Sensing data collecting and monitoring method for forest fire early warning

Abstract

The invention relates to the technical field of disaster early warning, in particular to a sensing data collecting and monitoring method for forest fire early warning, and the method comprises the steps: constructing a three-dimensional combustible environment model of a mountainous area; acquiring different sensor data of the to-be-monitored area, and performing feature value extraction on the acquired sensor data to obtain a multi-dimensional data risk feature value; according to the multi-dimensional data risk characteristic value, fire behavior initial judgment is carried out, the risk level is determined, and forest fire early warning is carried out based on the risk level; fusing all sensor data with the three-dimensional combustible environment model to obtain a three-dimensional fire situation model; and performing fire spreading prediction based on the machine learning model, calculating a fire danger index, and performing forest fire early warning according to the fire danger index. High-risk fire can be rapidly identified and an alarm is given, fire spreading prediction is carried out on medium-risk conditions, and mountain fire monitoring and early warning are dynamically carried out.

Description

Technical Field

[0001] This invention relates to the field of disaster early warning technology, specifically to a sensor data acquisition and monitoring method for wildfire early warning. Background Technology

[0002] Forest fires are a highly destructive natural disaster, and their prevention and monitoring currently rely mainly on two methods: traditional manual observation towers and satellite remote sensing technology.

[0003] However, despite their effectiveness in past fire prevention practices, these methods still have many limitations. Manual lookout tower observations are limited by factors such as field of vision and personnel fatigue, often resulting in significant delays in early warning. While satellite remote sensing technology has a wide coverage area, its sensitivity to small-scale, early-stage fires is insufficient, and its monitoring effectiveness is easily affected by weather conditions such as clouds and fog, leading to untimely or inaccurate information acquisition. Furthermore, while existing sensor network monitoring systems can detect and alert to fires using temperature sensors or smoke detectors, they lack multi-parameter, multi-level comprehensive monitoring capabilities, making it difficult to achieve dynamic perception and accurate prediction of fire development, and thus unable to provide sufficient support for early fire prevention and emergency decision-making. Summary of the Invention

[0004] To address the above problems, this invention provides a sensor data acquisition and monitoring method for wildfire early warning, comprising: Constructing a three-dimensional combustible environment model in mountainous areas: acquiring coordinate data, elevation data, and multispectral remote sensing images of the area to be monitored, constructing a three-dimensional terrain model of the area to be monitored; and extracting vegetation data of the area to be monitored and mapping it to the three-dimensional terrain model to obtain a three-dimensional combustible environment model in mountainous areas. Data from different sensors in the area to be monitored are acquired, including real-time temperature data, smoke concentration data, and meteorological data; feature values ​​are extracted from the acquired sensor data to obtain multidimensional data risk feature values. Based on the risk characteristic values ​​of multidimensional data, an initial assessment of the fire situation is made to determine the risk level, and a wildfire warning is issued based on the risk level. When the risk level is medium, all sensor data are fused with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model; by acquiring meteorological data for future periods, fire spread is predicted based on a machine learning model, the fire risk index is calculated, and wildfire warnings are issued based on the fire risk index.

[0005] The specific steps for constructing a three-dimensional combustible environment model in mountainous areas are as follows: The elevation data and high-resolution oblique camera images of the area to be monitored are acquired, including images at vertical and multiple oblique angles; and a coordinate system of the area to be monitored is established based on the geographic reference coordinates of the oblique camera images. Elevation data is converted into point cloud data, and the point cloud data is denoised and classified to generate a high-precision digital elevation model and digital surface model; image correction is performed on the tilted camera images; Based on the processed oblique camera images, aerial triangulation is performed through feature point extraction and matching to solve the exterior orientation elements and object coordinates of densified points in each image; dense matching is performed to match the corresponding points of each pixel in different images to generate a high-density three-dimensional point cloud, and the discrete three-dimensional point cloud is connected into a continuous three-dimensional network surface, which is the three-dimensional geometric shape of the terrain of the area to be monitored. The tilted camera images are mapped onto a 3D network surface based on the calculated exterior orientation elements, and then layered and fused according to the digital elevation model and digital surface model to obtain a 3D terrain model. The vegetation index is calculated based on the oblique camera images, and the vegetation information is classified to divide the area to be monitored into different combustible types. The combustible distribution map obtained by the classification is superimposed on the three-dimensional terrain model to obtain a three-dimensional combustible environment model of the mountainous area.

[0006] The multidimensional data risk characteristics include trend characteristics, abnormal propagation speed characteristics, wind speed characteristics, and humidity characteristics.

[0007] The trend characteristic value is obtained by linearly fitting the data sequences of all sensor nodes in the monitoring area, taking the maximum value of the slope of all fitted lines as the trend value of the sensor monitoring type, and then normalizing it linearly.

[0008] The specific operation of making an initial fire assessment based on multidimensional data risk characteristic values, determining the risk level, and issuing a wildfire warning based on the risk level is as follows: The maximum value of the trend feature value among all sensor monitoring types is selected as the maximum trend feature value for initial fire assessment. Set thresholds for fire trend, propagation speed, wind speed, and humidity respectively; compare the maximum change trend characteristic value with the calculated abnormal propagation speed characteristic value, wind speed characteristic value, and humidity characteristic value with the set thresholds respectively; If the maximum change trend characteristic value is greater than the fire trend threshold and the abnormal propagation speed characteristic value is greater than the propagation speed threshold, then the area to be monitored is determined to be a fire zone, which is at a high risk level, and a wildfire warning is immediately issued. If the maximum change trend characteristic value is greater than the fire trend threshold and the abnormal propagation speed characteristic value is less than or equal to the propagation speed threshold, a fire warning mark is made, and the thresholds of wind speed characteristic value and humidity characteristic value are compared. If the wind speed characteristic value is greater than the wind speed threshold or the humidity characteristic value is less than the humidity threshold, it is a medium risk level, and the fire spread prediction is made for the monitored area. If any of the above criteria are not met, the area to be monitored is determined to be a safe zone and is classified as low-risk.

[0009] The process of fusing all sensor data with a three-dimensional combustible material environment model to obtain a three-dimensional fire situation model is as follows: Based on real-time temperature data and smoke concentration data, determine the temperature anomaly area and smoke identification area of ​​the area to be monitored; superimpose the temperature anomaly area and smoke identification area onto the surface of the three-dimensional combustible environment model; and obtain the forecast meteorological data for future time periods, and convert it from discrete station data into meteorological field data covering the entire area to be monitored through spatial interpolation. The three-dimensional combustible environment model is meshed, and a prior ignition probability is assigned to each mesh based on vegetation index, combustible type, and slope aspect. According to Bayes' theorem, the posterior ignition probability of each mesh is calculated, and the two-dimensional matrix of the fire probability field is output. A three-dimensional fire situation model is obtained by fusing fire probability field and meteorological field data with the superimposed three-dimensional combustible environment model.

[0010] The machine learning model uses a convolutional long short-term memory network model.

[0011] The specific steps for predicting fire spread and calculating the fire risk index based on a machine learning model are as follows: Build a pre-trained machine learning model; The current moment is used as the input sequence of the three-dimensional fire situation model of multiple consecutive time periods. The input is fed into the machine learning model. The encoder processes the input sequence of past time periods, extracts and compresses the spatiotemporal features, and outputs the feature vector of past fire situation. The feature vector from the encoder is received, and the feature vector of the future fire state is predicted by the decoder. The machine learning model outputs the fire probability, spread direction and spread speed at each future time step based on the feature vector from the decoder. The fire risk index is calculated based on the fire probability, spread direction, and spread speed obtained from the output.

[0012] The monitoring and early warning system is based on the fire risk index. If the fire risk index is greater than or equal to the high-risk threshold, a fire zone is identified in the monitored area, and a fire alarm is immediately issued. If the fire risk index is less than the high-risk threshold but greater than or equal to the medium-risk threshold, sensor data is reacquired, and fire spread prediction is performed. If the fire risk index is less than the medium-risk threshold, the monitored area is identified as a safe zone with a low-risk level, and the fire warning marker is canceled.

[0013] This invention also provides a sensor data acquisition and monitoring system for wildfire early warning, comprising: The 3D model building module is used to construct a 3D combustible environment model in mountainous areas: it acquires coordinate data, elevation data and multispectral remote sensing images of the area to be monitored, constructs a 3D terrain model of the area to be monitored, and extracts vegetation data of the area to be monitored and maps it to the 3D terrain model to obtain a 3D combustible environment model in mountainous areas. The data feature extraction module is used to acquire different sensor data of the area to be monitored, including real-time temperature data, smoke concentration data, and meteorological data; and to extract feature values ​​from the acquired sensor data to obtain multi-dimensional data risk feature values. The initial fire assessment module is used to make an initial fire assessment based on multi-dimensional data risk characteristic values, determine the risk level, and issue wildfire warnings based on the risk level. The prediction and early warning module is used to fuse all sensor data with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model when the risk level is medium. By acquiring meteorological data for future periods, it predicts the spread of fire based on a machine learning model, calculates the fire risk index, and issues wildfire warnings based on the fire risk index.

[0014] Beneficial Effects: This invention provides a sensor data acquisition and monitoring method for wildfire early warning. By constructing a three-dimensional combustible environment model of mountainous areas, it can accurately present the topography and distribution of combustibles in the monitored area, overcoming the limitations of traditional two-dimensional models or single data. By acquiring data from different sensors and extracting features from multiple sources, multi-dimensional data risk feature values ​​are obtained. Combined with key indicators such as changing trends and abnormal propagation speed, initial fire situation assessments are made, enabling rapid identification of high-risk fires and issuance of alarms. Simultaneously, fire spread prediction is performed for medium-risk situations, reducing false alarms and missed alarms. All sensor data are fused with the three-dimensional combustible environment model to obtain a three-dimensional fire situation model. Combined with meteorological data, a machine learning model is used to predict fire spread and calculate the fire risk index. Based on the comparison results of the fire risk index with different thresholds, wildfire monitoring and early warning are dynamically implemented. Detailed Implementation

[0015] Exemplary embodiments of this disclosure will now be described in more detail.

[0016] Example This embodiment provides a sensor data acquisition and monitoring method for wildfire early warning. The specific implementation steps of the method are as follows: Constructing a three-dimensional combustible environment model in mountainous areas: acquiring coordinate data, elevation data, and multispectral remote sensing images of the area to be monitored, constructing a three-dimensional terrain model of the area to be monitored; and extracting vegetation data of the area to be monitored and mapping it to the three-dimensional terrain model to obtain a three-dimensional combustible environment model in mountainous areas. Data from different sensors in the area to be monitored are acquired, including real-time temperature data, smoke concentration data, and meteorological data; feature values ​​are extracted from the acquired sensor data to obtain multidimensional data risk feature values. Based on the risk characteristic values ​​of multidimensional data, an initial assessment of the fire situation is made to determine the risk level, and a wildfire warning is issued based on the risk level. When the risk level is medium, all sensor data are fused with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model; by acquiring meteorological data for future periods, fire spread is predicted based on a machine learning model, the fire risk index is calculated, and wildfire warnings are issued based on the fire risk index.

[0017] Step S1, constructing a three-dimensional combustible environment model in mountainous areas: acquiring coordinate data, elevation data, and multispectral remote sensing images of the area to be monitored, constructing a three-dimensional terrain model of the area to be monitored; and extracting vegetation data of the area to be monitored, mapping it to the three-dimensional terrain model to obtain a three-dimensional combustible environment model in mountainous areas. The specific operations are as follows: S101. Obtain elevation data and high-resolution oblique camera images of the area to be monitored, wherein the oblique camera images include images at vertical and multiple oblique angles; and establish a coordinate system for the area to be monitored based on the geographic reference coordinates inherent in the oblique camera images. S102. Convert the elevation data into point cloud data, and perform noise reduction and classification processing on the point cloud data to generate a high-precision digital elevation model and digital surface model; perform image correction on the tilted camera image; S103. Based on the processed oblique camera images, aerial triangulation is performed through feature point extraction and matching to solve the exterior orientation elements (position and attitude) and object coordinates of densified points for each image; dense matching is performed to match the corresponding points of each pixel in different images to generate a high-density three-dimensional point cloud, and the discrete three-dimensional point cloud is connected into a continuous three-dimensional network surface, which is the three-dimensional geometric shape of the terrain of the area to be monitored. S104. Map the tilted camera images onto the three-dimensional network surface according to the calculated exterior orientation elements, and then perform layer-by-layer fusion based on the digital elevation model and the digital surface model to obtain a three-dimensional terrain model. S105. Calculate the vegetation index based on the oblique camera image, classify the vegetation information, divide the area to be monitored into different combustible types, and overlay the combustible distribution map obtained by classification onto the three-dimensional terrain model to obtain a three-dimensional combustible environment model of the mountainous area.

[0018] Step S2: Acquire data from different sensors in the area to be monitored, including real-time temperature data, smoke concentration data, and meteorological data; extract feature values ​​from the acquired sensor data to obtain multidimensional data risk feature values; Real-time temperature data from the thermal infrared sensor deployed in the area to be monitored, smoke concentration data from the smoke detection sensor and the visible light camera, and real-time wind speed, wind direction, humidity, and temperature data from the meteorological sensor are acquired respectively.

[0019] The specific method for extracting feature values ​​from the acquired sensor data to obtain multidimensional data risk feature values ​​is as follows: The acquired data from different types of sensors are time-stamped and synchronized to obtain the data sequence of each sensor; Linear fitting is performed on the data sequences of all sensor nodes in the monitoring area, and the maximum value of the slope of all fitted lines is taken as the trend value of the sensor monitoring type. After linear normalization, the trend characteristic value is obtained. For any sensor monitoring type, calculate the standard deviation of the sensor data sequence of all sensor nodes in the monitored area at each monitoring time to obtain the data difference sequence; calculate the absolute value of the difference between each two adjacent data in the sensor data sequence, and sum all the absolute values ​​of the difference to obtain the anomaly propagation speed of the sensor monitoring type; calculate the average value of the anomaly propagation speed of all sensor monitoring types, and obtain the anomaly propagation speed characteristic value of the monitored area after linear normalization; Calculate the average wind speed and average relative humidity of all sensor nodes in the monitored area at all monitoring times, and obtain the wind speed characteristic value and humidity characteristic value after linear normalization. The changing trend characteristics, abnormal propagation speed characteristics, wind speed characteristics, and humidity characteristics of all sensor monitoring types are combined to form a multidimensional data risk characteristic value.

[0020] Step S3: Make an initial assessment of the fire situation based on the risk characteristic values ​​of multidimensional data, determine the risk level, and issue a wildfire warning based on the risk level. The specific operation is as follows: The maximum value of the trend feature value among all sensor monitoring types is selected as the maximum trend feature value for initial fire assessment. Set thresholds for fire trend, propagation speed, wind speed, and humidity respectively; compare the maximum change trend characteristic value with the calculated abnormal propagation speed characteristic value, wind speed characteristic value, and humidity characteristic value with the set thresholds respectively; If the maximum change trend characteristic value is greater than the fire trend threshold, and the abnormal propagation speed characteristic value is greater than the propagation speed threshold, then the area to be monitored is determined to be a fire zone, which is of high risk level. A fire alarm should be issued immediately, indicating that a certain sensor monitoring type has changed drastically and is spreading rapidly in space. This is direct proof of the occurrence and spread of open flames. If the maximum change trend characteristic value is greater than the fire trend threshold and the abnormal propagation speed characteristic value is less than or equal to the propagation speed threshold, a fire warning mark is made, and the thresholds of wind speed characteristic value and humidity characteristic value are compared. If the wind speed characteristic value is greater than the wind speed threshold or the humidity characteristic value is less than the humidity threshold, it is a medium risk level, and the fire spread prediction is made for the monitored area. If any of the above criteria are not met, the area to be monitored is determined to be a safe zone and is classified as low-risk.

[0021] S4. When there is a medium risk level, all sensor data are fused with the three-dimensional combustible environment model to obtain a three-dimensional fire situation model; by acquiring meteorological data for future periods, fire spread is predicted based on machine learning models, the fire risk index is calculated, and wildfire warnings are issued based on the fire risk index. The method for fusing all sensor data with a three-dimensional combustible material environment model to obtain a three-dimensional fire situation model is as follows: Based on real-time temperature data and smoke concentration data, determine the temperature anomaly area and smoke identification area of ​​the area to be monitored; superimpose the temperature anomaly area and smoke identification area onto the surface of the three-dimensional combustible environment model; and obtain the forecast meteorological data for future time periods, and convert it from discrete station data into meteorological field data covering the entire area to be monitored through spatial interpolation. The three-dimensional combustible environment model is meshed, and a prior ignition probability is assigned to each mesh based on vegetation index, combustible type, and slope aspect. According to Bayes' theorem, the posterior ignition probability of each mesh is calculated, and the two-dimensional matrix of the fire probability field is output. A three-dimensional fire situation model is obtained by fusing fire probability field and meteorological field data with the superimposed three-dimensional combustible environment model.

[0022] The specific method for predicting fire spread based on a machine learning model, calculating a fire risk index, and conducting monitoring and early warning based on the fire risk index is as follows: A trained machine learning model is constructed, which employs a convolutional long short-term memory network model that can simultaneously capture spatial features and the dependencies of event sequences.

[0023] The three-dimensional fire situation model of multiple consecutive time periods (including the past 6 hours and the next 6 hours) is used as the input sequence and input into the machine learning model. The encoder processes the input sequence of past time periods, extracts and compresses the spatiotemporal features, and outputs the feature vector of the past fire situation. The feature vector from the encoder is received, and the feature vector of the future fire state is predicted by the decoder. The machine learning model outputs the fire probability, spread direction and spread speed at each future time step based on the feature vector from the decoder. The fire risk index is calculated based on the fire probability, spread direction, and spread speed obtained from the output.

[0024] When the fire risk index is calculated, if the fire risk index is greater than or equal to the high-risk threshold, it is determined that there is a fire zone in the area to be monitored, and a fire alarm is immediately issued; if the fire risk index is less than the high-risk threshold but greater than or equal to the medium-risk threshold, sensor data is reacquired, and fire spread prediction is performed; if the fire risk index is less than the medium-risk threshold, the area to be monitored is determined to be a safe zone with a low-risk level, and the fire warning mark is canceled.

[0025] Finally, this embodiment also provides a sensor data acquisition and monitoring system for wildfire early warning, including: The 3D model building module is used to construct a 3D combustible environment model in mountainous areas: it acquires coordinate data, elevation data and multispectral remote sensing images of the area to be monitored, constructs a 3D terrain model of the area to be monitored, and extracts vegetation data of the area to be monitored and maps it to the 3D terrain model to obtain a 3D combustible environment model in mountainous areas. The data feature extraction module is used to acquire different sensor data of the area to be monitored, including real-time temperature data, smoke concentration data, and meteorological data; and to extract feature values ​​from the acquired sensor data to obtain multi-dimensional data risk feature values. The initial fire assessment module is used to make an initial fire assessment based on multi-dimensional data risk characteristic values, determine the risk level, and issue wildfire warnings based on the risk level. The prediction and early warning module is used to fuse all sensor data with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model when the risk level is medium. By acquiring meteorological data for future periods, it predicts the spread of fire based on a machine learning model, calculates the fire risk index, and issues wildfire warnings based on the fire risk index.

Claims

1. A sensor data acquisition and monitoring method for wildfire early warning, characterized in that, include: Constructing a three-dimensional combustible environment model in mountainous areas: acquiring coordinate data, elevation data, and multispectral remote sensing images of the area to be monitored, constructing a three-dimensional terrain model of the area to be monitored; and extracting vegetation data of the area to be monitored and mapping it to the three-dimensional terrain model to obtain a three-dimensional combustible environment model in mountainous areas. Data from different sensors in the area to be monitored are acquired, including real-time temperature data, smoke concentration data, and meteorological data; feature values ​​are extracted from the acquired sensor data to obtain multidimensional data risk feature values. Based on the risk characteristic values ​​of multidimensional data, an initial assessment of the fire situation is made to determine the risk level, and a wildfire warning is issued based on the risk level. When the risk level is medium, all sensor data are fused with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model; by acquiring meteorological data for future periods, fire spread is predicted based on a machine learning model, the fire risk index is calculated, and wildfire warnings are issued based on the fire risk index.

2. The sensor data acquisition and monitoring method according to claim 1, characterized in that, The specific steps for constructing a three-dimensional combustible environment model in mountainous areas are as follows: The elevation data and high-resolution oblique camera images of the area to be monitored are acquired, including images at vertical and multiple oblique angles; and a coordinate system of the area to be monitored is established based on the geographic reference coordinates of the oblique camera images. Elevation data is converted into point cloud data, and the point cloud data is denoised and classified to generate a high-precision digital elevation model and digital surface model; image correction is performed on the tilted camera images; Based on the processed oblique camera images, aerial triangulation is performed through feature point extraction and matching to solve the exterior orientation elements and object coordinates of densified points in each image; dense matching is performed to match the corresponding points of each pixel in different images to generate a high-density three-dimensional point cloud, and the discrete three-dimensional point cloud is connected into a continuous three-dimensional network surface, which is the three-dimensional geometric shape of the terrain of the area to be monitored. The tilted camera images are mapped onto a 3D network surface based on the calculated exterior orientation elements, and then layered and fused according to the digital elevation model and digital surface model to obtain a 3D terrain model. The vegetation index is calculated based on the oblique camera images, and the vegetation information is classified to divide the area to be monitored into different combustible types. The combustible distribution map obtained by the classification is superimposed on the three-dimensional terrain model to obtain a three-dimensional combustible environment model of the mountainous area.

3. The sensor data acquisition and monitoring method according to claim 1, characterized in that, The multidimensional data risk characteristics include trend characteristics, abnormal propagation speed characteristics, wind speed characteristics, and humidity characteristics.

4. The sensor data acquisition and monitoring method according to claim 3, characterized in that, The trend characteristic value is obtained by linearly fitting the data sequences of all sensor nodes in the monitoring area, taking the maximum value of the slope of all fitted lines as the trend value of the sensor monitoring type, and then normalizing it linearly.

5. The sensor data acquisition and monitoring method according to claim 4, characterized in that, The specific operation of making an initial fire assessment based on multidimensional data risk characteristic values, determining the risk level, and issuing a wildfire warning based on the risk level is as follows: The maximum value of the trend feature value among all sensor monitoring types is selected as the maximum trend feature value for initial fire assessment. Set thresholds for fire trend, propagation speed, wind speed, and humidity respectively; The maximum change trend characteristic value, the calculated abnormal propagation speed characteristic value, wind speed characteristic value, and humidity characteristic value are compared with the set thresholds respectively; If the maximum change trend characteristic value is greater than the fire trend threshold and the abnormal propagation speed characteristic value is greater than the propagation speed threshold, then the area to be monitored is determined to be a fire zone, which is at a high risk level, and a wildfire warning is immediately issued. If the maximum change trend characteristic value is greater than the fire trend threshold and the abnormal propagation speed characteristic value is less than or equal to the propagation speed threshold, a fire warning mark is made, and the thresholds of wind speed characteristic value and humidity characteristic value are compared. If the wind speed characteristic value is greater than the wind speed threshold or the humidity characteristic value is less than the humidity threshold, it is a medium risk level, and the fire spread prediction is made for the monitored area. If any of the above criteria are not met, the area to be monitored is determined to be a safe zone and is classified as low-risk.

6. The sensor data acquisition and monitoring method according to claim 1, characterized in that, The process of fusing all sensor data with a three-dimensional combustible material environment model to obtain a three-dimensional fire situation model is as follows: Based on real-time temperature data and smoke concentration data, determine the temperature anomaly area and smoke identification area of ​​the area to be monitored; superimpose the temperature anomaly area and smoke identification area onto the surface of the three-dimensional combustible environment model; and obtain the forecast meteorological data for future time periods, and convert it from discrete station data into meteorological field data covering the entire area to be monitored through spatial interpolation. The three-dimensional combustible environment model is meshed, and a prior ignition probability is assigned to each mesh based on vegetation index, combustible type, and slope aspect. According to Bayes' theorem, the posterior ignition probability of each mesh is calculated, and the two-dimensional matrix of the fire probability field is output. A three-dimensional fire situation model is obtained by fusing fire probability field and meteorological field data with the superimposed three-dimensional combustible environment model.

7. The sensor data acquisition and monitoring method according to claim 1, characterized in that, The machine learning model uses a convolutional long short-term memory network model.

8. The sensor data acquisition and monitoring method according to claim 7, characterized in that, The specific steps for predicting fire spread and calculating the fire risk index based on a machine learning model are as follows: Build a pre-trained machine learning model; The current moment is used as the input sequence of the three-dimensional fire situation model of multiple consecutive time periods. The input is fed into the machine learning model. The encoder processes the input sequence of past time periods, extracts and compresses the spatiotemporal features, and outputs the feature vector of past fire situation. The feature vector from the encoder is received, and the feature vector of the future fire state is predicted by the decoder. The machine learning model outputs the fire probability, spread direction and spread speed at each future time step based on the feature vector from the decoder. The fire risk index is calculated based on the fire probability, spread direction, and spread speed obtained from the output.

9. The sensor data acquisition and monitoring method according to claim 1, characterized in that, The monitoring and early warning system is based on the fire risk index. If the fire risk index is greater than or equal to the high-risk threshold, a fire zone is identified in the area to be monitored, and a fire alarm is immediately issued. If the fire risk index is less than the high-risk threshold but greater than or equal to the medium-risk threshold, sensor data is reacquired, and fire spread prediction is performed. If the fire risk index is less than the medium risk threshold, the area to be monitored is determined to be a safe zone with a low risk level, and the fire warning sign is cancelled.

10. A sensor data acquisition and monitoring system for wildfire early warning, characterized in that, include: The 3D model building module is used to construct a 3D combustible environment model in mountainous areas: it acquires coordinate data, elevation data and multispectral remote sensing images of the area to be monitored, constructs a 3D terrain model of the area to be monitored, and extracts vegetation data of the area to be monitored and maps it to the 3D terrain model to obtain a 3D combustible environment model in mountainous areas. The data feature extraction module is used to acquire different sensor data of the area to be monitored, including real-time temperature data, smoke concentration data, and meteorological data; and to extract feature values ​​from the acquired sensor data to obtain multi-dimensional data risk feature values. The initial fire assessment module is used to make an initial fire assessment based on multi-dimensional data risk characteristic values, determine the risk level, and issue wildfire warnings based on the risk level. The prediction and early warning module is used to fuse all sensor data with a three-dimensional combustible environment model to obtain a three-dimensional fire situation model when the risk level is medium. By acquiring meteorological data for future periods, it predicts the spread of fire based on a machine learning model, calculates the fire risk index, and issues wildfire warnings based on the fire risk index.