a wall or increased wall temperature near heating pipes), it will directly lead to a bias in the global background estimation, thereby causing distortion in the compensation data. In some alternative embodiments, background compensation can be achieved through a hybrid method, combining the advantages of the above three methods to form a combined scheme. For example, firstly, the global background reference value is calculated using the non-interested region method as the overall environmental benchmark; then, the moving minimum method is run separately for each sampling point, but the range of variation of its background estimate is constrained—limited to "global background reference value ± preset deviation threshold".This design retains the adaptability of individual sampling points to the local environment, while avoiding the problem of background estimation drifting infinitely and deviating from the overall environment due to long-term heating (such as continuous heat generation of the device) at a certain point through a global benchmark, thus achieving a balance between adaptability and stability.

[0043] Understandably, through the background temperature compensation mechanism, the system can achieve dynamic adaptation to ambient temperature: regardless of whether it's the sweltering heat of summer or the frigid cold of winter, even with significant differences in the ambient baseline temperature, as long as an abnormal temperature rise of 20°C occurs in the monitored area, the system can output a completely consistent alarm response. This fundamentally solves the problem of false alarms or missed alarms caused by seasonal changes and ambient temperature fluctuations, ensuring consistent judgments across seasons. The temperature difference data after background compensation can directly quantify the actual magnitude of the "abnormal temperature rise," rather than relying on absolute temperature values. This allows subsequent feature extraction (such as temperature rise rate and energy density calculation) and fire judgment logic to focus entirely on the real risk signals (such as abnormal increases in local heat sources), without needing to handle the interference caused by changes in ambient temperature, significantly simplifying algorithm complexity and improving judgment accuracy. The essence of this compensation process is to separate the "abnormal temperature rise" signal from the massive background signal of "ambient temperature": the small temperature rise differences that were originally masked by the ambient temperature (such as local low-temperature anomalies in the early stage of a fire) are highlighted, greatly reducing the interference of the background signal on the effective information, thereby significantly improving the data signal-to-noise ratio and laying the foundation for accurate identification of early fires.

[0044] Furthermore, the core function of outlier correction is to accurately identify and repair obvious erroneous data points (i.e., "bad pixels") caused by instantaneous and severe interference (such as insects briefly passing through the sensor's field of view or sudden electromagnetic pulse interference). These bad pixels are not true temperature changes, and if left untreated, they can easily lead to misjudgments. Therefore, targeted algorithms are needed to remove interference and restore the authenticity of the data. The bad pixel identification process is usually based on neighborhood difference comparison logic: taking a single sampling point as the center, the average temperature of the point and its eight surrounding sampling points is calculated. If the temperature deviation of the point from the average exceeds a preset threshold (such as 3 times the standard deviation, which can be flexibly adjusted according to the scenario), it is judged as an outlier and marked. Bad pixel repair is accomplished through data completion methods: the average or median of the surrounding valid sampling points can be used to directly replace the outlier, or the data can be calculated and completed based on the spatial temperature distribution characteristics of the surrounding points (such as linear interpolation or surface fitting) to ensure that the repaired data is consistent with the surrounding temperature trend.

[0045] Understandably, the key value of this correction process lies in preventing single, occasional interference events (such as brief insect obstruction) from directly triggering alarms, while ensuring the spatial continuity and logical rationality of the temperature matrix data. Its function is similar to editing an article to correct obvious typos, eliminating the misleading influence of "erroneous information" on subsequent judgments, and ensuring the accuracy of the data's "semantics" (temperature distribution trend), thus providing a reliable data foundation for subsequent feature extraction and fire situation determination.

[0046] The feature extraction module is used to extract time-domain and spatial-domain temperature feature parameters from the preprocessed data.

[0047] Among them, the temperature characteristic parameters include any one or more of the following: temperature rise rate, energy density, average temperature, maximum temperature, and spatial continuity index. The spatial continuity index is obtained by performing image connectivity analysis on the binarized high-temperature region.

[0048] It should be noted that the feature extraction module also includes a temperature change trend analysis submodule and an energy density calculation submodule. The temperature change trend analysis submodule is used to analyze the changing pattern of temperature characteristic parameters over time; the energy density calculation submodule is used to calculate the energy density distribution within the monitoring area based on the spatial distribution of the temperature sampling matrix.

[0049] Further explanation is needed regarding the specific calculation steps of the spatial continuity index. Step one: Binarization of the temperature matrix. The core purpose of binarization is to highlight high-temperature anomaly regions from the pre-processed clean temperature matrix, laying the foundation for subsequent connected component analysis. The specific implementation process is as follows: First, based on the temperature difference data after background temperature compensation, a high-temperature judgment threshold is dynamically set (for example, using the current background temperature as a baseline, superimposed with 30℃ as the high-temperature threshold; the threshold can be adaptively adjusted according to the application scenario); then, each pixel in the temperature matrix is ​​traversed, and the pixel temperature value is compared with the high-temperature threshold: if the pixel temperature is higher than the high-temperature threshold, the pixel is marked as "1" (representing a potential high-temperature region); if the pixel temperature is lower than or equal to the high-temperature threshold, the pixel is marked as "0" (representing a normal temperature region); after traversal, a binary matrix containing only "0" and "1" is output, and all pixels marked as "1" in the matrix together constitute the potential "fire zone" to be analyzed.

[0050] Step 2 in the calculation of spatial continuity index: Connectivity analysis of binary matrix. The core purpose of connectivity analysis is to identify the spatial clustering characteristics of "1" pixels in the binary matrix and distinguish between isolated high-temperature points and continuous high-temperature regions. The specific operation process is divided into two scans: ① First scan (label assignment and equivalent recording): Traverse the binary matrix from left to right and from top to bottom. When an unlabeled "1" pixel is detected, first check the upper neighbor pixel and the left neighbor pixel (i.e., spatially adjacent traversed pixels): If both neighbor pixels are "0", assign a new unique label (such as Label 1, Label 2, etc.) to the current "1" pixel; if any neighbor pixel is "1" and already has a label, assign the label of the neighbor pixel to the current "1" pixel; if the upper and left neighbor pixels are both "1" but have different labels, mark the current "1" pixel as one of the labels and record the "equivalence relationship" of the two different labels (indicating that they belong to the same connected component). ② Second Scan (Label Merging and Unification): Based on the equivalence relationships recorded in the first scan, the binary matrix is ​​traversed a second time, and all labels with equivalence relationships are uniformly replaced with the same label (e.g., equivalent Label 2 and Label 3 are both replaced with Label 2). The final output is a "label matrix". All "1" pixels with the same label in the matrix form an independent connected component, and each connected component corresponds to a potential continuous high-temperature region.

[0051] Step 3 of the spatial continuity index calculation: Connected Component Feature Extraction and Index Quantification. The core purpose of feature extraction and index quantification is to transform the spatial features of connected components into quantitative parameters that can be used for fire assessment. The specific operations are as follows: First, traverse all connected components in the label matrix and calculate the "connected area" of each connected component (i.e., the total number of "1" pixels contained in the connected component, which can be converted into the actual physical area by combining spatial mapping relationships); Second, select the "largest connected component" (i.e., the connected component with the largest connected area) from all connected components. Since real fires usually manifest as localized continuous heat accumulation, they are prone to forming large-scale continuous high-temperature areas. The features of the largest connected component can better reflect the potential fire risk; Finally, based on the above calculation results, generate at least one spatial continuity index. Optional indicators include: ① Maximum connected area (directly quantifies the scale of the largest continuous high-temperature area); ② Maximum connected area ratio (the ratio of the area of ​​the largest connected component to the area of ​​the entire monitoring area, reflecting the spatial proportion of the high-temperature area); ③ Shape irregularity (quantifies whether the shape of the high-temperature area conforms to the characteristics of fire spread by parameters such as the ratio of the perimeter and area of ​​the connected component and the area difference between the circumscribed rectangle and the actual area). The above indicators can accurately describe the spatial concentration of high-temperature areas, providing key spatial basis for subsequent fire assessment.

[0052] To clearly illustrate the application of spatial continuity indicators in fire situation assessment, a specific scenario example is provided below: Assuming the total number of pixels in the temperature matrix corresponding to the monitored area is 100, after binarization, there are a total of 15 pixels exceeding the high-temperature threshold (potential high-temperature points). The following analysis examines the judgment logic for three typical distribution scenarios: Scenario 1: After binarization of a locally compact clustered distribution, 15 high-temperature points are clustered together to form a compact connected domain (without scattered isolated points). In this case, the spatial continuity indicators are calculated as follows: ① The maximum connected area is 15 pixels (i.e., the area of ​​the unique connected domain); ② The percentage of the maximum connected area is 1.0 (maximum connected domain area ÷ total high-temperature point area = 15 ÷ 15). From the indicator characteristics, although the percentage of the maximum connected area is extremely high (indicating that the high-temperature points are not scattered), the absolute value of the maximum connected area is relatively small (only 15 pixels, corresponding to a small local area within the actual monitored area). Combined with the characteristic of "localized small-scale high-temperature clustering," the system can determine it as "localized overheating" (such as localized component heating in equipment) rather than a large-scale fire.

[0053] Scenario 2: After binarization, the 15 high-temperature points exhibit a linear or irregular spreading distribution, and all high-temperature points are interconnected, forming a narrow connected domain. The spatial continuity indicators are: ① The maximum connected area is 15 pixels (all high-temperature points belong to the same connected domain); ② The maximum connected area percentage is 1.0; ③ The irregularity value is high (because the connected domain exhibits a narrow spreading shape, the ratio of perimeter to area is large, or the area difference between the circumscribed rectangle and the actual connected domain is significant, consistent with the morphological characteristics of fire spread). Considering all indicators, the absolute size of the maximum connected area is moderate, accounting for 100% (no scattered high-temperature points), and the shape exhibits typical fire spread characteristics; therefore, the system can comprehensively determine it as a "fire".

[0054] Scenario 3: After binarization of a multi-regional dispersed distribution, 15 high-temperature points are dispersed into 3 independent small connected regions, each with an area of ​​5 pixels (no interconnection). The spatial continuity indicators are: ① Maximum connected area is 5 pixels (maximum area of ​​a single connected region); ② Maximum connected area percentage is 0.33 (maximum connected region area ÷ total high-temperature point area = 5 ÷ 15). Analysis of the indicator characteristics shows that the absolute value of the maximum connected area is small, and the percentage of the maximum connected area is low (indicating that the high-temperature points are dispersed across multiple regions), which does not conform to the spatial characteristics of a fire "localized clustering and gradual spread." It is more likely caused by multiple isolated interference sources in the environment (such as multiple small heat sources or transient electromagnetic interference). Therefore, the system can determine it as an "interference signal" to avoid false alarms.

[0055] Specifically, the feature extraction module plays a central control role in feature extraction, directly coordinating the collaborative operation of the temperature change trend analysis submodule and the energy density calculation submodule. For the received clean temperature matrix, the module can calculate basic features such as maximum temperature and average temperature by directly traversing the matrix. For key parameters such as energy density and temperature rise rate, the module obtains them by directly calling the output results of the energy density calculation submodule and the temperature change trend analysis submodule, eliminating the need for repetitive calculations and improving efficiency. After the feature data of each dimension is collected, the feature extraction module integrates the calculation results of all submodules (including basic features and key parameters), assembling them into a standardized one-dimensional feature vector. This ensures that the data format meets the requirements of subsequent processing and is ultimately accurately output to the fire situation determination logic module, providing a structured and complete feature input foundation for multi-parameter fusion determination.

[0056] Specifically, the core logic of the energy density calculation module is to quantify the degree of heat accumulation in the monitored area through the correlation between temperature and radiant energy. The specific calculation process is as follows: First, based on Planck's blackbody radiation law or a preset temperature-energy conversion model, the absolute temperature value of each pixel in the clean temperature matrix is ​​converted into the corresponding radiant energy value. Then, the radiant energy values ​​of all pixels in a single monitoring area are accumulated to obtain the total radiant energy of the area. Finally, combined with the physical area of ​​the monitored area (determined through the spatial mapping relationship calibrated by the system, such as the actual monitoring area corresponding to each pixel × the total number of pixels in the area), the average energy density of the area is obtained by calculating "total radiant energy ÷ physical area of ​​the area". This reflects the heat intensity per unit area and provides a quantitative basis for subsequent determination of "whether there is abnormal local heat accumulation".

[0057] The temperature change trend analysis module is used to accurately capture the rate of temperature change over time. In some optional embodiments, the temperature rise rate can be calculated by linear fitting. The specific implementation is as follows: First, a time window parameter is set, and temperature data from the most recent N consecutive sampling periods are selected (e.g., when the sampling period is 1 second, N=5, i.e., 5 sets of temperature data from the most recent 5 seconds are taken) to form a time-temperature data sequence; then, the least squares method is used to linearly fit the data sequence to construct a "temperature-time" linear equation. (where y is temperature, x is time, k is slope, and b is intercept); finally, the slope k of the fitted line is the temperature rise rate (unit: ℃ / second) within that time period. This method, through linear regression of multiple sets of historical data, can effectively offset the influence of random noise within a single sampling period. Compared with directly calculating the temperature difference between two adjacent samples, the obtained temperature rise rate is smoother and more stable, and can more realistically reflect the long-term temperature change trend, avoiding misjudgments caused by instantaneous data fluctuations.

[0058] The fire situation determination logic module is used to generate a fire situation determination signal based on the fusion analysis results of time-domain and spatial-domain temperature characteristic parameters.

[0059] The determination signals are tiered fire situation determination signals, including early warning signals, fire confirmation signals, and fire persistence signals. The fire situation determination employs a dual-time-window comparison mechanism, using the trend difference between short and long time windows to determine transient heat sources and persistent fires.

[0060] To clearly explain the multi-parameter fusion analysis mechanism of the fire situation determination logic module, the following will elaborate on three core fusion analysis strategies. Each strategy will clearly define its implementation logic, operation process, and parameter optimization methods to ensure the operability and accuracy of the determination process.

[0061] Rule-based fusion analysis, centered on a pre-defined rule base, determines fire situations through "feature threshold matching + logical condition combination." The specific process is as follows: Based on experimental data, industry standards, and field experience, multi-dimensional feature parameter thresholds and logical combination rules are pre-defined. Typical rule examples are as follows: If the following conditions are met: "temperature rise rate > threshold α", "energy density > threshold β", "spatial continuity index > threshold γ", and "duration of the above state > time Δt", then the fire is confirmed. If the conditions of "temperature rise rate > threshold α" or "maximum temperature > threshold θ" are met (threshold θ is set according to the environmental scenario, such as 80℃ for electrical equipment scenarios and 60℃ for civil spaces), it is determined to be a "suspected fire". If none of the above conditions are met, it is determined to be a "normal state".

[0062] The fire situation determination logic module receives the feature vector (including temperature rise rate, energy density, spatial continuity index, etc.) output by the feature extraction module in real time, compares each feature value with the corresponding threshold in the rule base one by one, and then judges whether the preset conditions are met according to the logical operator ("AND / OR"), and finally outputs the judgment result of "confirmed fire", "suspected fire" or "normal".

[0063] The initial thresholds (α, β, γ, Δt, θ, etc.) in the rule base are set based on experimental data and experience. In the later stages, dynamic thresholds and learning modules can be used to iteratively optimize the rules by combining environmental changes in actual operation and historical judgment results (such as false alarms / missed alarms) to ensure that the rules are adapted to the operational needs of different scenarios.

[0064] Weighted scoring fusion analysis achieves quantitative fire situation assessment by assigning weights to different characteristic parameters and calculating a comprehensive risk score. The core logic and operational steps are as follows: Based on historical fire data statistical analysis and the experience of domain experts, weight coefficients are assigned to key characteristic parameters (such as temperature rise rate T, energy density E, and spatial continuity S). , , ), and satisfy (For example, in the electrical equipment scenario, the weight of temperature rise rate is set to 0.4, energy density to 0.3, and spatial continuity to 0.3 to highlight the priority of abnormal temperature rise); Set two levels of judgment thresholds: warning threshold (e.g., 60 points) and alarm threshold (e.g., 80 points, with the total score set on a 100-point scale). The thresholds need to be adjusted in combination with the risk level of the scenario (e.g., the alarm threshold for the energy storage cabin scenario can be reduced to 70 points to improve sensitivity).

[0065] The fire situation assessment logic module acquires the normalized values ​​of each characteristic parameter in real time (converting characteristic values ​​of different dimensions into a unified range of 0-100), and substitutes them into the following formula to calculate the comprehensive risk score:

[0066] in, To calculate the overall risk score, , , These are the normalized eigenvalues ​​for temperature rise rate, energy density, and spatial continuity, respectively. , , These are the weighting coefficients.

[0067] If the overall risk score is greater than or equal to the alarm threshold, a "fire confirmed" response will be triggered (such as activating fire extinguishing devices or cutting off power). If the warning threshold is less than or equal to the comprehensive risk score and less than the alarm threshold, a "suspected fire" response will be triggered (such as an audible and visual warning, or a notification to the monitoring platform). If the overall risk score is less than the warning threshold, it is judged as "normal state".

[0068] The initial weight coefficients and judgment thresholds are set based on experience. In the later stages, they can be optimized by using dynamic thresholds and learning modules, combined with the judgment accuracy in actual operation (such as adjusting the weights to reduce the false alarm rate), to ensure the adaptability of the scoring system.

[0069] Machine learning fusion analysis learns the nonlinear mapping relationship between features and fire conditions by training mathematical models to achieve intelligent judgment. The specific process covers three stages: "model training - deployment - real-time judgment". Based on the system's computing power requirements and judgment accuracy requirements, a lightweight model (such as Support Vector Machine (SVM), Random Forest, or single-hidden-layer neural network) is selected to avoid complex models occupying too many embedded resources. A large amount of labeled historical data is collected, including feature vectors (containing multi-dimensional features such as temperature rise rate, energy density, spatial continuity, and maximum temperature) under states such as "fire", "local overheating", "interference", and "normal", to ensure that the dataset covers different scenarios (such as high-temperature environment and electromagnetic interference environment) and fault types. The selected model is trained using labeled datasets, and the internal parameters of the model (such as kernel function parameters of SVM and number of trees in random forest) are adjusted through cross-validation until the model's judgment accuracy (such as F1 score) on the test set reaches the preset standard (such as ≥95%). Finally, a lightweight model file (such as TensorFlow Lite format) is generated.

[0070] Deploy the trained model file to the fire situation determination logic module (embedded processor or edge computing unit) to ensure that the module has real-time loading and inference capabilities.

[0071] The real-time determination process is as follows: The fire determination logic module receives the feature vector output by the feature extraction module in real time and inputs it into the loaded machine learning model; the model outputs the determination result through inference: if it is a classification model, it directly outputs category labels such as "fire" and "normal"; if it is a probability model, it outputs the "fire probability value" (e.g., 92%); a probability determination threshold is set (e.g., 90%): if the fire probability is ≥ the threshold, it is determined as "confirmed fire"; if 50% ≤ probability < 90%, it is determined as "suspected fire"; if the probability < 50%, it is determined as "normal state".

[0072] Later, new data from the field operation (including judgment results and actual status) can be uploaded to the cloud through the communication module. The model can be incrementally trained regularly to optimize model parameters and improve the judgment accuracy in complex scenarios (such as adapting to feature shifts in extreme temperature and humidity environments).

[0073] As can be seen from the above introduction, the three fusion analysis strategies each have their own focus in terms of technical characteristics. The core advantage of rule-based fusion analysis lies in the strong interpretability of the judgment logic—each judgment result can correspond to a clear rule clause, which makes it easy for technicians to trace the judgment process and troubleshoot problems. At the same time, this strategy is easy to implement in hardware, requiring only basic logic operation units and has a small computational load, and can achieve millisecond-level response on low-power embedded platforms. However, the limitations of this strategy are also quite obvious. The setting of its rules and thresholds is highly dependent on expert experience and previous experimental data. For complex scenarios that are not predefined in the rule base (such as the superposition of multiple interference signals, irregular spread patterns of atypical fires), the judgment flexibility is insufficient, and it is difficult to characterize the implicit nonlinear correlation between feature parameters.

[0074] The advantage of weighted scoring fusion analysis lies in its high flexibility. By adjusting the weight coefficients of different feature parameters, it can quickly adapt to the judgment requirements of different application scenarios. For example, in industrial equipment monitoring scenarios, the weight of energy density can be increased to focus on local high heat accumulation, while in civil building scenarios, the weight of temperature rise rate can be increased to quickly capture sudden temperature changes. At the same time, the comprehensive risk score can intuitively quantify the risk level. Parameter adjustment does not require reconstructing the overall judgment logic; only the weights and thresholds need to be optimized. However, the initial weights of this strategy still need to be set based on experience, and the linear weighting calculation method cannot accurately reflect the complex nonlinear interaction relationship between features. For example, the synergistic effect of temperature rise rate and spatial continuity at different stages of fire spread may lead to judgment bias in some complex scenarios.

[0075] The core advantage of machine learning fusion analysis lies in its adaptability and adaptability to complex scenarios. Through training with a large amount of labeled historical data, the model can automatically learn the complex mapping relationship between feature parameters and fire status, including nonlinear patterns of multi-feature coupling. It can still maintain a high accuracy rate in complex scenarios such as atypical fires and multiple interferences, and its performance can be continuously optimized through incremental training. However, its limitations are mainly reflected in three aspects: First, model training requires the accumulation of sufficient labeled data covering different scenarios and different fault types, which results in high data collection and labeling costs. Second, the computational load of the model inference process is relatively large, requiring edge computing units or high-performance embedded processors to ensure real-time performance. Third, the "black box" nature of the model leads to poor interpretability of the judgment results, making it difficult to clarify the influence weight of individual feature parameters on the final judgment result, which is not conducive to fault tracing and logical verification.

[0076] In practical applications, the three fusion analysis strategies are not mutually exclusive, but can be combined to form a more efficient and accurate judgment scheme. For example, a collaborative model of "rule filtering + machine learning for fine-grained judgment" can be adopted: the system first performs preliminary screening of data through rule-based fusion analysis. For scenarios that clearly meet the rules of "normal state" or "clear fire situation", the judgment result is directly output to reduce the consumption of computing resources. For "suspected" scenarios where the feature parameters are at the threshold or there are multiple interference signals superimposed, the machine learning model is further invoked for deep reasoning. The accurate judgment result is output by combining the complex correlation between features. This ensures the real-time performance of conventional scenarios and improves the judgment accuracy of complex scenarios. For example, a hybrid framework of "machine learning-optimized weighted scoring" can be constructed: based on weighted scoring fusion analysis, but its weight coefficients no longer rely on human experience to set, but are trained on historical data through machine learning algorithms (such as gradient boosting trees and logistic regression) to automatically learn the optimal weights of each feature parameter under different environments and scenarios, forming a dynamic weight library; when the system is running, it automatically calls the matching weight coefficients to calculate the comprehensive risk score based on parameters such as the background temperature and scenario type of the current monitoring environment, which not only retains the advantages of intuitive quantification and convenient adjustment of weighted scoring, but also improves the scientific nature and scenario adaptability of weight setting through machine learning.

[0077] It's important to note that the fire assessment employs a dual-time-window comparison mechanism. This mechanism distinguishes between transient heat sources and ongoing fires by analyzing the trend differences between short and long time windows. The short time window is used to quickly detect any abrupt temperature increases with very high sensitivity, allowing for rapid calculation of a short-term temperature rise rate. The long time window is used to determine whether a short-term abrupt change is the start of a sustained process or merely an isolated event. The long time window can calculate a long-term temperature rise trend (e.g., the slope of the temperature-fitted line during this period). The core logic of this dual-time-scale comparison strategy is that a genuine fire not only exhibits short-term, violent outbreaks (abruptness) but also a sustained development process (trend), while most disturbances only exhibit abruptness without a trend.

[0078] The dual-time-window determination strategy compares and analyzes the instantaneous temperature changes within a short time window with the temperature trend characteristics within a long time window, and combines different combinations of the two to achieve accurate determination of the heat source properties. The specific application logic is as follows: In the face of instantaneous interference scenarios (such as a sudden hot wind blowing through the monitoring area or a short-term strong light shining on the sensor), a short time window will capture significant temperature fluctuations - at this time, the short-term temperature rise rate will increase sharply, even exceeding the preset alarm threshold, and the signal path progresses according to the logic of "detecting a short-term mutation → entering the short-term high threshold judgment"; however, the temperature trend within the long time window always remains stable, without an obvious upward slope or continuous fluctuations, only remaining near the ambient background temperature. Through the characteristic combination of "short-term mutation + long-term stability", the system can determine that the heat source is an instantaneous interference, and finally does not trigger an alarm action, effectively avoiding false alarms caused by accidental factors and ensuring the stability of the determination.

[0079] When a continuous fire breaks out in the monitoring area (such as the combustion of combustibles such as paper and wood), the short time window will continuously detect a high temperature rise rate - not only does the short-term temperature mutate, but it can also maintain a high value in subsequent sampling cycles, and the signal path gradually progresses from "detecting a short-term mutation → short-term high threshold judgment (not falling back) → confirming short-term continuous abnormality"; at the same time, the temperature trend within the long time window shows a clear upward characteristic, and as the fire develops, the long-term temperature rise slope continuously increases, and the signal evolves according to the logic of "detecting a long-term trend change → confirming long-term continuous temperature rise". Through the double characteristic verification of "short-term continuous abnormality + long-term continuous temperature rise", the system can determine the confirmed fire with a high confidence level and immediately trigger alarm and linkage control (such as starting a fire extinguishing device and cutting off the power supply) to ensure the timeliness and accuracy of the fire response.

[0080] In scenarios of slow smoldering or early overheating of equipment (such as slow heating of the wire insulation layer or slow smoldering of accumulated dust), the temperature change within the short time window is relatively gentle - the short-term temperature rise rate does not reach the alarm threshold, and the signal directly enters the path of "no short-term mutation → short-term low threshold observation"; however, the long time window will capture a slow but continuous temperature rise trend. Although the temperature change within a single sampling cycle is微小, it will form an obvious upward curve after accumulating for a period of time. Based on the characteristic combination of "no short-term mutation + long-term slow rise", the system will determine it as a potential risk and trigger a warning signal (such as pushing a warning message to the monitoring platform and starting local audible and visual alerts), prompting the operation and maintenance personnel to pay attention to the status of this area, achieving early warning of fire or overheating hazards, and争取时间 for hazard disposal.

[0081] Under normal operating conditions (without abnormal heat sources or interference factors in the monitored area), the temperature within a short time window shows no drastic fluctuations, oscillating slightly around the ambient background temperature without triggering any short-term anomaly detection. Similarly, the temperature trend within a long time window remains stable, with a long-term temperature rise slope close to zero, closely matching the ambient background temperature model. By combining these "short-term stability + long-term stability" characteristics, the system determines that the monitored area is in a normal state, maintaining only routine sampling and monitoring without executing any alarms or warnings, ensuring low-power, interference-free operation when there is no risk.

[0082] Furthermore, the processing and control unit also includes a dynamic threshold and learning module, which is used to adaptively adjust the judgment threshold and fusion feature weights for fire determination based on changes in ambient temperature and historical fire data.

[0083] The dynamic threshold and learning module includes: an input feature set for caching historical temperature feature parameters; a feature normalization unit for normalizing temperature feature parameters; an online learning algorithm unit for outputting optimized judgment thresholds and feature weights based on the normalized feature parameters and feedback signals using an online machine learning algorithm; a parameter update unit for smoothly updating the optimized parameters to the fire judgment logic module; and a local storage unit for non-volatile storage of system parameters and historical data.

[0084] Specifically, the core function of the input feature set is to provide a high-quality training data source for the system's self-learning process. It does not store historical data indefinitely, but rather employs a dynamic storage mechanism using a sliding window or circular queue—caching only the feature vectors (including multi-dimensional parameters such as temperature rise rate, energy density, and spatial continuity indices) output by all feature extraction modules within the most recent preset time period (e.g., the past 24 hours). This design ensures that the training data covers recent environmental characteristics and operational status, reflecting the actual situation of the current scenario, while avoiding problems such as excessive storage resource consumption and decreased data timeliness caused by excessive data accumulation. This provides a precise and fresh input data foundation for subsequent learning algorithms.

[0085] The key role of the feature normalization unit is to address the judgment bias caused by differences in the dimensions and numerical ranges of different feature parameters. For example, the unit of temperature rise rate is "℃ / second" (the numerical range may be 0-10), the unit of energy density is "W / m²" (the numerical range may be 0-1000), and the spatial continuity index may be a dimensionless ratio (the numerical range is 0-1). If the original data is used directly for training, features with large numerical ranges tend to dominate the learning process, weakening the role of other features. Therefore, this unit performs standardization (such as Z-Score normalization) or normalization (such as Min-Max scaling) to uniformly map all feature parameters to the same numerical range (such as 0-1 or -1-1), ensuring that the learning algorithm can fairly and efficiently evaluate the impact of each feature on the judgment result, thereby improving the accuracy and stability of model training.

[0086] The online learning algorithm unit is a crucial component for achieving system self-optimization. Its function is to analyze historical error cases (false alarms and missed alarms) to summarize experience and generate better judgment parameters (thresholds, weights). This unit does not run continuously but initiates the learning process through an event-triggered mechanism. Specific triggering events fall into three categories: first, maintenance personnel confirm through the human-machine interface that an alarm is a false alarm (e.g., an erroneous alarm caused by interference signals); second, the system discovers missed alarms (i.e., an actual fire occurred but the system did not trigger an alarm) through multi-sensor linkage verification (e.g., combining smoke sensor data with on-site manual confirmation); and third, routine optimization performed by the system at a preset cycle (e.g., weekly) (fine-tuning parameters based on recent normal operation data and a small number of abnormal cases).

[0087] Taking the learning process triggered by a false alarm event as an example, the specific process is as follows: When a false alarm is confirmed, the system first extracts the feature data sequence of a specific time period before and after the false alarm event (such as 10 seconds before the false alarm to 5 seconds after the false alarm) from the input feature set, and labels these data as "negative samples" (clearly marking that the feature pattern does not correspond to a real fire). Then, the learning algorithm performs in-depth analysis of the feature pattern of the negative sample. If it finds that its typical feature is "high temperature rise rate, but energy density is significantly lower than the conventional fire threshold, and poor spatial continuity index" (such as interference caused by fast-moving small heat sources such as lighters), the algorithm will adjust the judgment parameters accordingly. At the threshold level, it may slightly increase the judgment threshold of energy density and spatial continuity index to avoid low-energy, dispersed temperature anomalies triggering alarms. At the weight level, it may reduce the weight of temperature rise rate in the fusion judgment, while increasing the weight of energy density and spatial continuity, strengthening the consideration of "degree of heat accumulation" and "spatial distribution characteristics", and ensuring that when similar feature patterns are encountered in the future, the probability of the system outputting "non-fire" is significantly increased.

[0088] The core design goal of the parameter update unit is to ensure the stability of the system's decision-making performance. Instead of directly overwriting old parameters with new ones, it uses a smooth transition strategy to iterate parameters. Specifically, it calculates the new parameters using a weighted average formula: "New parameter = β × Old parameter + (1-β) × New parameter learned," where β is a smoothing factor (usually close to 1, such as 0.9). This approach allows the new parameters to take effect gradually and in small increments. For example, if the temperature rise rate accounts for 0.4 of the old weights, and the new weight learned is 0.2, with β set to 0.9, the updated weight will be 0.9 × 0.4 + 0.1 × 0.2 = 0.38, only 0.02 lower than the old value. This strategy effectively avoids drastic fluctuations in the system's decision-making logic caused by sudden parameter changes, preventing the introduction of new false alarms and false negatives, and ensuring the system maintains stable operation throughout the learning and optimization process.

[0089] The local storage unit uses non-volatile storage media (such as FLASH flash memory) to persistently store the system's "learning results" and key operational data. The specific storage content includes three categories: First, all judgment parameters currently used by the system, such as thresholds for each feature and weight coefficients for fusion judgments, ensuring that the latest parameters can be directly loaded after a power outage and restart, avoiding the loss of accumulated intelligence. Second, representative false positive and false negative case feature data (including labels), which can serve as samples for subsequent incremental learning, helping the algorithm to further optimize. Third, metadata about the learning process, such as the number of learning triggers, the parameter adjustment range for each learning iteration, and changes in judgment accuracy after learning, providing data support for operations and maintenance personnel to evaluate the system's self-learning effectiveness and troubleshoot problems.

[0090] Understandably, this module, through dynamic adaptation and continuous learning mechanisms, endows the system with significant environmental adaptability and intelligent evolution capabilities. On the one hand, it can automatically adjust the judgment criteria for the unique environmental characteristics of the deployment scenario (such as the large temperature difference brought about by the changing seasons, the periodic temperature fluctuations caused by the alternation of day and night, and the inherent heat source patterns of specific areas—such as the heat dissipation of equipment in industrial workshops and the operation mode of heating and cooling systems in civil buildings). This ensures that the system can accurately identify fire signals and avoid environmental interference under any complex environmental conditions, maintaining a high judgment accuracy rate. On the other hand, with the help of the "lifelong learning" function, the system can achieve continuous optimization—no longer a fixed-function device that is "fixed at the factory," but one that can learn from false alarms and missed alarms during operation, continuously improve the judgment logic through parameter iteration, and form an intelligent agent that continuously evolves over time. The core value of this design lies in three aspects: First, the system performance will not degrade due to long-term operation or environmental changes. On the contrary, with the accumulation of experience data, the accuracy and reliability of judgment will gradually improve, effectively extending the practical value cycle of the equipment. Second, it significantly reduces the reliance on expert experience—no technicians are required to repeatedly debug parameters and optimize rules for different scenarios, reducing the complexity of on-site deployment and the manpower cost of later maintenance. Third, through autonomous adaptation and self-optimization, it avoids performance bottlenecks caused by insufficient environmental adaptability, allowing the system to play a stable role in diverse application scenarios, further expanding the applicability of the technology.

[0091] The output interface unit 104 is connected to the processing and control unit 103 and is used to receive the judgment signal and convert the judgment signal into a corresponding control signal for output to an external actuator. The output interface unit 104 is connected to an external actuator 107, which includes an audible and visual alarm, an automatic fire extinguishing device, a power cut-off device, and other safety linkage devices.

[0092] Specifically, the output interface unit primarily serves for signal conversion and electrical isolation. Decision signals from the processing and control unit are typically low-voltage, low-current digital signals, such as 3.3V or 5V TTL levels, possibly transmitted via GPIO (General Purpose Input / Output) or specific communication protocols (such as I2C or SPI). Control signals output by the output interface unit to drive external actuators are usually switching signals requiring high voltage and high current, such as switching on / off 220V AC or 24V DC. Electrical isolation is to prevent interference from high voltage, surge currents, etc., in the external actuator circuitry from feeding back to the core processing and control unit, causing system restarts, crashes, or even hardware damage. The most common and effective method is to use optocouplers or relays.

[0093] To clearly illustrate the hardware structure and working mechanism of the output interface unit, we will now provide an example based on a specific module type: The output interface unit is typically designed as an integrated board, and its core includes three types of functional modules: relay output module, solid-state relay output module, and open-circuit gate output module. The appropriate output method can be flexibly selected according to the type of external actuator (such as high-voltage equipment, low-voltage devices, or signal indicating devices). The specific working principle is as follows: The relay output module is mainly used to drive external actuators requiring high voltage and high current control (such as solenoid valves in automatic fire extinguishing systems and power disconnect contactors). Its operation is based on electromagnetic induction and mechanical contact action: the microcontroller (MCU) of the processing and control unit outputs a low-level control signal through the GPIO pin. This signal first drives the transistor or MOSFET (metal-oxide-semiconductor field-effect transistor) on the board to conduct; after the transistor / MOSFET conducts, the relay coil receives operating current and generates electromagnetic attraction, pulling the internal normally open contact to close; after the contact closes, the power supply circuit or control circuit of the external actuator is connected, thereby triggering the preset action (such as energizing and starting the fire extinguishing solenoid valve, or disconnecting the contactor to cut off the equipment power). The advantage of this module is that the mechanical contacts can withstand higher voltage and current (such as AC220V, DC24V / 10A) and have good electrical isolation performance, making it suitable for control in high-voltage scenarios.

[0094] The solid-state relay output module adopts a fully electronic structure (no mechanical contacts), suitable for scenarios requiring high switching speeds, frequent operation, or where mechanical wear is a concern (such as ventilation system fan control and high-frequency alarm device drive). Its working principle is based on power electronic devices and opto-isolation technology: the module integrates an optocoupler isolation unit and power electronic switching devices such as thyristors (SCRs) and triacs (TRIACs). When the microcontroller's GPIO outputs a control signal, the optocoupler unit first converts the electrical signal into an optical signal, achieving electrical isolation between the control terminal and the high-voltage output terminal (preventing strong electrical interference from entering the control circuit). The optical signal triggers the power electronic devices to conduct, thereby connecting the circuit of the external actuator. Compared to traditional relays, this module has no mechanical wear, a fast switching response speed (microseconds), no operating noise, and a longer lifespan, making it suitable for scenarios requiring long-term, high-frequency operation.

[0095] OC (Open Collector) gate modules are primarily used to drive low-voltage DC small devices (such as LED status indicators, small buzzers, and signal indicators). Their working principle is based on the switching characteristics of a transistor: the module's output is equivalent to the collector of an NPN transistor, with the emitter directly grounded. When the microcontroller's GPIO outputs a high-level signal, the transistor's base receives current and saturates, allowing the external device (such as an LED) to form a loop with the OC gate output through an external pull-up resistor. Current flows through the pull-up resistor and the external device into the transistor's collector, ultimately grounding from the emitter, driving the device to operate (e.g., lighting an LED or emitting a buzzer). When the GPIO output is low, the transistor is cut off, the loop is broken, and the device stops working. The advantages of this module are its simple circuit structure, controllable output current (typically supporting mA-level current), and ability to directly drive low-voltage, low-power devices without requiring additional power amplifier units, saving board space and cost.

[0096] The communication module 105 is used to upload system status and fire situation assessment data to a monitoring platform or cloud server. The communication module supports wired interfaces or wireless communication protocols, including at least one of RS485, CAN, NB-IoT, Wi-Fi, and LoRa communication protocols.

[0097] Specifically, the communication module 105 uploads key information such as local fire situation assessment results (e.g., "fire confirmed"), real-time characteristic data (temperature rise rate, energy density, etc.), and system status (online, offline, self-check anomaly) to a remote monitoring center, cloud platform, or user's mobile app in real time. This enables unattended operation and centralized monitoring, allowing managers to monitor the safety status of multiple monitoring points remotely. Once a fire is confirmed, in addition to local audible and visual alarms, the communication module immediately sends the highest-priority alarm information to the cloud platform, which can automatically push SMS, phone, and app notifications to relevant personnel. This creates a multi-layered, three-dimensional alarm network, significantly shortening alarm response time and providing valuable time for firefighting and evacuation. It also uploads device self-check information (e.g., lens contamination, sensor drift) to prompt maintenance; the cloud platform can collect massive amounts of operational data for big data analysis, optimizing global algorithm models and threshold parameters. This enables predictive maintenance and global intelligent upgrades, moving from "single-point intelligence" to "network intelligence." The platform can use the communication module to perform parameter configuration, firmware upgrades, and restarts on remote devices. This greatly reduces the deployment and maintenance costs of the system, eliminating the need for on-site technical personnel.

[0098] The power supply and self-test module 106 is used to power the system and realize automatic calibration and health status monitoring. The power supply and self-test module 106 integrates non-uniformity correction function, background temperature adaptive update function, contamination detection function, and temperature drift compensation function.

[0099] Specifically, during the manufacturing process of temperature sampling arrays, even using the same technology, there will be slight hardware differences among thousands of sensor pixels—even under identical infrared radiation conditions, the output signals of each pixel will still deviate. This deviation leads to fixed "pockmark" noise in the temperature matrix generated by the system, severely interfering with the accuracy of temperature measurement. The core objective of the non-uniformity correction function is to eliminate the inherent baseline error (zero drift error) and gain error of each pixel, ensuring that all pixels maintain a consistent output response to the same temperature. The specific implementation is as follows: upon power-on or at a preset cycle (e.g., once daily), the system quickly switches to a uniform temperature blackbody reference source at the sensor front end via a built-in mechanism (usually an electric shutter). At this time, the output values ​​of all pixels at this uniform temperature are collected, and combined with the standard temperature of the blackbody reference source, a unique correction coefficient (including offset and gain value) for each pixel is calculated, and a corresponding correction parameter table is established and stored locally. During subsequent normal temperature measurement, the system uses the raw data collected from each pixel in real time to call its corresponding correction coefficient for calculation, thereby outputting an accurate temperature value after eliminating non-uniformity deviations.

[0100] Because ambient temperature dynamically changes with seasons (spring, summer, autumn, winter), time of day (day and night), and spatial scenarios (indoor and outdoor transitions), if the system uses a fixed "normal temperature" benchmark, it can easily lead to a decrease in the sensitivity to identify "abnormal temperature rises." For example, in summer when the ambient temperature is 35℃, a local temperature of 55℃ might be considered abnormal, while in winter when the ambient temperature is 5℃, a local temperature of 55℃ would clearly be a fire signal. The background temperature adaptive update function continuously learns and dynamically updates the background temperature in the current environment, providing an accurate reference benchmark for judging abnormal temperature rises. Specifically, the system statistically analyzes the temperature data of preset "non-concern areas" (such as walls, ceilings, and other static areas determined to have no fire risk) or covers all pixels within the monitoring area in real time. By calculating the moving average temperature (such as the average value within a sliding window) or the moving minimum temperature, a dynamically updated background temperature model is established. When the ambient temperature drifts slowly (such as changes in room temperature caused by day-night temperature differences), the background temperature model is updated synchronously to ensure that the benchmark temperature used in subsequent judgments always matches the current environment.

[0101] In harsh environments such as industrial workshops and kitchens, the optical windows (e.g., lenses, protective covers) of temperature sensors are easily covered by contaminants such as oil, dust, and particulate matter, causing infrared energy to attenuate during transmission. Even if there is an abnormal high temperature in the monitored area, the infrared signal received by the sensor will be weakened, resulting in generally lower temperature readings and increasing the risk of missed fire detection. The core objective of the contamination detection function is to automatically identify whether the transmittance of the optical window has decreased and to provide timely warnings of maintenance needs. Specifically, during the Non-Uniformity Correction (NUC) process, the system simultaneously acquires the current measured temperature of the blackbody reference source and compares it with the standard measured value during factory calibration or historical measured values ​​under clean optical window conditions. If the current measured temperature is significantly lower than the standard value, and the readings of all pixels show a synchronous downward trend (excluding single pixel failures), it is determined that the optical window is contaminated. Further, based on the magnitude of temperature attenuation (e.g., a measured value more than 5°C lower than the standard value indicates mild contamination, and more than 10°C lower indicates severe contamination), the contamination level is classified, and corresponding maintenance prompt signals are triggered (e.g., a warning is pushed for mild contamination, and an audible and visual alarm is triggered for severe contamination).

[0102] Temperature sensor performance slowly "drifts" over time due to factors such as component aging and changes in operating temperature (e.g., sensor overheating from prolonged operation). Even with a clean optical window and stable ambient temperature, sensor readings will gradually deviate from the true temperature, potentially leading to threshold failure over time. Temperature drift compensation addresses this long-term drift and maintains sensor accuracy. Specifically, the system pre-establishes a compensation function model for sensor readings and key influencing factors (including sensor temperature and cumulative operating time). During periodic non-uniformity correction (NUC), the system compares the collected blackbody reference source measurements with the model's predictions, iteratively optimizing the compensation function parameters using a fitting algorithm to update the drift model. During normal temperature measurement, the system continuously collects the sensor's temperature and cumulative operating time, inputting these data into the updated drift model to perform compensation calculations on the corrected temperature data for each pixel, ultimately outputting an accurate temperature value that eliminates drift errors.

[0103] Understandably, the non-uniformity correction function ensures the accuracy of temperature data from the source, which is the quality foundation for all intelligent judgments of the system; the background temperature adaptive update function enables the system to adapt to different seasons and climates, avoiding sluggishness due to high ambient temperature in summer and oversensitivity due to low ambient temperature in winter, and always maintaining consistent sensitivity; the contamination detection function enables predictive maintenance; the temperature drift compensation function ensures the measurement stability and accuracy of the system during long-term operation (months or even years), extending the effective life of the system.

[0104] The multi-point temperature sampling array fire detection system is suitable for at least one application in building fire protection, industrial safety, energy storage equipment protection, rail transit environmental monitoring, and smart home security.

[0105] The multi-point temperature sampling array fire detection system proposed in this application adopts a non-rotating structure, enabling multi-point parallel sampling without the need for mechanical moving parts. This design eliminates the risk of failure caused by rotational wear and positional offset, greatly improving the system's lifespan and stability. During long-term operation, performance degradation or failure will not occur due to wear of mechanical parts, reducing maintenance costs and downtime. It is suitable for applications with extremely high reliability requirements. In data acquisition, only temperature data is collected, without involving images or visual information. This non-imaging scheme avoids privacy leaks that may arise from image acquisition, while effectively monitoring fires and ensuring the safety of personnel and property. Combining multi-dimensional parameters such as temperature rise rate, energy density, maximum temperature, and spatial continuity, a comprehensive threshold and time-based method are used. The continuous logic significantly reduces false alarms and missed alarms. Through comprehensive analysis of multiple parameters, it can more accurately judge the fire situation and avoid misjudgments that may be caused by judging a single parameter. The judgment threshold and weight are automatically adjusted according to environmental changes and historical data to achieve algorithm self-adaptation. This not only allows the system to maintain stable judgment performance during long-term operation, but also continuously optimizes the judgment parameters to improve the ability to judge the fire situation and reduce the occurrence of false alarms and missed alarms. The communication module, power supply and self-test module, and dynamic threshold and learning module have been expanded. These modular hardware designs give the system good scalability and can be adapted to different scenarios. The combination of multi-point parallel sampling and embedded computing makes the response time less than 1-2 seconds, which can realize early fire identification and automatic linkage control, improving the efficiency and accuracy of fire response.

[0106] This addresses issues related to response speed, structural complexity, cost, and privacy protection.

[0107] Next, referring to the accompanying drawings, the method for determining fire conditions using a multi-point temperature sampling array according to embodiments of this application is described.

[0108] Figure 2 This is a flowchart illustrating the fire detection method using a multi-point temperature sampling array according to an embodiment of this application.

[0109] In step S101, temperature data from multiple discrete spatial sampling points within the monitoring area are simultaneously collected by the temperature sampling array unit 101 to generate a temperature sampling matrix.

[0110] In step S102, the temperature matrix is ​​preprocessed and spatially mapped to establish the correspondence between sampling points and physical monitoring areas.

[0111] Data preprocessing includes noise suppression, background temperature compensation, and outlier correction for the temperature sampling matrix. Spatial mapping is used to establish the correspondence between the position (array coordinates) of each pixel on the sensor array and the physical area in the actual monitoring space.

[0112] Specifically, the spatial mapping is implemented as follows: Before the temperature sampling array leaves the factory, a rigorous optical calibration process determines the actual spatial direction or region pointed to by each pixel. This correspondence is pre-calculated and stored as a "lookup table." During spatial mapping, based on this pre-stored lookup table, the data of each pixel in the cleanroom temperature matrix is ​​quickly mapped to its corresponding physical area (e.g., grouping multiple adjacent pixels into the same logical monitoring area). For example, an 8x8 array may be mapped to four sectors (S1, S2, S3, S4) in the monitoring space, with each sector consisting of 4x4 pixels.

[0113] In step S103, multiple temperature feature parameters are extracted and calculated, including time-domain feature parameters and spatial-domain feature parameters.

[0114] In step S104, the fire status is determined based on the fusion change trend of temporal and spatial characteristic parameters.

[0115] Among them, the fire situation assessment adopts a dual time scale comparison strategy that combines short-term and long-term windows, and makes decisions by comprehensively considering the sudden changes in the short-term window and the trends in the long-term window.

[0116] In step S105, if it is determined that the fire condition continues to be met, a control signal is generated and output to the external actuator.

[0117] Furthermore, in some optional embodiments, the method also includes dynamically updating the fire determination threshold and feature weights through an online learning mechanism to achieve adaptive optimization based on historical operating data and real-time environmental parameters.

[0118] The online learning mechanism uses machine learning algorithms to incrementally train the input feature set and outputs optimized judgment thresholds and multi-feature fusion weights.

[0119] It should be noted that the foregoing explanation of the embodiment of the multi-point temperature sampling array fire determination system also applies to the multi-point temperature sampling array fire determination method of this embodiment, and will not be repeated here.

[0120] The fire detection method using a multi-point temperature sampling array proposed in this application adopts a non-rotating structure, enabling multi-point parallel sampling without the need for mechanical moving parts. This design eliminates the risk of failure caused by rotational wear and positional offset, greatly improving the system's lifespan and stability. During long-term operation, performance degradation or failures due to mechanical component wear will not occur, reducing maintenance costs and downtime. This method is suitable for applications with extremely high reliability requirements. In data acquisition, only temperature data is collected, without involving images or visual information. This non-imaging scheme avoids privacy leaks that may arise from image acquisition, while effectively monitoring fires and ensuring the safety of personnel and property. Combining multi-dimensional parameters such as temperature rise rate, energy density, maximum temperature, and spatial continuity, a comprehensive threshold and time-based method are used. The continuous logic significantly reduces false alarms and missed alarms. Through comprehensive analysis of multiple parameters, it can more accurately judge the fire situation and avoid misjudgments that may be caused by judging a single parameter. The judgment threshold and weight are automatically adjusted according to environmental changes and historical data to achieve algorithm self-adaptation. This not only allows the system to maintain stable judgment performance during long-term operation, but also continuously optimizes the judgment parameters to improve the ability to judge the fire situation and reduce the occurrence of false alarms and missed alarms. The communication module, power supply and self-test module, and dynamic threshold and learning module have been expanded. These modular hardware designs give the system good scalability and can be adapted to different scenarios. The combination of multi-point parallel sampling and embedded computing makes the response time less than 1-2 seconds, which can realize early fire identification and automatic linkage control, improving the efficiency and accuracy of fire response.

[0121] This addresses issues related to response speed, structural complexity, cost, and privacy protection.

[0122] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0123] When the processor 302 executes the program, it implements the multi-point temperature sampling array fire determination method provided in the above embodiments.

[0124] Furthermore, the vehicle also includes: Communication interface 303 is used for communication between memory 301 and processor 302.

[0125] The memory 301 is used to store computer programs that can run on the processor 302.

[0126] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0127] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0128] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0129] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0130] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0131] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0132] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0133] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0134] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A multi-point temperature sampling array fire detection system, characterized in that, include: A temperature sampling array unit is fixedly installed within the monitoring area to synchronously collect temperature data from multiple discrete spatial sampling points within the monitoring area and output a temperature sampling matrix. An optical focusing component is disposed at the sensing front end of the temperature sampling array unit to expand the detection field of view or enhance the energy convergence of a specific sensitive area; A processing and control unit, electrically connected to the temperature sampling array unit, is used to receive the temperature sampling matrix and run a fire situation determination algorithm, and generate a determination signal based on the fire situation determination result. The processing and control unit includes: The data preprocessing module is used to perform noise suppression, background temperature compensation, and outlier correction on the temperature sampling matrix. The feature extraction module is used to extract time-domain and spatial-domain temperature feature parameters from the preprocessed data; The fire situation determination logic module is used to generate a fire situation determination signal based on the fusion analysis results of the time domain and spatial domain temperature characteristic parameters; An output interface unit, connected to the processing and control unit, is used to receive the determination signal and convert the determination signal into a corresponding control signal for output to an external execution device; The communication module is used to upload system status and fire situation assessment data to the monitoring platform or cloud server; The power supply and self-test module is used to power the system and to achieve automatic calibration and health status monitoring. The system is a static non-scanning architecture, without mechanical scanning or imaging components. The temperature sampling array unit is a non-imaging multi-point temperature sensor array. The fire determination is based on multi-parameter fusion analysis in the time and spatial domains.

2. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The feature extraction module further includes a temperature change trend analysis submodule and an energy density calculation submodule, wherein... The temperature change trend analysis submodule is used to analyze the change pattern of the temperature characteristic parameters over time; The energy density calculation submodule is used to calculate the energy density distribution within the monitoring area based on the spatial distribution of the temperature sampling matrix.

3. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The temperature characteristic parameters include any one or more of the following: temperature rise rate, energy density, average temperature, maximum temperature, and spatial continuity index. The spatial continuity index is obtained by performing image connectivity analysis on the binarized high-temperature region.

4. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The processing and control unit also includes a dynamic threshold and learning module, which is used to adaptively adjust the judgment threshold and fusion feature weights for fire determination based on changes in ambient temperature and historical fire data.

5. The multi-point temperature sampling array fire detection system according to claim 4, characterized in that, The dynamic threshold and learning module includes: Input feature set, used to cache historical temperature feature parameters; The feature normalization unit is used to normalize the temperature feature parameters; The online learning algorithm unit is used to output optimized judgment thresholds and feature weights based on normalized feature parameters and feedback signals through online machine learning algorithms. The parameter update unit is used to smoothly update the optimized parameters to the fire determination logic module. Local storage units are used to store system parameters and historical data in a non-volatile manner.

6. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The output interface unit is connected to an external actuator, which includes an audible and visual alarm, an automatic fire extinguishing device, a power cut-off device, and other safety linkage equipment.

7. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The communication module supports wired interfaces or wireless communication protocols, including at least one of RS485, CAN, NB-IoT, Wi-Fi, and LoRa communication protocols.

8. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The power supply and self-test module integrates non-uniformity correction, background temperature adaptive update, contamination detection, and temperature drift compensation functions.

9. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The judgment signal output by the processing and control unit is a graded fire judgment signal, including a warning signal, a fire confirmation signal, and a fire persistence signal.

10. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The fire situation determination adopts a dual-time window comparison mechanism, which determines transient heat sources and continuous fires by the trend difference between short-time windows and long-time windows.

11. The multi-point temperature sampling array fire detection system according to claim 1, characterized in that, The system is applicable to at least one of the following applications: building fire protection, industrial safety, energy storage equipment protection, rail transit environmental monitoring, and smart home security.

12. A method for determining the fire situation using a multi-point temperature sampling array, characterized in that, The method is executed by a processing and control unit and includes the following steps: Temperature data from multiple discrete spatial sampling points within the monitoring area are simultaneously collected by a temperature sampling array unit to generate a temperature sampling matrix. The temperature matrix is ​​preprocessed and spatially mapped to establish the correspondence between sampling points and physical monitoring areas; Extract and calculate multiple temperature feature parameters, including time-domain and spatial-domain feature parameters; Fire status is determined based on the fusion and change trends of the time-domain and spatial-domain feature parameters. If the fire conditions are determined to continue, a control signal is generated and output to the external actuator.

13. The fire situation determination method using a multi-point temperature sampling array according to claim 12, characterized in that, The fire situation determination adopts a dual-timescale comparison strategy that combines short-term and long-term windows, and makes decisions by comprehensively considering the sudden changes within the short-term window and the trends within the long-term window.

14. The fire situation determination method using a multi-point temperature sampling array according to claim 12, characterized in that, It also includes dynamically updating the fire assessment threshold and feature weights through an online learning mechanism, and achieving adaptive optimization based on historical operating data and real-time environmental parameters.

15. The fire situation determination method using a multi-point temperature sampling array according to claim 14, characterized in that, The online learning mechanism uses machine learning algorithms to incrementally train the input feature set and outputs optimized judgment thresholds and multi-feature fusion weights.

Images