Forest fire detection and early warning system and method based on multi-modal fusion
Through the multimodal fusion forest fire detection method, the full-dimensional data analysis model and pre-fire abnormal feature library are used to achieve timely and accurate detection and early warning of forest fires, and the problems of small monitoring range and low efficiency in traditional methods are solved.
Patent Information
- Application Number
- CN202510614067.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-13
- Publication Date
- 2025-08-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional forest fire detection methods have small monitoring range and low efficiency, and cannot predict forest fires in a timely and accurate manner and issue alarms.
By extracting full-dimensional data of the forest environment, including physical environment, gas composition and vegetation physiological data, a multimodal fusion data analysis model is established, and a database of pre-fire abnormal characteristics is constructed in real time to trigger an alarm.
It improves the accuracy of forest fire detection and the rapidity of early warning, can obtain forest environment changes from multiple angles, eliminate seasonal interference, and improves the accuracy of early warning models.
Smart Images
Figure CN120526518A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to computer science and technology and forest fire prevention, and in particular to a forest fire detection and early warning system and method based on multimodal fusion. Background Art
[0002] Forest fires can cause serious damage to the ecological environment, economy, and society. They destroy vast swaths of forest resources, reduce biodiversity, release large amounts of greenhouse gases like carbon dioxide, impact the global climate, and threaten the lives and property of surrounding residents. Therefore, timely and accurate forest fire detection and early warning are crucial. However, traditional forest fire detection methods are limited by their small monitoring range and low efficiency, making it difficult to predict forest fires and issue timely warnings. Summary of the Invention
[0003] The purpose of the present invention is to provide a forest fire detection and early warning system and method based on multimodal fusion to solve the problems raised in the prior art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a forest fire detection and early warning method based on multimodal fusion, the monitoring and early warning method comprising the following steps: Step S1: extracting historical full-dimensional data of the forest environment, wherein the historical full-dimensional data includes physical environment data, gas composition data, and vegetation physiological data; classifying and storing the full-dimensional data according to natural seasonal states to form a spatiotemporal data set; The physical environment data includes temperature, humidity, wind speed and direction, light intensity and soil moisture content; the gas composition data includes oxygen concentration, carbon dioxide concentration and carbon monoxide concentration; the vegetation physiological data includes chlorophyll content, transpiration rate and trunk sap flow changes; the natural seasonal state is divided into four seasons: spring, summer, autumn and winter.
[0005] Step S2: Establish a data analysis model, analyze the spatiotemporal data set, calculate the variance of the data, and establish a set of historical forest environment curves in different natural seasonal states; The data analysis model is established to classify the data in the spatiotemporal data set into different types, forming independent data sets for the same type of data, using the same type of data sets as input data for the data analysis model, and calculating and outputting the data curve of the data type; the calculation formula is:
[0006] s is the sample standard deviation, n is the number of sample data, x i is the i-th data point, is the average value of the overall data; The historical forest environment curve set is established based on the degree of correlation between the curves to form a historical forest environment curve set; the calculation formula is:
[0007] where x i and y i There are two different types of data sets. is x i The corresponding mean, It is y i The corresponding mean, n is the number of data in the data set.
[0008] Step S3: Collect full-dimensional data of the forest environment in real time to form a real-time data set. Input the real-time data set into the data analysis model to establish a real-time forest environment curve set. Obtain a historical forest environment curve set corresponding to the same historical natural seasonal state as the current natural seasonal state for comparison. Mark time segments with different curve slopes as abnormal in the real-time forest environment curve set to form an abnormal database. Collect full-dimensional forest environment data in real time at a fixed frequency; when establishing a real-time forest environment curve set, perform continuity correction on discrete real-time data; compare the real-time forest environment curve set with the historical forest environment curve set under the same natural season status; if the slope of a real-time forest environment curve is different from the slope of the historical forest environment curve under the same natural season status, mark the corresponding segment in the real-time forest environment curve set as an anomaly to form an anomaly database.
[0009] Step S4: extract historical full-dimensional data before the occurrence of historical fires, construct a set of historical forest environment curves before the fire, extract data features from the historical forest environment curves before the fire in different natural seasons, and construct a library of abnormal features before the fire in different natural seasons; When extracting the data features of the historical forest environment curve, the historical full-dimensional data before the historical fire occurred is subjected to dimensionality reduction processing, and the main data features are extracted by linearly transforming the variance of the past original data; a timestamp is added to each data feature, and the data are classified and stored according to the natural seasonal state, forming a pre-fire abnormal feature library for different natural seasonal states.
[0010] Step S5: extract data features from the anomaly database and compare them with the anomaly feature database before the fire in the same natural season state. If the similarity of the data features of the two exceeds a threshold, an alarm is issued.
[0011] When extracting data features from the anomaly database, the data features from the anomaly database are compared with the pre-fire anomaly feature library under the same natural seasonal conditions. If the calculated similarity exceeds a threshold, an alarm is triggered. The formula for calculating similarity is:
[0012] Among them, X is the feature vector of the abnormal database, Y is the feature vector of the abnormal feature database before the fire, and Sim(X,Y) is the similarity between X and Y.
[0013] Furthermore, a forest fire detection and early warning method based on multimodal fusion is characterized in that: the monitoring and early warning system includes: a data acquisition and classification module, a data analysis and modeling module, a real-time data comparison and marking module, a pre-fire feature construction module and a feature comparison and alarm module; The data collection and classification module is used to extract the full-dimensional historical data of the forest environment and store it according to the natural seasonal state to form a spatiotemporal data set; the data analysis and modeling module is used to establish a data analysis model to analyze the spatiotemporal data set and establish a set of historical forest environment curves in different natural seasonal states; the real-time data comparison and marking module is used to collect the full-dimensional data of the forest environment in real time, compare it with historical data and mark anomalies to form an anomaly database; the pre-fire feature construction module is used to extract historical pre-fire data to construct a pre-fire anomaly feature library in different natural seasonal states; the feature comparison alarm module is used to extract data features from the anomaly database and compare them with the pre-fire anomaly feature library, and trigger an alarm if the threshold is exceeded; The output end of the data acquisition and classification module is connected to the input end of the data analysis and modeling module; the output end of the data analysis and modeling module is connected to the input end of the real-time data comparison and marking module; the output end of the real-time data comparison and marking module is connected to the input end of the feature comparison alarm module; the output end of the pre-fire feature construction module is connected to the input end of the feature comparison alarm module.
[0014] The data collection and classification module includes a full-dimensional data collection unit and a seasonal classification storage unit; The full-dimensional data acquisition unit is used to extract full-dimensional data on the physical and gaseous components of the forest environment and the physiological characteristics of vegetation; the seasonal classification storage unit is used to classify and store the collected full-dimensional data according to the natural seasonal state to construct a spatiotemporal dataset.
[0015] The data analysis and modeling module includes a data type division unit and a curve association construction unit; the real-time data comparison and marking module includes a real-time data collection unit and an abnormal comparison and marking unit; The data type classification unit is used to classify the data in the spatiotemporal dataset to form an independent dataset as the model input; the curve association construction unit is used to output the data curve according to the input data, build an association network and form a set of historical forest environment curves; the real-time data collection unit is used to collect full-dimensional data at a fixed frequency and correct and establish a real-time forest environment curve; the anomaly comparison and marking unit is used to compare the real-time forest environment curve with the historical curve, mark the anomaly and form an anomaly database.
[0016] The pre-fire feature construction module includes a historical data dimension reduction unit and a feature classification storage unit; the feature comparison alarm module includes an abnormal feature extraction unit and a similarity alarm triggering unit; The historical data dimensionality reduction unit is used to reduce the dimensionality of the full-dimensional data before the historical fire and extract the main data features; the feature classification storage unit is used to add timestamps to the data features and classify and store them according to the natural seasonal status to build a feature library; the abnormal feature extraction unit is used to extract the data features of the abnormal database; the similarity alarm triggering unit is used to compare the abnormal database features with the pre-fire feature library, and trigger an alarm when the similarity exceeds the threshold.
[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention realizes multimodal data fusion by extracting full-dimensional data of the forest environment, including physical environment, gas composition and vegetation physiology. It can obtain forest environment changes from multiple angles and improve the accuracy of forest fire detection and early warning.
[0018] 2. The present invention constructs an abnormal feature library of different natural seasonal conditions by extracting historical pre-fire data, and performs pattern matching based on historical fire laws. It can quickly locate data patterns that meet fire precursors, thereby improving the accuracy of early warning.
[0019] 3. The present invention processes historical data by classifying them according to natural seasonal conditions, establishes a set of forest environment curves exclusive to each season, eliminates interference from seasonal periodic fluctuations, and improves the accuracy of the early warning model. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 This is a flow chart of a forest fire detection and early warning method based on multimodal fusion according to the present invention; Figure 2 This is a structural diagram of a forest fire detection and early warning system based on multimodal fusion in the present invention. DETAILED DESCRIPTION
[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0022] Example 1: Figure 1 As shown, the present invention provides a technical solution, a forest fire detection and early warning method based on multimodal fusion, the monitoring and early warning method includes the following steps: Step S1: extracting historical full-dimensional data of the forest environment, wherein the historical full-dimensional data includes physical environment data, gas composition data, and vegetation physiological data; classifying and storing the full-dimensional data according to natural seasonal states to form a spatiotemporal data set; The physical environment data include temperature, humidity, wind speed and direction, light intensity and soil moisture content; the gas composition data include oxygen concentration, carbon dioxide concentration and carbon monoxide concentration; the vegetation physiological data include chlorophyll content, transpiration rate and trunk sap flow changes; Step S2: Establish a data analysis model, analyze the spatiotemporal data set, calculate the variance of the data, and establish a set of historical forest environment curves in different natural seasonal states; The data analysis model is established to classify the data in the spatiotemporal data set into different types, forming independent data sets for the same type of data, using the same type of data sets as input data for the data analysis model, and calculating and outputting the data curve of the data type; the calculation formula is:
[0023] s is the sample standard deviation, n is the number of sample data, x i is the i-th data point, is the average value of the overall data; The historical forest environment curve set is established based on the degree of correlation between the curves to form a historical forest environment curve set; the calculation formula is:
[0024] where x i and y i There are two different types of data sets. is x i The corresponding mean, It is y i The corresponding mean, n is the number of data in the data set.
[0025] Step S3: Collect full-dimensional data of the forest environment in real time to form a real-time data set. Input the real-time data set into the data analysis model to establish a real-time forest environment curve set. Obtain a historical forest environment curve set corresponding to the same historical natural seasonal state as the current natural seasonal state for comparison. Mark time segments with different curve slopes as abnormal in the real-time forest environment curve set to form an abnormal database. Collect full-dimensional forest environment data in real time at a fixed frequency; when establishing a real-time forest environment curve set, perform continuity correction on discrete real-time data; compare the real-time forest environment curve set with the historical forest environment curve set under the same natural season status; if the slope of a real-time forest environment curve is different from the slope of the historical forest environment curve under the same natural season status, mark the corresponding segment in the real-time forest environment curve set as an anomaly to form an anomaly database.
[0026] Step S4: extract historical full-dimensional data before the occurrence of historical fires, construct a set of historical forest environment curves before the fire, extract data features from the historical forest environment curves before the fire in different natural seasons, and construct a library of abnormal features before the fire in different natural seasons; When extracting the data features of the historical forest environment curve, the historical full-dimensional data before the historical fire occurred is subjected to dimensionality reduction processing, and the main data features are extracted by linearly transforming the variance of the past original data; a timestamp is added to each data feature, and the data are classified and stored according to the natural seasonal state, forming a pre-fire abnormal feature library for different natural seasonal states.
[0027] Step S5: extract data features from the anomaly database and compare them with the pre-fire anomaly feature database under the same natural seasonal conditions. If the similarity of the data features of the two exceeds a threshold, an alarm is issued.
[0028] When extracting data features from the anomaly database, the data features from the anomaly database are compared with the pre-fire anomaly feature library under the same natural seasonal conditions. If the calculated similarity exceeds a threshold, an alarm is triggered. The formula for calculating similarity is:
[0029] Among them, X is the feature vector of the abnormal database, Y is the feature vector of the abnormal feature database before the fire, and Sim(X,Y) is the similarity between X and Y.
[0030] For example, in large forests, various sensors are deployed in protected forest areas to collect full-dimensional data of the forest environment over a long period of time, recording daily average temperatures between 25°C and 30°C, average humidity between 55% and 67%, and light intensity between 500 and 1000 μmol·m -2 ·s -1 Soil moisture content remained within the range of 25%-35%. Regarding gas composition data, oxygen concentrations remained stable at 25.5%-26%, carbon dioxide concentrations were 320-420 ppm, and carbon monoxide concentrations were usually below 10 ppm. Vegetation physiological data showed chlorophyll content between 40-60 SPAD values, and transpiration rates of 2-4 mmol·m -2 ·s -1, trunk sap flow varies between 0.1 and 0.3 L / h. This data is categorized and stored according to summer natural seasonal conditions, forming a spatiotemporal dataset. The data analysis and modeling module processes this data and, based on the association network, establishes a set of historical forest environmental curves. At 1:00 PM on a particular day, the real-time temperature recorded was 33°C and the humidity was 60%. After continuity correction, these data were compared with the set of historical summer forest environmental curves. The slope of the temperature curve at this time was significantly higher than that of the historical curves. This time segment was marked as an anomaly and stored in the anomaly database. The pre-fire feature construction module extracts historical pre-fire data to construct an anomaly feature library. The anomaly database is compared with the pre-fire anomaly feature library and, using a similarity calculation formula, the resulting similarity is 0.8, exceeding the pre-set threshold of 0.7. The system immediately issues a fire warning, notifying forest rangers to promptly investigate fire hazards, thereby effectively ensuring forest safety.
[0031] Example 2, as Figure 2 As shown, the present invention provides a forest fire detection and early warning system based on multimodal fusion, which includes: a data acquisition and classification module, a data analysis and modeling module, a real-time data comparison and marking module, a pre-fire feature construction module and a feature comparison and alarm module; The data collection and classification module is used to extract the full-dimensional historical data of the forest environment and store it according to the natural seasonal state to form a spatiotemporal data set; the data analysis and modeling module is used to establish a data analysis model to analyze the spatiotemporal data set and establish a set of historical forest environment curves in different natural seasonal states; the real-time data comparison and marking module is used to collect the full-dimensional data of the forest environment in real time, compare it with historical data and mark anomalies to form an anomaly database; the pre-fire feature construction module is used to extract historical pre-fire data to construct a pre-fire anomaly feature library in different natural seasonal states; the feature comparison alarm module is used to extract data features from the anomaly database and compare them with the pre-fire anomaly feature library, and trigger an alarm if the threshold is exceeded; The output end of the data acquisition and classification module is connected to the input end of the data analysis and modeling module; the output end of the data analysis and modeling module is connected to the input end of the real-time data comparison and marking module; the output end of the real-time data comparison and marking module is connected to the input end of the feature comparison alarm module; the output end of the pre-fire feature construction module is connected to the input end of the feature comparison alarm module.
[0032] The data collection and classification module includes a full-dimensional data collection unit and a seasonal classification storage unit; The full-dimensional data acquisition unit is used to extract full-dimensional data on the physical and gaseous components of the forest environment and the physiological characteristics of vegetation; the seasonal classification storage unit is used to classify and store the collected full-dimensional data according to the natural seasonal state to construct a spatiotemporal dataset.
[0033] The data analysis and modeling module includes a data type division unit and a curve association construction unit; the real-time data comparison and marking module includes a real-time data collection unit and an abnormal comparison and marking unit; The data type classification unit is used to classify the data in the spatiotemporal dataset to form an independent dataset as the model input; the curve association construction unit is used to output the data curve according to the input data, build an association network and form a set of historical forest environment curves; the real-time data collection unit is used to collect full-dimensional data at a fixed frequency and correct and establish a real-time forest environment curve; the anomaly comparison and marking unit is used to compare the real-time forest environment curve with the historical curve, mark the anomaly and form an anomaly database.
[0034] The pre-fire feature construction module includes a historical data dimension reduction unit and a feature classification storage unit; the feature comparison alarm module includes an abnormal feature extraction unit and a similarity alarm triggering unit; The historical data dimensionality reduction unit is used to reduce the dimensionality of the full-dimensional data before the historical fire and extract the main data features; the feature classification storage unit is used to add timestamps to the data features and classify and store them according to the natural seasonal status to build a feature library; the abnormal feature extraction unit is used to extract the data features of the abnormal database; the similarity alarm triggering unit is used to compare the abnormal database features with the pre-fire feature library, and trigger an alarm when the similarity exceeds the threshold.
[0035] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.
Claims
1. A forest fire detection and early warning method based on multimodal fusion, characterized by: The monitoring and early warning method comprises the following steps: Step S1: extracting historical full-dimensional data of the forest environment, wherein the historical full-dimensional data includes physical environment data, gas composition data, and vegetation physiological data; classifying and storing the full-dimensional data according to natural seasonal states to form a spatiotemporal data set; Step S2: Establish a data analysis model, analyze the spatiotemporal data set, calculate the variance of the data, and establish a set of historical forest environment curves in different natural seasonal states; Step S3: Collect full-dimensional data of the forest environment in real time to form a real-time data set. Input the real-time data set into the data analysis model to establish a real-time forest environment curve set. Obtain a historical forest environment curve set corresponding to the same historical natural seasonal state as the current natural seasonal state for comparison. Mark time segments with different curve slopes as abnormal in the real-time forest environment curve set to form an abnormal database. Step S4: extract historical full-dimensional data before the occurrence of historical fires, construct a set of historical forest environment curves before the fire, extract data features from the historical forest environment curves before the fire in different natural seasons, and construct a library of abnormal features before the fire in different natural seasons; Step S5: extract data features from the anomaly database and compare them with the anomaly feature database before the fire in the same natural season state. If the similarity of the data features of the two exceeds a threshold, an alarm is issued.
2. The forest fire detection and early warning method based on multimodal fusion according to claim 1, characterized in that: The specific steps of step S1 are as follows: the physical environment data include temperature, humidity, wind speed and direction, light intensity and soil moisture content; the gas composition data include oxygen concentration, carbon dioxide concentration and carbon monoxide concentration; the vegetation physiological data include chlorophyll content, transpiration rate and trunk sap flow changes.
3. The forest fire detection and early warning method based on multimodal fusion according to claim 2, characterized in that: The specific steps of step S2 are as follows: the data analysis model is established, the data in the spatiotemporal data set is divided into types, the same type of data is formed into an independent data set, the same type of data set is used as the input data of the data analysis model, and the data curve of the type of data is calculated and output; the calculation formula is: s is the sample standard deviation, n is the number of sample data, x i is the i-th data point, is the average value of the overall data; The historical forest environment curve set is established based on the degree of correlation between the curves to form a historical forest environment curve set; the calculation formula is: where x i and y i There are two different types of data sets. is x i The corresponding mean, It is y i The corresponding mean, n is the number of data in the data set.
4. The forest fire detection and early warning method based on multimodal fusion according to claim 3 is characterized by: The specific steps of step S3 are as follows: collecting full-dimensional forest environment data in real time at a fixed frequency; performing continuity correction on discrete real-time data when establishing a real-time forest environment curve set; comparing the real-time forest environment curve set with a historical forest environment curve set under the same natural season state; if the slope of a real-time forest environment curve is different from the slope of the historical forest environment curve under the same natural season state, marking the corresponding segment in the real-time forest environment curve set as an abnormality to form an abnormality database.
5. The forest fire detection and early warning method based on multimodal fusion according to claim 4 is characterized in that: The specific steps of step S4 are as follows: receiving the anomaly database, extracting the data features of the historical forest environment curve, performing dimensionality reduction processing on the historical full-dimensional data before the historical fire, and extracting the main data features by linearly transforming the variance of the past original data; adding a timestamp to each data feature, classifying and storing them according to the natural seasonal state, and forming a pre-fire anomaly feature library for different natural seasonal states.
6. The forest fire detection and early warning method based on multimodal fusion according to claim 5, characterized in that: The specific steps of step S5 are as follows: when extracting data features from the abnormal database, the data features of the abnormal database are compared with the abnormal feature library before the fire in the same natural season state. If the calculated similarity exceeds the threshold, an alarm is triggered; the formula for calculating the similarity is: Among them, X is the feature vector of the abnormal database, Y is the feature vector of the abnormal feature database before the fire, and Sim(X,Y) is the similarity between X and Y.
7. A forest fire detection and early warning system based on multimodal fusion, applied to the forest fire detection and early warning method based on multimodal fusion according to any one of claims 1 to 6, characterized in that: The monitoring and early warning system includes: a data acquisition and classification module, a data analysis and modeling module, a real-time data comparison and marking module, a pre-fire feature construction module and a feature comparison and alarm module; The data acquisition and classification module is used to extract the full-dimensional historical data of the forest environment and store it according to the natural seasonal state to form a spatiotemporal data set; the data analysis and modeling module is used to establish a data analysis model to analyze the spatiotemporal data set and establish a set of historical forest environment curves in different natural seasonal states; the real-time data comparison and marking module is used to collect the full-dimensional data of the forest environment in real time, compare it with the historical data and mark anomalies to form an anomaly database; the pre-fire feature construction module is used to extract the historical pre-fire data to construct a pre-fire anomaly feature library in different natural seasonal states; the feature comparison alarm module is used to extract the data features of the anomaly database and compare them with the pre-fire anomaly feature library, and trigger an alarm if the threshold is exceeded; The output end of the data acquisition and classification module is connected to the input end of the data analysis and modeling module; the output end of the data analysis and modeling module is connected to the input end of the real-time data comparison and marking module; the output end of the real-time data comparison and marking module is connected to the input end of the feature comparison alarm module; the output end of the pre-fire feature construction module is connected to the input end of the feature comparison alarm module.
8. The forest fire detection and early warning system based on multimodal fusion according to claim 7, characterized in that: The data acquisition and classification module includes a full-dimensional data acquisition unit and a seasonal classification storage unit; The full-dimensional data acquisition unit is used to extract full-dimensional data on the physical, gaseous composition and vegetation physiology of the forest environment; the physical environment data includes temperature, humidity, wind speed and direction, light intensity and soil moisture content; the gas composition data includes oxygen concentration, carbon dioxide concentration and carbon monoxide concentration; the vegetation physiological data includes chlorophyll content, transpiration rate and trunk sap flow changes; the seasonal classification storage unit is used to classify and store the collected full-dimensional data according to natural seasonal status to construct a spatiotemporal dataset.
9. The forest fire detection and early warning system based on multimodal fusion according to claim 7, characterized in that: The data analysis and modeling module includes a data type division unit and a curve association construction unit; the real-time data comparison and marking module includes a real-time data collection unit and an abnormal comparison and marking unit; The data type classification unit is used to classify the data in the spatiotemporal data set into independent data sets as model input; the curve association construction unit is used to output data curves based on the input data, construct an association network and form a set of historical forest environment curves; the real-time data collection unit is used to collect full-dimensional data at a fixed frequency and correct and establish a real-time forest environment curve; the anomaly comparison and marking unit is used to compare the real-time forest environment curve with the historical curve, mark anomalies and form an anomaly database.
10. The forest fire detection and early warning system based on multimodal fusion according to claim 7, characterized in that: The pre-fire feature construction module includes a historical data dimension reduction unit and a feature classification storage unit; the feature comparison alarm module includes an abnormal feature extraction unit and a similarity alarm triggering unit; The historical data dimensionality reduction unit is used to perform dimensionality reduction processing on the full-dimensional data before the historical fire occurred to extract the main data features; the feature classification storage unit is used to add timestamps to the data features and classify and store them according to the natural seasonal state to build a feature library; the abnormal feature extraction unit is used to extract the data features of the abnormal database; the similarity alarm triggering unit is used to compare the abnormal database features with the pre-fire feature library, and trigger an alarm if the similarity exceeds the threshold.