Forest fire point detection method based on KAN network and multi-sensor fusion

Abstract

The invention discloses a forest fire point detection method based on a KAN network and multi-sensor fusion. The forest fire point detection method comprises the following steps: S10, configuring a plurality of sensors in fire point detection equipment for data acquisition; s20, transmitting the acquired data to a multi-sensor fusion model based on a KAN network for data fusion, and fusing the data into one piece of data as decision data for forest fire point detection; and S30, performing fire point generation early warning processing by using the decision data of forest fire point detection. According to the method, the detection of the fire points in a large forest can be more accurately realized.

Description

Technical Field

[0001] The invention belongs to the technical field of forest fire monitoring, and in particular relates to a forest fire point detection method based on a KAN network and multi-sensor fusion. Background Art

[0002] Forest fire is a natural disaster that is sudden, destructive, and difficult to deal with and rescue. Forest fire not only causes the loss of some trees, but also damages the forest ecological environment, and causes problems such as air pollution and damage to people's property. Therefore, early warning of forest and grassland fires is crucial.

[0003] At present, most of the protection against forest fires is through drone technology, satellite and remote video monitoring, ground patrols and manual observation, which are generally costly. The rapid development of the electronics industry has promoted the research and development of various sensors. For the monitoring of forest fire points, multiple sensors can be used for monitoring, such as smoke sensors, temperature sensors, etc. Using a single sensor may cause false detection, so it is usually possible to try to use multiple sensors for simultaneous monitoring, but the various data brought by multiple sensors are often not well integrated and decisions cannot be made.

[0004] For multi-sensor data fusion, there are usually Kalman fusion, first-order low-pass filtering, etc., but these fusion methods can only be used for limited sensors, and the fusion method is single. Using neural networks for data fusion, the common use of multi-layer perceptron (MLP), but as the amount of data increases, the number of parameters of the MLP model will also increase sharply, resulting in an increase in computing costs, and the MLP model is a black box model, which makes it difficult to explain how data fusion and decision-making are performed internally. Many traditional data fusion algorithms are more based on fixed fusion methods, and may make wrong judgments for some emergencies. In addition, the numerical ranges of different sensors may be different, so traditional fusion algorithms may also need to perform data conversion, which makes it difficult to take into account the characteristics of all sensor data. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a forest fire point detection method based on KAN network and multi-sensor fusion, which can more accurately realize the detection of fire points in large forests.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a forest fire point detection method based on KAN network and multi-sensor fusion, comprising the steps of S10, multiple sensors are configured in the fire point detection equipment to collect data; S20, transmitting the collected data to a multi-sensor fusion model based on the KAN network for data fusion, and fusing the data into one data as decision data for forest fire point detection; S30, utilizing the decision data of the forest fire point detection to perform fire point occurrence early warning processing.

[0007] Furthermore, the process of establishing a multi-sensor fusion model based on the KAN network includes: Construct a sample data set, including data from multiple sensors in different environmental conditions, and corresponding descriptions of forest fire sites; The sensor data is passed into the data fusion KAN network model, and then the original sensor data is refined using the learnable activation function; after multiple data refinements and fusions, the measurement data of multiple sensors are finally fused into one data as the decision data for forest fire point detection.

[0008] Furthermore, constructing a sample data set includes: data under normal environment, dry environment and high temperature environment, or data under normal conditions, suspected fire points, low-concentration smoke and high-concentration smoke conditions; and the corresponding fused data is the probability of fire point occurrence.

[0009] Furthermore, the decision data output by the model is compared with the label corresponding to the fused data, the corresponding error is calculated, and then the error is used to adjust the learnable activation function to fit the input multi-sensor data to the label of the fused data; after multiple adjustments, the error gradually decreases, and the entire model will fit the multi-sensor input data and label data; After adjusting the activation function multiple times, the KAN network model that fuses multi-sensor data is obtained.

[0010] Furthermore, the multi-sensor fusion based on the KAN network includes the following steps: First, for the data obtained from multiple different sensors, each data is passed through three learnable activation functions; Then, the values ​​of the corresponding positions of each sensor data after being calculated by the activation function are accumulated, and the data of multiple sensors are fused into 3 data; After that, two fused data are obtained by calculating two different learnable activation functions on the three fused data respectively; Finally, the learnable activation functions are calculated separately and then fused into one data, which is the final fusion data of multiple sensor data and serves as the decision data for forest fire point detection.

[0011] Furthermore, the method of accumulating the values ​​of the corresponding positions of each sensor data after the activation function calculation and fusing the data of multiple sensors into three data includes: (1) adding the values ​​of the first sensor data corresponding to the first learnable activation function, the values ​​of the second sensor data corresponding to the first learnable activation function, and the values ​​of the first learnable activation function corresponding to the Nth sensor data; (2) adding the values ​​of the first sensor data corresponding to the second learnable activation function, the values ​​of the second sensor data corresponding to the second learnable activation function, and the values ​​of the second learnable activation function corresponding to the Nth sensor data; (3) adding the values ​​of the first sensor data corresponding to the third learnable activation function, the values ​​of the second sensor data corresponding to the third learnable activation function, and the values ​​of the third learnable activation function corresponding to the Nth sensor data.

[0012] The beneficial effects of adopting this technical solution are: The present invention detects forest fires by using multiple sensors to collect data in the current environment to more accurately determine whether there is a fire at the current location. However, the multiple data from multiple sensors cannot be well integrated and decisions cannot be made. Therefore, the present invention implements multi-sensor data fusion based on the KAN network to achieve accurate detection of forest fires.

[0013] The present invention uses the KAN network to fuse the data of multiple sensors, learns the relationship between the multi-sensor data and the fire point situation, and can well supervise the numerical characteristics of different sensor data. In addition, KAN has a strong ability to fit nonlinear functions, so it can easily find the nonlinear mapping relationship between multi-sensor data and fused label data, thereby achieving better fire point detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flowchart of a forest fire point detection method based on KAN network and multi-sensor fusion.

[0015] Figure 2 This is a flow chart of model training in an embodiment of the present invention.

[0016] Figure 3 Schematic diagram of multi-sensor fusion based on KAN network in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described below with reference to the accompanying drawings.

[0018] In this embodiment, see Figure 1As shown, the present invention proposes a forest fire point detection method based on KAN network and multi-sensor fusion, including the steps of S10, multiple sensors are configured in the fire point detection equipment to collect data; S20, transmitting the collected data to a multi-sensor fusion model based on the KAN network for data fusion, and fusing the data into one data as decision data for forest fire point detection; S30, utilizing the decision data of the forest fire point detection to perform fire point occurrence early warning processing.

[0019] As an optimization solution of the above embodiment, Figure 2 As shown in the figure, the process of establishing a multi-sensor fusion model based on the KAN network includes: Construct a sample data set, including data from multiple sensors in different environmental conditions, and corresponding descriptions of forest fire sites; The sensor data is passed into the data fusion KAN network model, and then the original sensor data is refined using the learnable activation function; after multiple data refinements and fusions, the measurement data of multiple sensors are finally fused into one data as the decision data for forest fire point detection.

[0020] Specifically, the sample data set includes: data under normal environment, dry environment and high temperature environment, or data under normal conditions, suspected fire points, low concentration smoke and high concentration smoke; and the corresponding fusion data is the probability of fire point occurrence. Among them, the high temperature environment refers to the temperature close to the ignition point of the plant.

[0021] Specifically, the decision data output by the model is compared with the label corresponding to the fused data, the corresponding error is calculated, and then the error is used to adjust the learnable activation function to fit the input multi-sensor data to the label of the fused data; after multiple adjustments, the error gradually decreases, and the entire model will fit the multi-sensor input data and label data; After adjusting the activation function multiple times, the KAN network model that fuses multi-sensor data is obtained.

[0022] As an optimization solution of the above embodiment, Figure 3 As shown, the multi-sensor fusion based on KAN network includes the following steps: First, for the data obtained from multiple different sensors, each data is passed through three learnable activation functions; Then, the values ​​of the corresponding positions of each sensor data after being calculated by the activation function are accumulated, and the data of multiple sensors are fused into 3 data; Specifically, the method of accumulating the values ​​of the corresponding positions of each sensor data after the activation function calculation and fusing the data of multiple sensors into three data includes: (1) adding the values ​​of the first sensor data corresponding to the first learnable activation function, the values ​​of the second sensor data corresponding to the first learnable activation function, and the values ​​of the first learnable activation function corresponding to the Nth sensor data; (2) adding the values ​​of the first sensor data corresponding to the second learnable activation function, the values ​​of the second sensor data corresponding to the second learnable activation function, and the values ​​of the second learnable activation function corresponding to the Nth sensor data; (3) adding the values ​​of the first sensor data corresponding to the third learnable activation function, the values ​​of the second sensor data corresponding to the third learnable activation function, and the values ​​of the third learnable activation function corresponding to the Nth sensor data.

[0023] Afterwards, two fused data are obtained by calculating two different learnable activation functions on the three fused data.

[0024] Finally, the learnable activation functions are calculated separately and then fused into one data, which is the final fusion data of multiple sensor data and serves as the decision data for forest fire point detection.

[0025] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A forest fire point detection method based on KAN network and multi-sensor fusion, characterized in that: Included steps S10, multiple sensors are configured in the fire point detection equipment to collect data; S20, transmitting the collected data to a multi-sensor fusion model based on the KAN network for data fusion, and fusing the data into one data as decision data for forest fire point detection; S30, utilizing the decision data of the forest fire point detection to perform fire point occurrence early warning processing.

2. According to claim 1, a forest fire point detection method based on KAN network and multi-sensor fusion is characterized in that: The process of establishing a multi-sensor fusion model based on the KAN network includes: Construct a sample data set, including data from multiple sensors in different environmental conditions, and corresponding descriptions of forest fire sites; The sensor data is passed into the data fusion KAN network model, and then the original sensor data is refined using the learnable activation function; after multiple data refinements and fusions, the measurement data of multiple sensors are finally fused into one data as the decision data for forest fire point detection.

3. The forest fire point detection method based on KAN network and multi-sensor fusion according to claim 2 is characterized in that: The sample data set constructed includes: data under normal environment, dry environment and high temperature environment, or data under normal conditions, suspected fire points, low-concentration smoke and high-concentration smoke conditions; and the corresponding fused data is the probability of fire point occurrence.

4. The forest fire point detection method based on KAN network and multi-sensor fusion according to claim 2 is characterized in that: Compare the decision data output by the model with the label of the corresponding fused data, calculate the corresponding error, and then use the error to adjust the learnable activation function to fit the input multi-sensor data to the label of the fused data; after multiple adjustments, the error gradually decreases, and the entire model will fit the multi-sensor input data and label data; After adjusting the activation function multiple times, the KAN network model that fuses multi-sensor data is obtained.

5. The forest fire point detection method based on KAN network and multi-sensor fusion according to claim 4 is characterized in that: Multi-sensor fusion based on KAN network includes the following steps: First, for the data obtained from multiple different sensors, each data is passed through three learnable activation functions; Then, the values ​​of the corresponding positions of each sensor data after being calculated by the activation function are accumulated, and the data of multiple sensors are fused into 3 data; After that, two fused data are obtained by calculating two different learnable activation functions on the three fused data respectively; Finally, the learnable activation functions are calculated separately and then fused into one data, which is the final fusion data of multiple sensor data and serves as the decision data for forest fire point detection.

6. The forest fire point detection method based on KAN network and multi-sensor fusion according to claim 5 is characterized in that: The method of accumulating the values ​​calculated by the activation function at the corresponding position of each sensor data and fusing the data of multiple sensors into three data includes: (1) adding the values ​​processed by the first learnable activation function corresponding to the first sensor data, the values ​​processed by the first learnable activation function corresponding to the second sensor data, and the values ​​processed by the first learnable activation function corresponding to the Nth sensor data; (2) adding the values ​​processed by the second learnable activation function corresponding to the first sensor data, the values ​​processed by the second learnable activation function corresponding to the second sensor data, and the values ​​processed by the second learnable activation function corresponding to the Nth sensor data; and (3) adding the values ​​processed by the third learnable activation function corresponding to the first sensor data, the values ​​processed by the third learnable activation function corresponding to the second sensor data, and the values ​​processed by the third learnable activation function corresponding to the Nth sensor data.

Images