Building fire alarm method and system

Abstract

The invention relates to the technical field of fire alarm, and particularly discloses a building fire alarm method and system, and the method comprises the following steps: S1, obtaining shot images and thermal images of all monitoring points in a building, and combining a pre-established target object library to determine a region needing to be monitored; s2, collecting a smoldering sample image set of each target object, classifying the smoldering sample image set, and training a smoldering thermal diffusion model of each type of target object by using a neural network model so as to simulate thermal diffusion characteristics of the smoldering thermal diffusion model; s3, dividing the monitoring area into pixel points, and detecting whether hot spots exist or not; if a hot spot is found, acquiring an initial thermal image of the hot spot, continuously monitoring, then calling a corresponding smoldering thermal diffusion model to generate a comparison image set, and judging whether the hot spot is a smoldering source or not by comparing the real-time thermal image with a model prediction result; according to the method, accurate detection of the smoldering source in the early stage of the building fire is realized by combining image recognition and a thermal diffusion model.

Description

Technical Field

[0001] The present invention relates to the technical field of fire alarms, and in particular to a building fire alarm method and system. Background Art

[0002] With the booming construction industry, modern buildings are becoming increasingly complex and diverse, with highly diversified interior layouts, finishing materials, and electrical equipment. In this environment, fire hazards are becoming more subtle and elusive. Smoldering fires, as a highly concealed form of fire, are more complex and difficult to detect than open flames.

[0003] While existing fire warning technologies can monitor and warn of fires to a certain extent, they exhibit numerous limitations when dealing with smoldering fires. Traditional smoke detectors primarily rely on changes in smoke concentration to trigger an alarm. However, smoldering fires often don't produce significant amounts of smoke in their early stages, making it difficult for smoke detectors to detect fire signals in a timely manner. Temperature sensors, on the other hand, are a common fire warning device. Their operating principle is to detect the occurrence of a fire based on significant changes in ambient temperature. However, in the initial stages of a smoldering fire, a relatively low localized temperature rise typically occurs. This temperature change can be slow and easily masked by normal temperature fluctuations in the surrounding environment, making it difficult for temperature sensors to accurately identify and issue a timely alarm. Summary of the Invention

[0004] The purpose of the present invention is to provide a building fire alarm method and system to solve the following technical problems.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A building fire alarm method comprises the following steps: Step S1: Acquire all monitoring points in the building, obtain captured images and thermal images at the monitoring points; establish a target object library, wherein the target object library contains a number of target objects; and determine a monitoring area based on the target object library, the captured images, and the thermal images; Step S2: Obtaining a smoldering sample image set of each target object, and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and obtaining a smoldering heat diffusion model for each target object in the class; Step S3: Divide the monitoring area into a number of pixels, determine whether a hot spot exists in each pixel, and if a hot spot exists, obtain an initial thermal image of the hot spot, and monitor the hot spot in real time to obtain a set of thermal images of the hot spot; Determine a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; input the initial thermal image into the comparative diffusion model to obtain a comparative image set; compare the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source.

[0006] As a further solution of the present invention: the process of obtaining the captured image and the thermal image includes: A thermal imaging monitoring point is set at each monitoring point, and the monitoring point is used to obtain the captured image within the monitoring range, and the thermal imaging monitoring point is used to obtain the thermal imaging map of the monitoring range.

[0007] As a further solution of the present invention: the process of establishing the target object library includes: Obtain sample images of all target objects, extract image features of all sample images of the target objects based on image recognition technology, record them as sample image features, associate each sample image feature with its corresponding target object, and obtain a target object library.

[0008] As a further solution of the present invention: the process of determining the monitoring area includes: Identify all objects in the captured image, and record the area occupied by the object in the captured image as the object area; extract image features of the object area based on image feature extraction technology, and record them as object features; compare the object features with all sample image features in the target object library, obtain the similarity between the object features and each sample image feature, screen out several sample image features whose similarity with the object features exceeds a preset similarity threshold, and select the sample image feature with the highest similarity to the object feature from the several sample image features, and record them as target image features; obtain the target object corresponding to the target image feature in the target object library, and record the area occupied by the object in the captured image as the monitoring area.

[0009] As a further solution of the present invention, the process of acquiring the smoldering sample image set of the target object includes: A monitoring time period of a preset length is set, and the moment when the target object just begins to smolder is recorded as the smoldering start moment. From the smoldering start moment of the target object, a video stream of the target object in the monitoring time period after the smoldering start moment is obtained; the video stream is divided into a number of image frames at equal intervals, and a smoldering sample image set is obtained from all the image frames.

[0010] As a further solution of the present invention, the process of dividing the smoldering sample image set of each target object into several categories includes: Obtain the material of each target object and the physical properties of each material, including ignition point, specific heat capacity and porosity; preset the weight coefficient of each physical property and obtain the property value of each material , where w i Represents the weight coefficient of the i-th physical attribute, P i represents the value of the i-th physical property of the material, where n is the total number of physical properties; A plurality of attribute value intervals are set, and a plurality of materials whose attribute values ​​belong to the same attribute value interval are recorded as one category; and a plurality of target objects whose materials belong to the same category are recorded as the same category.

[0011] As a further solution of the present invention: the process of obtaining the smoldering heat diffusion model includes: Acquire a smoldering area of ​​the target object within the image frame, and obtain temperature characteristics and geometric characteristics of the smoldering area, wherein the temperature characteristics include a maximum temperature, a temperature gradient, and a temperature change rate of the smoldering area, and the geometric characteristics include a boundary change, an area, and a diffusion rate of the smoldering area; According to the monitoring time period, a time sequence is established, and the time node corresponding to each image frame in the time sequence is obtained; the temperature characteristics and geometric characteristics at each time node are recorded as a training set, and the training set is input into the initial model. The initial model is trained to obtain a smoldering heat diffusion model of the target object.

[0012] As a further solution of the present invention: a building fire alarm system comprising: Data acquisition module: including monitoring points and thermal imaging monitoring points. A thermal imaging monitoring point is set at each monitoring point. The monitoring point is used to obtain the captured images within the monitoring range, and the thermal imaging monitoring point is used to obtain the thermal imaging map of the monitoring range; Establishing a target object library, the target object library containing a plurality of target objects; determining a monitoring area based on the target object library, the captured image, and the thermal image; Fitting module: Obtaining a smoldering sample image set of each target object and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and obtaining a smoldering heat diffusion model of each target object in the class; Monitoring module: Divides the monitoring area into a number of pixels, determines whether a hot spot exists in each pixel, obtains an initial thermal image of the hot spot if a hot spot exists, and monitors the hot spot in real time to obtain a set of thermal images of the hot spot; Determine a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; input the initial thermal image into the comparative diffusion model to obtain a comparative image set; compare the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source.

[0013] Beneficial effects of the present invention: By combining thermal imaging technology with a neural network model, the present invention can effectively identify early smoldering sources, solving the problem that traditional smoke detectors and temperature sensors have difficulty detecting smoldering fires in their early stages. A smoldering heat diffusion model is used to conduct real-time comparative analysis of hot spots, and the slope of the regression line is used to determine whether they are smoldering sources, significantly improving the accuracy and timeliness of fire warnings. Classification and training models are based on the physical properties of the target object's material (such as ignition point, specific heat capacity, etc.), making the smoldering heat diffusion model more targeted and improving its adaptability in different scenarios. Image recognition is used to automatically determine the monitoring area, and the combination of pixel-level hot spot detection and region growing algorithms reduces manual intervention, enabling automated monitoring and early warning of fire hazards. Model training is performed using comprehensive temperature and geometric features, enhancing the comprehensiveness and reliability of fire predictions. Through technological innovation, the present invention significantly improves the early warning capabilities of building fires, especially smoldering fires, and has important practical application value. Furthermore, the present invention is suitable for complex building environments, can accurately locate fire hazards and issue real-time warnings, providing efficient technical support for fire prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below with reference to the accompanying drawings.

[0015] Figure 1 It is a schematic diagram of the steps of a building fire alarm method of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] See also Figure 1 As shown, the present invention is a building fire alarm method, comprising the following steps: Step S1: Acquire all monitoring points in the building, set up thermal imaging monitoring points at each monitoring point, the monitoring points are used to acquire images within the monitoring range, and the thermal imaging monitoring points are used to acquire thermal images of the monitoring range; Establishing a target object library, the target object library containing a plurality of target objects; identifying all target objects in the captured image based on image recognition technology and the target object library, and marking the area occupied by the target objects in the thermal image as a monitoring area; According to the building plan, monitoring points are deployed in key areas such as corridors, rooms, and equipment rooms to ensure that each monitoring point has complete field of view. Each monitoring point integrates a visible light camera (resolution ≥1080p) and an infrared thermal imager (temperature measurement range -20°C to 550°C, accuracy ±2°C). A synchronization controller ensures that the timestamps of the two types of images are aligned. The visible light camera captures RGB images at a frame rate of 5fps, and the thermal imager simultaneously outputs temperature matrix data. All data is pre-processed (denoising and distortion correction) by edge computing nodes and uploaded to the central server. As a preferred embodiment of the present invention, the target objects are all objects in the building that have a smoldering risk; Collect multi-angle sample images of common flammable objects in buildings. The target objects include wooden furniture, curtains, electrical equipment, etc. Each type of object contains at least 200 annotated images, and the sample images are annotated with bounding boxes and material labels. As a preferred embodiment of the present invention, the process of establishing the target object library includes: Obtain sample images of all target objects, extract image features of all sample images of the target objects based on image recognition technology, record them as sample image features, associate each sample image feature with its corresponding target object, and obtain a target object library; Use ResNet-50 or YOLOv7 models to extract image features (such as SIFT, HOG, or deep features) and build a feature vector database. Use the cosine similarity algorithm to compare real-time image features with features in the database, set a similarity threshold (such as 0.85), and filter out potential target objects. Perform non-maximum suppression (NMS) on the matching results to eliminate duplicate detections. In a preferred embodiment of the present invention, the process of identifying the target object in the captured image includes: Identify all objects in the captured image, and record the area occupied by the object in the captured image as the object area; extract image features of the object area based on image feature extraction technology, and record them as object features; compare the object features with all sample image features in the target object library, obtain the similarity between the object features and each sample image feature, screen out several sample image features whose similarity with the object features exceeds a preset similarity threshold, and select the sample image feature with the highest similarity to the object feature from the several sample image features, and record it as the target image feature; obtain the target object corresponding to the target image feature in the target object library, and record the area occupied by the object in the captured image as the monitoring area; The coordinates of the target object identified in the visible light image are mapped to the thermal image to generate an ROI (region of interest) mask. Adaptive threshold segmentation is performed on the temperature data within the ROI to eliminate interference from environmental heat sources. Monitoring weights are assigned based on the fire risk level of the target object (such as the material ignition point and historical fire data). High-risk areas use a higher sampling frequency (10fps), while low-risk areas (reduced to 1fps) save computing power. The high-risk area includes the distribution box, and the low-risk area includes the metal bracket. Apply HSV color space analysis to visible light images to identify bright or shadowed areas and dynamically adjust the temperature difference sensitivity of thermal imaging. If the temperature fluctuation in a certain area is less than ±1°C for 30 seconds, it is marked as a steady-state environment and real-time monitoring is suspended. Step S2: Obtaining a smoldering sample image set of each target object, and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and training the initial model to obtain a smoldering heat diffusion model for each target object in the class; In a preferred embodiment of the present invention, the process of acquiring the smoldering sample image set of the target object includes: A monitoring period of a preset length is set, the moment when the target object begins to smolder is recorded as the smoldering start time, and a video stream of the target object within the monitoring period after the smoldering start time is obtained from the smoldering start time; the video stream is divided into a number of image frames at equal intervals, and a smoldering sample image set is obtained from all the image frames; It should be noted that a 1:1 building material testing platform was built in a controlled laboratory environment, equipped with a constant temperature and humidity system. The temperature range in the controlled laboratory environment was 23±2°C and the humidity range was 50±5%RH. An industrial-grade thermal imager was used to collect data synchronously with a 4K high-speed camera. Three groups of control experiments were set up for each target material, namely normal state, initial smoldering stage, and open flame conversion stage. In a preferred embodiment of the present invention, the process of dividing the smoldering sample image set of each target object into several categories includes: Obtain the material of each target object and the physical properties of each material, including ignition point, specific heat capacity and porosity; preset the weight coefficient of each physical property and obtain the property value of each material , where w i Represents the weight coefficient of the i-th physical attribute, P i represents the value of the i-th physical property of the material, where n is the total number of physical properties; Setting a number of attribute value intervals, recording a number of materials whose attribute values ​​belong to the same attribute value interval as one category; and recording a number of target objects whose materials belong to the same category as the same category; The ignition point determines the smoldering starting temperature threshold, the specific heat capacity reflects the heat accumulation rate, and the porosity affects the oxygen diffusion efficiency; The fire risk attribute value of the material is calculated using a linear weighted formula, where the weight coefficient is determined by the hierarchical analysis method or principal component analysis; the Av value is divided into 3 to 5 intervals using K-means clustering or equal frequency binning method; In a preferred embodiment of the present invention, the process of obtaining the smoldering heat diffusion model includes: Acquire a smoldering area of ​​the target object within the image frame, and obtain temperature characteristics and geometric characteristics of the smoldering area, wherein the temperature characteristics include a maximum temperature, a temperature gradient, and a temperature change rate of the smoldering area, and the geometric characteristics include a boundary change, an area, and a diffusion rate of the smoldering area; Establishing a time sequence according to the monitoring time period and obtaining a time node corresponding to each image frame in the time sequence; recording the temperature features and geometric features at each time node as a training set, inputting the training set into the initial model, training the initial model, and obtaining a smoldering heat diffusion model of the target object; It should be noted that for different materials, a dedicated smoldering heat diffusion model is generated for quantitative prediction. A complete characteristic representation of the smoldering process is constructed by simultaneously analyzing the thermal imaging temperature field, visible light morphological changes, and material physical parameters. A 3D convolutional neural network is used to capture the spatiotemporal evolution of temperature diffusion, including the direction and velocity of heat flow, and the combustion trend is predicted in combination with material properties. The heat transfer equation is embedded in the neural network loss function to constrain the model to conform to physical laws, and the diffusion consistency loss is used to ensure that the predicted temperature change rate is consistent with the actual thermodynamic process. Step S3: Divide the monitoring area into a number of pixels, determine whether a hot spot exists in each pixel, and if a hot spot exists, obtain an initial thermal image of the hot spot, and monitor the hot spot in real time to obtain a set of thermal images of the hot spot; Determining a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; inputting the initial thermal image into the comparative diffusion model to obtain a comparative image set; comparing the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source; The monitoring area is divided into a high-resolution grid. The temperature of each pixel is collected in real time using a thermal imager. An adaptive threshold algorithm is used to identify abnormally high temperature areas, i.e., hot spots. Candidate hot spots are tracked in time series. A valid hot spot is only considered valid if it persists and expands in area for five consecutive frames (approximately one second). In a preferred embodiment of the present invention, the process of determining whether there is a hot spot in each pixel includes: Obtaining the average temperature of each pixel in the smoldering area of ​​the target object at the start of smoldering; obtaining the minimum value of the average temperature of each target object, and setting a temperature threshold based on the minimum value; if the temperature of any pixel in the monitoring area exceeds the temperature threshold, then marking the pixel as a hot spot; It is worth noting that if a hot spot may be composed of multiple adjacent pixels, a region growing algorithm is used to merge connected regions above a threshold into one hot spot; The process of setting the temperature threshold according to the minimum value includes: Under normal building operation, the temperature data of each monitoring point is continuously collected during the monitoring period, and the smoldering starting temperature characteristics of various materials are obtained through laboratory simulation. The real-time threshold value is calculated using the sliding window percentile method. The temperature threshold value T th =T min +k× σ , where T min is the minimum value, σ The standard deviation of the temperature at the monitoring point over the past 24 hours is used to eliminate the impact of environmental fluctuations. k is the sensitivity coefficient, which defaults to 2.5 and can be adjusted to 1.8 in high-risk areas. As a preferred embodiment of the present invention, the initial thermal image of the hot spot is a thermal image when the hot spot is first discovered; In a preferred embodiment of the present invention, the process of obtaining the thermal image set of the hot spot includes: The thermal imaging monitoring point acquires a thermal image of the monitoring area in real time to obtain a plurality of real-time thermal images, and all the real-time thermal images are recorded as a thermal image set; As a preferred embodiment of the present invention, the process of obtaining the comparison image set includes: The contrast diffusion model uses the initial thermal image as the thermal image corresponding to the monitoring area at the start of smoldering; the timestamp corresponding to the initial thermal image is used as the initial time, and the contrast diffusion model generates images of the monitoring area at each time point within the monitoring period after the initial time, which are recorded as simulated images, and a contrast image set is obtained from all the simulated images; In a preferred embodiment of the present invention, the process of determining whether the hot spot is a smoldering source includes: Numbering each real-time thermal image in the thermal image set and numbering each simulated image in the comparison image set; obtaining a similarity between each thermal image and simulated image with the same number, and recording it as an image similarity; A comparison similarity threshold is set, and the similarity difference corresponding to each number is obtained in real time. The similarity difference C = Sm - Sm', where Sm is the image similarity and Sm' is the comparison similarity threshold; the similarity difference corresponding to each number is obtained, and each number and its corresponding similarity difference are fitted based on the least squares method to obtain a regression line; the slope of the regression line is obtained. If the slope is greater than 0, the hot spot is a smoldering source; otherwise, the hot spot is not a smoldering source; Use the structural similarity index (SSIM) or Hausdorff distance to compare the measured thermal images with the model predictions frame by frame. If the similarity is greater than 0.8 and the error gradient is consistent, it is determined to be a smoldering source. If the similarity is less than 0.8 and the diffusion direction is inconsistent, it is determined to be an interference source. The numbering is used to eliminate the time offset between model prediction and measured data through dynamic time warping; It is understandable that when the hot spot is a smoldering source, the location coordinates of the monitoring area in the building are obtained, and an early warning is sent to the relevant staff or platform to remind the relevant staff or platform that there is a smoldering fire hazard at the location coordinates.

[0018] A building fire alarm system, comprising: Data acquisition module: including monitoring points and thermal imaging monitoring points. A thermal imaging monitoring point is set at each monitoring point. The monitoring point is used to obtain the captured images within the monitoring range, and the thermal imaging monitoring point is used to obtain the thermal imaging map of the monitoring range; Establishing a target object library, the target object library containing a plurality of target objects; determining a monitoring area based on the target object library, the captured image, and the thermal image; Fitting module: Obtaining a smoldering sample image set of each target object and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and obtaining a smoldering heat diffusion model of each target object in the class; Monitoring module: Divides the monitoring area into a number of pixels, determines whether a hot spot exists in each pixel, obtains an initial thermal image of the hot spot if a hot spot exists, and monitors the hot spot in real time to obtain a set of thermal images of the hot spot; Determine a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; input the initial thermal image into the comparative diffusion model to obtain a comparative image set; compare the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source.

[0019] The above is a detailed description of one embodiment of the present invention. However, the content described is a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A building fire alarm method, characterized in that: The following steps are involved: Step S1: Acquire all monitoring points in the building, and obtain the captured images and thermal images at the monitoring points; Establishing a target object library, the target object library containing a plurality of target objects; determining a monitoring area based on the target object library, the captured image, and the thermal image; Step S2: Obtaining a smoldering sample image set of each target object, and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and obtaining a smoldering heat diffusion model for each target object in the class; Step S3: Divide the monitoring area into a number of pixels, determine whether a hot spot exists in each pixel, and if a hot spot exists, obtain an initial thermal image of the hot spot, and monitor the hot spot in real time to obtain a set of thermal images of the hot spot; Determine a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; input the initial thermal image into the comparative diffusion model to obtain a comparative image set; compare the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source.

2. A building fire alarm method according to claim 1, characterized in that: In step S1, the process of obtaining the captured image and the thermal image includes: A thermal imaging monitoring point is set at each monitoring point, and the monitoring point is used to obtain the captured image within the monitoring range, and the thermal imaging monitoring point is used to obtain the thermal imaging map of the monitoring range.

3. A building fire alarm method according to claim 1, characterized in that: In step S1, the process of establishing the target object library includes: Obtain sample images of all target objects, extract image features of all sample images of the target objects based on image recognition technology, record them as sample image features, associate each sample image feature with its corresponding target object, and obtain a target object library.

4. A building fire alarm method according to claim 1, characterized in that: In step S1, the process of determining the monitoring area includes: Identify all objects in the captured image, and record the area occupied by the object in the captured image as the object area; extract image features of the object area based on image feature extraction technology, and record them as object features; compare the object features with all sample image features in the target object library, obtain the similarity between the object features and each sample image feature, screen out several sample image features whose similarity with the object features exceeds a preset similarity threshold, and select the sample image feature with the highest similarity to the object feature from the several sample image features, and record them as target image features; obtain the target object corresponding to the target image feature in the target object library, and record the area occupied by the object in the captured image as the monitoring area.

5. A building fire alarm method according to claim 1, characterized in that: In step S2, the process of acquiring the smoldering sample image set of the target object includes: A monitoring time period of a preset length is set, and the moment when the target object just begins to smolder is recorded as the smoldering start moment. From the smoldering start moment of the target object, a video stream of the target object in the monitoring time period after the smoldering start moment is obtained; the video stream is divided into a number of image frames at equal intervals, and a smoldering sample image set is obtained from all the image frames.

6. A building fire alarm method according to claim 1, characterized in that: In step S2, the process of dividing the smoldering sample image set of each target object into several categories includes: Obtain the material of each target object and the physical properties of each material, including ignition point, specific heat capacity and porosity; preset the weight coefficient of each physical property and obtain the property value of each material , where w i Represents the weight coefficient of the i-th physical attribute, P i represents the value of the i-th physical property of the material, where n is the total number of physical properties; A plurality of attribute value intervals are set, and a plurality of materials whose attribute values ​​belong to the same attribute value interval are recorded as one category; and a plurality of target objects whose materials belong to the same category are recorded as the same category.

7. A building fire alarm method according to claim 1, characterized in that: In step S2, the process of obtaining the smoldering heat diffusion model includes: Acquire a smoldering area of ​​the target object within the image frame, and obtain temperature characteristics and geometric characteristics of the smoldering area, wherein the temperature characteristics include a maximum temperature, a temperature gradient, and a temperature change rate of the smoldering area, and the geometric characteristics include a boundary change, an area, and a diffusion rate of the smoldering area; According to the monitoring time period, a time sequence is established, and the time node corresponding to each image frame in the time sequence is obtained; the temperature characteristics and geometric characteristics at each time node are recorded as a training set, and the training set is input into the initial model. The initial model is trained to obtain a smoldering heat diffusion model of the target object.

8. A building fire alarm method according to claim 1, characterized in that: In step S3, the process of determining whether the hot spot is a smoldering source includes: The thermal imaging monitoring point acquires a thermal image of the monitoring area in real time to obtain a plurality of real-time thermal images, and all the real-time thermal images are recorded as a thermal image set; The contrast diffusion model uses the initial thermal image as the thermal image corresponding to the monitoring area at the start of smoldering; the timestamp corresponding to the initial thermal image is used as the initial time, and the contrast diffusion model generates images of the monitoring area at each time point within the monitoring period after the initial time, which are recorded as simulated images, and a contrast image set is obtained from all the simulated images; Numbering each real-time thermal image in the thermal image set and numbering each simulated image in the comparison image set; obtaining a similarity between each thermal image and simulated image with the same number, and recording it as an image similarity; A comparison similarity threshold is set, and the similarity difference corresponding to each number is obtained in real time. The similarity difference C = Sm-Sm', where Sm is the image similarity and Sm' is the comparison similarity threshold. The similarity difference corresponding to each number is obtained, and each number and its corresponding similarity difference are fitted based on the least squares method to obtain a regression line. The slope of the regression line is obtained. If the slope is greater than 0, the hot spot is a smoldering source; otherwise, the hot spot is not a smoldering source.

9. A building fire alarm system, characterized in that: include: Data acquisition module: including monitoring points and thermal imaging monitoring points. A thermal imaging monitoring point is set at each monitoring point. The monitoring point is used to obtain the captured images within the monitoring range, and the thermal imaging monitoring point is used to obtain the thermal imaging map of the monitoring range; Establishing a target object library, the target object library containing a plurality of target objects; determining a monitoring area based on the target object library, the captured image, and the thermal image; Fitting module: Obtaining a smoldering sample image set of each target object and dividing the smoldering sample image set of each target object into several classes; establishing an initial model based on a neural network model, inputting the smoldering sample image set of all target objects in the same class into the initial model, and obtaining a smoldering heat diffusion model of each target object in the class; Monitoring module: Divides the monitoring area into a number of pixels, determines whether a hot spot exists in each pixel, obtains an initial thermal image of the hot spot if a hot spot exists, and monitors the hot spot in real time to obtain a set of thermal images of the hot spot; Determine a smoldering heat diffusion model corresponding to the monitoring area, recorded as a comparative diffusion model; input the initial thermal image into the comparative diffusion model to obtain a comparative image set; compare the thermal image set with the comparative image set to determine whether the hot spot is a smoldering source.

Images