Ancient village Internet of Things fire-fighting remote monitoring system supporting multi-mode early warning

Abstract

The invention discloses an ancient village internet-of-things fire-fighting remote monitoring system supporting multi-mode early warning, and belongs to the technical field of intelligent fire fighting. Comprising a data detection module, an edge calculation module, a communication network module, an early warning response module, a virtual reality simulation drilling module and a three-dimensional visual monitoring module. The system is provided with an early warning judgment module, scene features are constructed through multi-modal data under multiple scenes, scene feature fusion is achieved, the smoke temperature level, the temperature change rate in a building and the diffusion rate are automatically judged, the fire-fighting early warning truth rate is improved through datamation analysis of an open fire recognition threshold value, the possibility of false alarm is reduced, and the fire-fighting early warning efficiency is improved. A consecutive fire-fighting early warning response operation generated subsequently due to false alarm of the system is avoided; the problem that at present, fire early warning accuracy in ancient villages is not enough, and consequently early warning response misoperation causes damage to grass and tree buildings of the ancient villages is solved.

Description

Technical Field

[0001] This invention belongs to the field of smart fire protection technology, specifically relating to an IoT-based remote fire monitoring system for ancient villages that supports multimodal early warning. Background Technology

[0002] Ancient buildings and original villages, as important carriers of history and culture, are mostly constructed of wood or thatched roofs. Due to their age, they lack fire resistance. Furthermore, ancient villages and buildings are usually densely packed and lack standard fire lanes. Once a fire breaks out, it can spread rapidly, causing irreparable damage to cultural heritage and threatening the lives of villagers. Therefore, for ancient villages and buildings, prevention is better than cure. Accurate fire monitoring of ancient villages and buildings to nip potential fire hazards in the bud is far more important than timely response when a fire occurs. Traditional fire hazard monitoring and investigation relies on human intervention, with personnel assigned to specific areas within ancient villages to conduct monitoring and investigation at fixed times and locations. However, human monitoring is prone to omissions, untimely monitoring, and slow response after fire hazards are discovered. With the development of Internet of Things (IoT) technology, fire monitoring can be made intelligent and unmanned. Through remote monitoring sensors and wireless network data transmission, unmanned fire monitoring can be directly achieved. To allow tourists to immerse themselves in the ethnic customs and original lifestyle of ancient villages, the original living scenes of the residents are still preserved in the villages. However, the original residential buildings usually use firewood in stoves as the source of fire for cooking. The smoke produced during the burning of firewood is detected by the fire detection module of the system, which can easily be misjudged as a fire, leading to a series of subsequent fire emergency actions controlled by the system. Although existing fire monitoring systems use indicators such as smoke temperature sequence and indoor temperature change rate to determine the authenticity of fire warnings, these judgments are based on intuitive scene phenomena and do not have accurate data-driven analysis of intuitive scenes. In essence, the judgment of the authenticity of a fire is still based on intuitive scene phenomena combined with experience. This judgment method has a high misjudgment rate. Summary of the Invention

[0003] This invention provides an IoT-based remote fire monitoring system for ancient villages that supports multimodal early warning. The system includes an early warning discrimination module that constructs scene features from multimodal data across multiple scenarios, achieving scene feature fusion and automatically judging smoke and temperature levels, building temperature change rates and diffusion rates, and open flame identification thresholds. Through data analysis, the accuracy of fire warnings is improved, reducing the possibility of false alarms and preventing subsequent consecutive fire warning response operations caused by false alarms. This solves the problem of damage to vegetation and buildings in ancient villages caused by inaccurate early warning systems.

[0004] To achieve the above-mentioned technical objectives, the present invention is implemented through the following technical solution: A remote fire monitoring system for ancient villages using the Internet of Things (IoT) that supports multimodal early warning includes: The data detection module is used to simultaneously detect light, smoke, temperature, and sound multimodal monitoring data on both the exterior and interior of buildings in ancient villages. The edge computing module is used to preprocess the collected multimodal data. Based on the preprocessed multimodal data, a scene feature model is constructed to determine the authenticity of the early warning. When it is determined to be a real fire early warning, the algorithm model is used to analyze the early warning level and location based on the preprocessed multimodal data. The communication network module, employing hybrid networking technology, transmits multimodal data collected by the data detection module to the edge computing module via a communication protocol; simultaneously, it precisely matches the fire early warning analysis results from the edge computing module to the fire response equipment in the early warning response module via the communication protocol. The early warning response module has multiple levels of fire response equipment. It receives fire early warning analysis results from the edge computing module, generates matching response commands, and controls the corresponding fire response equipment to complete the command actions through the response commands. The virtual reality simulation exercise module integrates multimodal data collected by the data detection module, including 3D scanning information of ancient village buildings. It constructs a digital twin based on the BIM-GIS 3D model and uses fluid dynamics to simulate the fire and smoke diffusion path by combining on-site meteorological monitoring data. This allows for early prediction of the fire's development trend, early generation of response instructions for the early warning response module, and early generation of resource scheduling plans, thereby improving response accuracy and shortening rescue time. The 3D visualization monitoring module is used to model the fire response equipment and buildings in the ancient village under the early warning response module by point, and connect with the edge computing module. When a fire occurs, the 3D visualization platform immediately roams to the early warning location, forms an associated early warning sequence diagram according to the early warning sequence, and the fire response equipment set around the early warning sequence diagram is put into a standby state.

[0005] Preferably, the light, smoke, temperature, and sound multimodal data are captured and collected by setting up a multispectral high-altitude observation camera, a smoke detector, a temperature detector, and a fire sound and light alarm, respectively. Using the outer perimeter of the ancient village as a boundary, a multispectral high-altitude observation camera is installed at each apex of the geometric shape formed by the boundary. The multispectral high-altitude observation camera is used to capture and collect visual images of smoke, firelight, and flying fire in the ancient village. Within each building of the ancient village, the rooms are divided into two sections based on whether they are interconnected. Interconnected rooms are considered as the same space, and smoke detectors and temperature detectors are installed in each space. Smoke detectors and temperature detectors are also installed at equal intervals around the perimeter of each building. The fire alarm with sound and light is installed on the exterior of the building.

[0006] Preferably, the multimodal data preprocessing is divided into preprocessing of smoke sensor, temperature sensor time series data and video stream data from multispectral high-altitude observation cameras; The time-series data preprocessing first uses exponentially weighted average (EWMA) for noise filtering, then uses linear interpolation to fill in missing values, and uses the IQR rule to identify and remove outliers; finally, it uses Min-Max to normalize the data. The video stream data is first converted from color frames to grayscale frames by grayscale conversion, retaining only the luminance channel to reduce data volume while preserving the contrast features of smoke, flames, and background. Histogram averaging is used to improve image clarity in low-light environments, preventing smoke from being misjudged due to insufficient light. Frames are extracted at fixed intervals to reduce data volume and ensure that no key dynamics are missed. The MOG2 algorithm is used to separate smoke, fire, walls, and equipment. Smoke and fire areas are extracted and Gaussian filtering is applied to these areas for noise reduction. The optical flow field of adjacent frames is calculated to capture the movement trend of smoke and the flickering characteristics of fire. Smoke detector, temperature sensor, and video stream data are matched with frames within the same time period based on the frame timestamps of the video stream to ensure spatiotemporal consistency of the analysis.

[0007] Preferably, the method for determining the authenticity of the early warning is as follows: 1) Distinguishing between smoke and flame characteristics: Smoke differentiation analysis based on spectral and morphological characteristics: Smoke from daily life and smoke from fires can be distinguished by the difference in the proportion of blue light in the scattered light and the difference in the spectral distribution of infrared radiation intensity. The OpenCV contour detection algorithm is used to distinguish between residential smoke and fire smoke based on the diffusion speed and contour shape of the smoke. The flickering feature frequency is extracted by calculating the grayscale changes of the pixels at the edge of the flame using the optical flow method of visual monitoring. The flickering feature frequency is then used to determine whether the flame is from domestic use or a fire. Temperature distribution is captured by thermal infrared sensing, and the temperature field of domestic fire and fire temperature field are distinguished by calculating the standard deviation of the temperature gradient. 2) Scene-adaptive dynamic weight adjustment mechanism: Based on the distinguishing features of smoke, temperature, and flame in 1), we will construct different scenario libraries and continuously update them in future monitoring by combining the smoke, temperature, and flame feature data with the corresponding new life scenarios. 3) Machine learning algorithm model: The input layer integrates smoke, temperature, flames, and corresponding scene features; The feature extraction layer processes temporal data using a 3D convolutional neural network (3D-CNN), capturing the changes in smoke morphology over 10 consecutive frames. The kernel size of the 3D-CNN is set to 3*3*3. A long short-term memory network (LSTM) is used to capture temperature change trends, with 64 hidden layer units in the LSTM. The decision-making layer outputs three categories of results: "normal", "suspected", and "fire" based on the judgment of the feature extraction layer. The modal features and corresponding classification thresholds are automatically adjusted based on scene updates and complexity; for example, in sacrificial scenes or bonfire parties held in ancient villages, the "fire" recognition threshold is increased to avoid misjudgment.

[0008] Preferably, the analysis model for the warning level and warning location is as follows: The smoke detectors and heat detectors deployed inside and outside each building in the ancient village are numbered, and the accurate location coordinates of each group of numbered smoke detectors and heat detectors are determined. The number of each group of smoke detectors and heat detectors is matched with its location coordinates. After receiving the sensing signals from the smoke detectors and heat detectors, the processor input terminal in the edge computing module determines the number of the smoke detectors and heat detectors based on the input port. The location coordinates are then matched with the corresponding location coordinates, which are the warning locations. Smoke and flame video stream images are captured by a multispectral high-altitude observation camera. After preprocessing the visual images of the video stream, frame feature data of smoke and flame are extracted, including the contour shape of smoke and flame and the smoke diffusion speed. Combined with temperature detection by a temperature sensor, the dynamic trend of temperature rise is analyzed. Spatial features of the spatial contour shape of flame, smoke, and diffusion range are extracted by a convolutional neural network (CNN). Temporal features of smoke diffusion speed and temperature rise dynamic trend are processed by a long short-term memory network (LSTM). Combining spatial and temporal features, the severity level of the fire is output through a fully connected layer.

[0009] Preferably, the communication network module includes an edge-cloud collaboration unit. After the edge computing module completes 90% of the real-time data preprocessing and analysis, the edge-cloud collaboration unit uploads key result data, such as warning level and warning location, to the cloud platform for backup through an encrypted channel. This prevents the key result data from being lost or distorted during transmission from the communication network module to the warning response module, which would otherwise fail to receive complete warning level and warning location information and thus be unable to make an accurate emergency response.

[0010] Preferably, the data detection module further includes: an electricity fingerprint recognition and collection unit, in which an electricity fingerprint collection and recognition module is installed at the distribution box of each household on the electricity circuit of the ancient village; the power can only be turned on after successful fingerprint collection and recognition; the fingerprint information is transmitted to the control center for record through the communication network module; for the use of high-power electrical products such as electric vehicles and electric stoves, the building corresponding to the fingerprint information is locked and displayed in three dimensions on the three-dimensional visualization monitoring module for key monitoring.

[0011] Preferably, the early warning response module includes an emergency linkage control unit, which is used to link fire response equipment via a communication protocol; The fire response equipment includes: Each household is equipped with an independent indoor fire hydrant, achieving one hydrant per household; the ancient village is divided into several concentric rings by a circle or geometric diagram, and several sets of outdoor fire hydrants are reasonably arranged in each concentric ring according to the size of the perimeter of each concentric ring. The automatic sprinkler head installed indoors adopts a downspray nozzle structure, with a heat-gathering cover installed below the nozzle. When a fire occurs indoors, the automatic sprinkler head sprays water to extinguish the fire, and at the same time, the heat-gathering cover quickly gathers the radiant heat generated. Several fire monitors are set up outdoors in the ancient village. The fire monitors are evenly distributed within the ancient village, and the spray area of ​​the several fire monitors can completely cover the area of ​​the ancient village. Similarly, the aforementioned indoor fire hydrants, outdoor fire hydrants, automatic sprinkler heads, and fire monitors are all numbered, and each number corresponds to an accurate location coordinate; based on the warning location information, the fire-fighting facilities at the corresponding locations are activated for fire emergency response; A fire-fighting water tank is located at a high point around the ancient village. A level sensor is installed inside the fire-fighting water tank. An inlet pressure sensor, an inlet flow meter, and a flow regulating valve are installed at the inlet of the fire-fighting water tank. An outlet flow meter, an emergency shut-off valve, and an outlet flow valve are installed at the outlet of the fire-fighting water tank. The level sensor, inlet pressure sensor, inlet flow meter, and outlet flow meter are connected to the input terminal of the control center processor via a communication network module. The flow regulating valve and the emergency shut-off valve are connected to the output terminal of the control center processor via the communication network module.

[0012] Preferably, the early warning response module further includes a flying fire defense unit. The flying fire defense unit uses outdoor fire hydrants in the boundary area of ​​the early warning response module as the first line of defense and fire monitors as the second line of defense. It uses several multispectral high-altitude observation cameras set in the data detection module as monitoring means to perform multispectral detection and flying fire monitoring in the area. After the location area of ​​the flying fire is determined by the edge computing module, the early warning response module controls the fire hydrants and fire monitors in the flying fire area to enter the emergency response state to extinguish the flying fire.

[0013] Preferably, the meteorological monitoring data includes: wind direction, wind speed, air temperature and humidity, and air pressure; outliers in the collected meteorological data need to be removed, and the meteorological data is input into the fire dynamics simulator (FDS) to perform fluid dynamics simulation on the direction, degree, and range of fire spread; rescue plans are deployed in advance based on the simulation results, fire response equipment is prepared in advance through the early warning response module, and rescue and evacuation routes are planned.

[0014] The beneficial effects of this invention are: This invention provides an IoT-based remote fire monitoring system for ancient villages that supports multimodal early warning. By constructing scene features and matching captured modal features such as smoke and flames, it determines whether a real fire has occurred. If a real fire has occurred, it then analyzes the fire warning level and location. If it is determined to be residential smoke and flames, no further measures are taken. This invention integrates multimodal data and performs dynamic analysis of smoke and flame features, and then adapts it to residential and fire scenarios. Compared to relying on a single feature and intuitive scene combined with human experience for judgment, this invention can significantly improve the accuracy of fire authenticity judgment and avoid misjudgment leading to subsequent chain control responses. Given the flammability of buildings in ancient villages, a system for monitoring hazardous electricity use has been implemented. Users are required to have their fingerprints collected and verified before using electricity; the fingerprint information is transmitted to the control center for record-keeping. If high-power electricity is used, the building corresponding to that fingerprint will be locked and closely monitored to ensure an immediate early warning response in case of fire, allowing for timely firefighting and preventing further escalation. Furthermore, once a fingerprint is registered in the control center and located in its corresponding building, the coordinated early warning and response module will issue a preparatory command to the fire response equipment below, enabling emergency response and firefighting at any time. The water level, inlet pressure, inlet flow rate, and outlet flow rate of the fire water tank are monitored. When no fire occurs, the water level monitoring ensures that the water level in the fire water tank is always above the safe threshold to avoid a mismatch between the water supply and the outlet flow rate, which could lead to a lack of water and cause the fire to escalate. During fire fighting, the inlet and outlet flow rate are monitored and controlled in coordination with the inlet and outlet flow rate regulating valves according to the warning level, i.e., the severity of the fire. This ensures that the inlet and outlet flow rate are balanced or that the inlet flow rate is greater than the outlet flow rate. When the outlet flow rate decreases slightly, the outlet water pressure can be increased to improve the outlet water pressure and coverage of the fire hydrants, thereby improving the efficiency of fire fighting. Active monitoring is conducted specifically for flying sparks, and a dual-line defense system is set up to extinguish them, preventing the fire from spreading across areas and increasing the burden on fire response. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a system structure block diagram of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1 A remote fire monitoring system for ancient villages using the Internet of Things (IoT) that supports multimodal early warning includes: The data detection module uses multispectral high-altitude observation cameras, smoke detectors, temperature detectors, and fire alarms to simultaneously detect light, smoke, temperature, and sound multimodal data on both the exterior and interior of the ancient village buildings. The specific setup is as follows: the outer perimeter of the ancient village is used as the boundary line to delineate the boundary range of the ancient village. At each apex of the geometric shape formed by the boundary line, a multispectral high-altitude observation camera is set up. The multispectral high-altitude observation camera is used to capture and collect visual images of smoke, firelight and flying fire in the ancient village. For each building in the ancient village, the boundary is whether the rooms in each building are interconnected. Interconnected rooms are considered as the same space. Smoke detectors and heat detectors are installed in each space, with the smoke detectors and heat detectors centrally located on the ceiling of the house. Smoke detectors and heat detectors are also installed at equal intervals around the perimeter of each building, with the smoke detectors and heat detectors installed under the eaves around the four sides of the building. Fire alarms with sound and light are installed on the exterior of the building or in the center of the roof. The edge computing module is used to preprocess the acquired multimodal data. The preprocessed data is divided into smoke and temperature time series data and video stream data from multispectral high-altitude surveillance cameras. For time-series data preprocessing, the exponentially weighted average (EWMA) is first used for noise filtering; then, missing values ​​are filled in by linear interpolation. If there are many consecutive missing values, it is marked as a device failure, and the device needs to be checked and the data collected again, and replaced with the historical average value of the corresponding sensor. Outliers are identified and removed by the IQR rule; finally, the data is normalized by Min-Max to prepare for subsequent multimodal feature fusion. For video stream data, color frames are first converted to grayscale frames by grayscale conversion, retaining only the luminance channel to reduce data volume while preserving the contrast features of smoke, flames, and background. Histogram averaging is used to improve image clarity in low-light environments, preventing smoke from being misjudged due to insufficient light. Frames are extracted at fixed intervals of 1-2 seconds to reduce data volume and ensure no key dynamics are missed. The MOG2 algorithm is used to separate smoke, flames, building walls, equipment, and surrounding interference. Smoke and flame areas are extracted and Gaussian filtering is applied to these areas for noise reduction, eliminating false targets caused by floating objects or similar objects in the environment. The optical flow field of adjacent frames is calculated to capture the movement trend of smoke and the flickering characteristics of flames. Multimodal data fusion of smoke, temperature and video stream data uses the frame timestamp of the video stream as a benchmark to match smoke, temperature and video frames within the same time period. One video frame usually contains a smoke value and a temperature value, ensuring the spatiotemporal consistency of the analysis. Before determining the fire warning level and location, it is necessary to first determine the authenticity of the warning caused by smoke and fire. If it is just smoke and fire caused by daily life, it is a false warning and there is no need to conduct further analysis on the warning level and location. If it is determined to be a real fire warning, then further analysis on the warning level and location is required. The methods for determining the authenticity of early warnings are as follows: 1) Distinguishing between smoke and flame characteristics: Smoke differentiation analysis based on spectral and morphological characteristics: Smoke from domestic use and smoke from fires are distinguished by the difference in the proportion of blue light in the scattered light and the spectral distribution of infrared radiation intensity. Generally, smoke from domestic use has a higher proportion of blue light, while smoke from fires due to incomplete combustion has a higher infrared radiation intensity in the 8-14μm band. The spectral characteristics of smoke are collected by a combination of visible light and infrared smoke detectors, and then a classification model is trained using support vector machine (SVM) to distinguish between fire smoke and domestic smoke. The OpenCV contour detection algorithm is used to distinguish between domestic smoke and fire smoke based on the diffusion speed and contour shape of the smoke. Fire smoke usually spreads rapidly and has an irregular contour, while smoke from domestic fires usually rises steadily and has a more regular contour. The circularity and centroid movement trajectory parameters of the smoke region are extracted and the analysis is performed to distinguish between domestic fire smoke and fire smoke. The flickering feature frequency is extracted by calculating the grayscale changes of the flame edge pixels using the optical flow method of visual monitoring. The flickering feature frequency is used to determine whether the flame is produced by domestic fire or a fire. Under normal circumstances, the flickering frequency of domestic fire is low, between 1 and 10 Hz, and the amplitude of the fluctuation is small. On the other hand, the flickering frequency of fire flame is above 15 Hz and the amplitude of the fluctuation is large because the fire burns rapidly and violently. The flickering frequency will gradually increase according to the size of the fire spread. Temperature distribution is captured by thermal infrared sensors, and the temperature field of domestic cooking or fire is distinguished by calculating the standard deviation of the temperature gradient. Usually, the temperature field of domestic cooking is relatively uniform, while the temperature field gradient changes significantly when a fire occurs. 2) Scene-adaptive dynamic weight adjustment mechanism: Based on the distinguishing features of smoke, temperature, and flame in 1), we can construct different scenario libraries. These scenario libraries can include: real fire scenarios, daily fire use scenarios, and sacrificial scenarios. In future monitoring, we will combine the smoke, temperature, and flame feature data with the corresponding new life scenarios and continuously update them. 3) Machine learning algorithm model: The input layer integrates smoke, temperature, flames, and corresponding scene features, specifically the smoke spectral features, smoke contour features, temperature change rate, and fire flash frequency mentioned above. The feature extraction layer processes the temporal data of smoke and temperature using a 3D convolutional neural network (3D-CNN). Here, the smoke contour, spectral feature changes, and flicker frequency changes of fire and temperature change rate are set in 10 consecutive frames of video stream images. The kernel size of the 3D-CNN is set to 3*3*3. The Long Short-Term Memory (LSTM) network is combined to capture the temperature change trend, and the number of hidden layer units of the LSTM is set to 64. The decision-making layer outputs three categories of results: "normal", "suspected", and "fire" based on the judgment of the feature extraction layer. When it is judged as suspected, the location of the fire feature is enlarged and extracted for re-extraction and identification, while human confirmation is used as a double insurance. The modal features and corresponding classification thresholds are automatically adjusted based on scene updates and complexity; for example, in sacrificial scenes or bonfire parties held in ancient villages, the "fire" recognition threshold is increased to avoid misjudgment.

[0019] Once a fire is confirmed to be real, it is necessary to analyze the fire's warning level and location. Smoke and heat detectors deployed inside and outside each building in the ancient village are numbered, and the accurate location coordinates of each group of numbered smoke and heat detectors are determined. The number of each group of smoke and heat detectors is then matched with its location coordinates. After receiving the sensing signals from the smoke and heat detectors, the processor input in the edge computing module determines the detector's number based on the input port. The location coordinates corresponding to these numbers are then used as the warning location. Smoke and flame video stream images are captured using a multispectral high-altitude observation camera. Preprocessing of the video stream visual images extracts frame feature data of smoke and flames, including their contour shapes and smoke diffusion speed. Temperature is captured using a temperature sensor, and the dynamic trend of temperature rise is analyzed. Spatial features of the flame contour, smoke contour, and diffusion range are extracted using a convolutional neural network (CNN). Temporal features of smoke diffusion speed and temperature rise dynamic trends are processed using a long short-term memory network (LSTM). Combining spatial and temporal features, a fully connected layer outputs the fire severity level. Based on fire expert fire severity assessment standards, the following level distinctions are established: If the temperature is less than 60℃, the smoke concentration is low, and there is no obvious flame, it is assessed as the initial stage of fire, level 1; Temperature 60~300℃, smoke concentration moderate, flame area <1m 2 If so, it is assessed as the fire development stage, Level 2; Temperature > 300℃, high smoke concentration, flame area > 5m² 2 If so, it is assessed as a period of intense fire, Level 3; After determining the fire warning level and location, the results are transmitted to the fire response equipment in the warning response module via the communication network module using precise matching of the communication protocol. The communication network module adopts a hybrid networking technology and is also responsible for transmitting the multimodal data collected by the aforementioned data detection module to the edge computing module via the communication protocol. As a preferred implementation, the communication network module includes an edge-cloud collaboration unit. After the edge computing module completes 90% of the real-time data preprocessing and analysis, the edge-cloud collaboration unit uploads key result data, such as warning level and warning location, to the cloud platform for backup through an encrypted channel. This avoids the loss or distortion of key result data during transmission from the communication network module to the warning response module, which would prevent the warning response module from receiving complete warning level and warning location information and thus being unable to make an accurate emergency response. The aforementioned early warning response module includes an emergency linkage control unit, which links the fire response equipment via a communication protocol. The fire response equipment includes: Each household is equipped with an independent indoor fire hydrant, achieving one hydrant per household; the ancient village is divided into several concentric rings by a circle or geometric diagram, and several sets of outdoor fire hydrants are reasonably arranged in each concentric ring according to the size of the perimeter of each concentric ring. The automatic sprinkler heads installed indoors adopt a downspray nozzle structure, with a heat-gathering cover installed below the nozzle. When a fire occurs indoors, the automatic sprinkler heads spray water to extinguish the fire, and at the same time, the heat-gathering cover quickly gathers the radiant heat generated. Several fire monitors are set up outdoors in the ancient village. The fire monitors are evenly distributed in the ancient village. The spray area of ​​several fire monitors can completely cover the area of ​​the ancient village. Similarly, a number of indoor fire hydrants, outdoor fire hydrants, automatic sprinkler heads, and fire monitors are assigned numbers, and each number corresponds to an accurate location coordinate. Based on the early warning location information, the emergency linkage control unit sends control commands to the equipment at the corresponding location coordinates, enabling the fire-fighting facilities at the corresponding location to respond to the fire emergency and complete the fire extinguishing operation. The fire water tank, serving as a centralized water supply station for fire-fighting equipment, plays a crucial role in extinguishing fires. Located at a high point around the ancient village, the fire water tank is equipped with a level sensor. At the inlet, there is an inlet pressure sensor, an inlet flow meter, and a flow regulating valve. At the outlet, there is an outlet flow meter, an emergency shut-off valve, and an outlet flow valve. The level sensor, inlet pressure sensor, inlet flow meter, and outlet flow meter are connected to the control center processor input via a communication network module. The emergency shut-off valve is connected to the output of the control center processor via a communication network module. When there is no fire, the liquid level sensor in the fire water tank monitors the water level in real time. When the water level is lower than the safety line, it needs to be replenished in time to ensure that the water level is always above the safety level. During the fire fighting process, the inlet and outlet flow meters monitor the inlet and outlet flow of the water tank and adjust the inlet and outlet flow through the inlet flow valve and outlet flow valve according to the warning level to ensure that the inlet and outlet are balanced or that the inlet flow is greater than the outlet flow, so as to avoid the inlet supply and outlet flow not reaching an balance.

[0020] The virtual reality simulation module integrates multimodal data collected by the data detection module, including 3D scanning information of ancient village buildings. A digital twin is constructed based on a BIM-GIS 3D model. Combined with on-site meteorological monitoring data, including wind direction, wind speed, air temperature and humidity, and air pressure, outlier values ​​are removed. Fluid dynamics is used to simulate the spread of fire and smoke, allowing for early prediction of fire development trends. This embodiment employs a Fire Dynamics Simulator (FDS). The aforementioned data is input into the simulator to perform fluid dynamics simulations of the fire's spread direction, intensity, and extent. Based on the simulation results, response commands for the early warning response module can be generated in advance, along with resource scheduling plans. Safe rescue routes and evacuation routes are planned, improving response accuracy and shortening rescue time. The 3D visualization monitoring module is used to model the fire response equipment under the early warning response module and the buildings in the ancient village by point, and connects with the edge computing module. When a fire occurs, the 3D visualization platform immediately roams to the early warning location and forms an associated early warning sequence diagram according to the early warning sequence. The fire response equipment set around the early warning sequence diagram is put into standby status under the command of the emergency linkage control unit. Example 2

[0021] Based on Embodiment 1, the data detection module in Embodiment 1 further includes: an electricity fingerprint recognition and acquisition unit. An electricity fingerprint recognition and acquisition module is installed at the distribution box of each household on the electricity circuit of the ancient village. The power switch can only be turned on after successful fingerprint recognition. The fingerprint information is transmitted to the control center for record-keeping through the communication network module. One fingerprint information corresponds to one building. For the use of high-power electrical products such as electric vehicles and electric stoves, the building corresponding to the fingerprint information is locked and displayed in three dimensions on the three-dimensional visualization monitoring module, focusing on the temperature, smoke and video images of the household. Once a fire occurs due to high-power electricity use, it can be detected and dealt with immediately. Example 3

[0022] Based on Example 1, a flying fire defense unit is set up under the early warning response module. The outdoor fire hydrants in the boundary area of ​​the early warning response module in Example 1 are used as the first line of defense, and the fire monitors are used as the second line of defense. Several multispectral high-altitude observation cameras set in the data detection module are used as monitoring means to perform multispectral detection and flying fire monitoring in the area. After the location area of ​​the flying fire is determined by the edge computing module, the early warning response module controls the fire hydrants and fire monitors in the flying fire area to enter the emergency response state to extinguish the flying fire.

Claims

1. A remote fire monitoring system for ancient villages using the Internet of Things (IoT) that supports multimodal early warning, characterized in that: include: The data detection module is used to simultaneously detect light, smoke, temperature, and sound multimodal monitoring data on both the exterior and interior of buildings in ancient villages. The edge computing module is used to preprocess the collected multimodal data, and to judge the authenticity of the early warning by constructing a scene feature model based on the preprocessed multimodal data. Once a real fire warning is issued, the warning level and location are analyzed using an algorithm model based on the pre-processed multimodal data. The communication network module, employing hybrid networking technology, transmits multimodal data collected by the data detection module to the edge computing module via a communication protocol; simultaneously, it precisely matches the fire early warning analysis results from the edge computing module to the fire response equipment in the early warning response module via the communication protocol. The early warning response module has multiple levels of fire response equipment. It receives fire early warning analysis results from the edge computing module, generates matching response commands, and controls the corresponding fire response equipment to complete the command actions through the response commands. The virtual reality simulation exercise module integrates multimodal data collected by the data detection module, including 3D scanning information of ancient village buildings, and constructs a digital twin based on the BIM-GIS 3D model; combined with on-site meteorological monitoring data, it uses fluid dynamics to simulate fire and smoke diffusion paths, predicts the development trend of fire in advance, generates response instructions for the early warning response module in advance, and generates resource scheduling plans in advance. The 3D visualization monitoring module is used to model the fire response equipment and buildings in the ancient village under the early warning response module by point, and connect with the edge computing module. When a fire occurs, the 3D visualization platform immediately roams to the early warning location, forms an associated early warning sequence diagram according to the early warning sequence, and the fire response equipment set around the early warning sequence diagram is put into a standby state.

2. The IoT-based remote fire monitoring system for ancient villages supporting multimodal early warning as described in claim 1, characterized in that, The light, smoke, temperature, and sound multimodal data are captured and collected by setting up multispectral high-altitude observation cameras, smoke detectors, temperature detectors, and fire sound and light alarms, respectively. Using the outer perimeter of the ancient village as a boundary, a multispectral high-altitude observation camera is installed at each apex of the geometric shape formed by the boundary. The multispectral high-altitude observation camera is used to capture and collect visual images of smoke, firelight, and flying fire in the ancient village. Within each building of the ancient village, the rooms are divided into two sections based on whether they are interconnected. Interconnected rooms are considered as the same space, and smoke detectors and temperature detectors are installed in each space. Smoke detectors and temperature detectors are also installed at equal intervals around the perimeter of each building. The fire alarm with sound and light is installed on the exterior of the building.

3. The IoT-based remote fire monitoring system for ancient villages supporting multimodal early warning as described in claim 1, characterized in that, The multimodal data preprocessing is divided into preprocessing of smoke and temperature sensing time-series data and video stream data from multispectral high-altitude observation cameras; The time-series data preprocessing first uses EWMA for noise filtering, then linear interpolation to fill in missing values, and IQR rule to identify and remove outliers; finally, Min-Max is used to normalize the data. The video stream data is first converted from color frames to grayscale frames by grayscale conversion, retaining only the luminance channel to reduce the amount of data, while preserving the contrast features of smoke, flames and background. Histogram averaging is used to improve image clarity in low-light environments and prevent smoke from being misjudged due to insufficient light. Frames are extracted at fixed intervals to reduce data volume and ensure that no key dynamics are missed. The MOG2 algorithm is used to separate smoke, fire, walls, and equipment. Smoke and fire areas are extracted and Gaussian filtering is applied to these areas for noise reduction. Calculate the optical flow field of adjacent frames to capture the motion trend of smoke and the flickering characteristics of firelight; Smoke detector, temperature sensor, and video stream data are matched with frames within the same time period based on the frame timestamps of the video stream to ensure spatiotemporal consistency of the analysis.

4. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The method for determining the authenticity of the early warning is as follows: 1) Distinguishing between smoke and flame characteristics: Smoke differentiation analysis based on spectral and morphological characteristics: Smoke from daily life and smoke from fires can be distinguished by the difference in the proportion of blue light in the scattered light and the difference in the spectral distribution of infrared radiation intensity. The OpenCV contour detection algorithm is used to distinguish between residential smoke and fire smoke based on the diffusion speed and contour shape of the smoke. The flickering feature frequency is extracted by calculating the grayscale changes of the pixels at the edge of the flame using the optical flow method of visual monitoring. The flickering feature frequency is then used to determine whether the flame is from domestic use or a fire. Temperature distribution is captured by thermal infrared sensing, and the temperature field of domestic fire and fire temperature field are distinguished by calculating the standard deviation of the temperature gradient. 2) Scene-adaptive dynamic weight adjustment mechanism: Based on the distinguishing features of smoke, temperature, and flame in 1), we will construct different scenario libraries and continuously update them in future monitoring by combining the smoke, temperature, and flame feature data with the corresponding new life scenarios. 3) Machine learning algorithm model: The input layer integrates smoke, temperature, flames, and corresponding scene features; The feature extraction layer processes temporal data using 3D-CNN, capturing the smoke morphology changes across 10 consecutive frames. The kernel size of the 3D-CNN is set to 3*3*3. LSTM is then used to capture temperature change trends, with the number of hidden layer units in the LSTM set to 64. The decision-making layer outputs three categories of results based on the judgment of the feature extraction layer: "normal", "suspected", and "fire".

5. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The analysis model for warning levels and warning locations is as follows: The smoke detectors and heat detectors deployed inside and outside each building in the ancient village are numbered, and the accurate location coordinates of each group of numbered smoke detectors and heat detectors are determined. The number of each group of smoke detectors and heat detectors is matched with its location coordinates. After receiving the sensing signals from the smoke detectors and heat detectors, the processor input terminal in the edge computing module determines the number of the smoke detectors and heat detectors based on the input port. The location coordinates are then matched with the corresponding location coordinates, which are the warning locations. Smoke and flame video stream images are captured by a multispectral high-altitude observation camera. After preprocessing the visual images of the video stream, frame feature data of smoke and flame are extracted, the contour shape of smoke and flame and the smoke diffusion speed are extracted, and temperature is captured by a temperature sensor to analyze the dynamic trend of temperature rise. Spatial features of the spatial contour shape of flame, smoke and diffusion range are extracted by CNN. The temporal characteristics of smoke diffusion rate and temperature rise dynamics are processed by LSTM; the severity level of the fire is output through a fully connected layer by combining spatial and temporal characteristics.

6. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The communication network module includes an edge-cloud collaboration unit. After the edge computing module completes 90% of the real-time data preprocessing and analysis, the edge-cloud collaboration unit uploads the key result data to the cloud platform for backup through an encrypted channel.

7. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The data detection module also includes an electricity fingerprint recognition and collection unit. An electricity fingerprint recognition and collection module is installed at the distribution box of each household on the electricity circuit of the ancient village. The power can only be turned on after the fingerprint is successfully collected and recognized. The fingerprint information is transmitted to the control center for record through the communication network module. For the use of high-power electrical products, the building corresponding to the fingerprint information is locked and displayed in three dimensions on the three-dimensional visualization monitoring module for key monitoring.

8. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The early warning response module includes an emergency linkage control unit, which is used to link fire response equipment through a communication protocol. The fire response equipment includes: Each household is equipped with an independent indoor fire hydrant, achieving one hydrant per household; the ancient village is divided into several concentric rings by a circle or geometric diagram, and several sets of outdoor fire hydrants are reasonably arranged in each concentric ring according to the size of the perimeter of each concentric ring. The automatic sprinkler head installed indoors adopts a downspray nozzle structure, with a heat-concentrating cover installed below the nozzle. Several fire monitors are set up outdoors in the ancient village. The fire monitors are evenly distributed within the ancient village, and the spray area of ​​the fire monitors completely covers the area of ​​the ancient village. Similarly, the aforementioned indoor fire hydrants, outdoor fire hydrants, automatic sprinkler heads, and fire monitors are all numbered, and each number corresponds to an accurate location coordinate; based on the warning location information, the fire-fighting facilities at the corresponding locations are activated for fire emergency response; A fire-fighting water tank is located at a high point around the ancient village. A level sensor is installed inside the fire-fighting water tank. An inlet pressure sensor, an inlet flow meter, and a flow regulating valve are installed at the inlet of the fire-fighting water tank. An outlet flow meter, an emergency shut-off valve, and an outlet flow valve are installed at the outlet of the fire-fighting water tank. The level sensor, inlet pressure sensor, inlet flow meter, and outlet flow meter are connected to the input terminal of the control center processor via a communication network module. The flow regulating valve and the emergency shut-off valve are connected to the output terminal of the control center processor via the communication network module.

9. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The early warning response module also includes a flying fire defense unit. The flying fire defense unit uses outdoor fire hydrants in the boundary area of ​​the early warning response module as the first line of defense and fire monitors as the second line of defense. It uses several multispectral high-altitude observation cameras set in the data detection module as monitoring means to perform multispectral detection and flying fire monitoring in the area. After the location area of ​​the flying fire is determined by the edge computing module, the early warning response module controls the fire hydrants and fire monitors in the flying fire area to enter the emergency response state to extinguish the flying fire.

10. A remote fire monitoring system for ancient villages supporting multimodal early warning via the Internet of Things, as described in claim 1, is characterized in that, The meteorological monitoring data includes wind direction, wind speed, air temperature and humidity, and air pressure. Outliers in the collected meteorological data need to be removed, and the meteorological data is input into FDS to perform fluid dynamics simulation of the fire.

Images