CN-120875091-B - Forest fire temperature downscaling method and system based on machine learning and energy conservation constraint

CN120875091BCN 120875091 BCN120875091 BCN 120875091BCN-120875091-B

Abstract

The invention discloses a forest fire temperature downscaling method and a system based on machine learning and energy conservation constraint, comprising the following steps of firstly, data preparation and pretreatment; and step three, correcting the temperature downscaling result based on the energy conservation constraint. According to the invention, the light temperature of the sensor end with the spatial resolution of 50m is obtained by utilizing a XGBoost machine learning model, then the light temperature of the sensor end with the spatial resolution of 50m is corrected by calculating an energy conservation factor, the problem of distortion of a temperature scale reduction result when an energy conservation principle is not considered is weakened, and the scale reduction precision of the forest fire temperature is effectively improved.

Inventors

YE JIANGXIA
XU MENG
WANG YANXIA
ZHANG GUI
YANG YONGKE
KONG LEI

Assignees

西南林业大学
中南林业科技大学

Dates

Publication Date: 20260508
Application Date: 20250810

Claims (7)

1. A forest fire temperature downscaling method based on machine learning and energy conservation constraint is characterized by comprising the following steps: Step one, preparing remote sensing images and auxiliary data: collecting data of a forest fire scene, wherein the data of the forest fire scene comprises optical wave band data, near infrared wave band data and middle infrared wave band data of a satellite image, and slope direction and gradient data of the forest fire scene, wherein the spatial resolution of the near infrared wave band data is H m, the spatial resolution of the middle infrared wave band data is L m, and L > H Resampling the data of a forest fire scene to obtain two sets of grid image data with spatial resolutions of H m and L m respectively, and calculating to obtain the sensor end bright temperature with the spatial resolution of L m according to the middle infrared band data, wherein the grid image data comprises seven effective factors including altitude, gradient, slope direction, normalized vegetation index NDVI, earth surface coverage type, GF-4 satellite red light band reflectivity and near infrared band reflectivity according to the sensor end bright temperature of L m; Step two, temperature downscaling based on a machine learning algorithm: According to the grid image data with the spatial resolution of L m and the corresponding sensor end bright temperature with the spatial resolution of L m, a training set is constructed, a XGBoost temperature downscaling model is constructed, the training set is input into the XGBoost temperature downscaling model for training to obtain a trained XGBoost temperature downscaling model, the grid image data with the spatial resolution of H m is input into the trained XGBoost temperature downscaling model to obtain each pixel sensor end bright temperature with the spatial resolution of H m; step three, correcting a temperature downscaling result based on energy conservation constraint: Correcting the bright temperature of each pixel sensor end with the spatial resolution of H m based on energy conservation constraint to obtain corrected bright temperature of each pixel sensor end with the H m spatial resolution; ; as a constraint factor for conservation of energy, To the corrected first The temperature of each pixel is calculated, The pixel number of the raster image data with the spatial resolution of H m; To be the first before correction The temperature of each pixel.
2. A forest fire temperature downscaling method based on machine learning and energy conservation constraints as claimed in claim 1, wherein h=50, l=400.
3. The method for reducing the temperature of a forest fire based on machine learning and energy conservation constraints according to claim 1, wherein the specific steps of the first step are as follows: The method comprises the steps of 1-1, obtaining GF-4 satellite remote sensing images of a forest fire scene, obtaining radiation brightness values after radiation calibration of a middle infrared band, obtaining reflectivity data of an optical band and reflectivity data of a near infrared band through atmospheric correction, obtaining normalized vegetation index NDVI according to the reflectivity data of a red light band and the reflectivity data of a near infrared band of the optical band, performing supervision classification by using a support vector machine algorithm image with the reflectivity data of the optical band, the reflectivity data of the near infrared band and the normalized vegetation index NDVI as inputs, downloading a digital elevation model DEM with 90 m spatial resolution, obtaining slope and gradient data of the forest fire scene through calculation in ArcGIS software, and resampling the reflectivity data of the optical band and the reflectivity data of the near infrared band, the normalized vegetation index NDVI, the surface coverage type and the slope and gradient data of the forest fire scene to obtain two sets of grid image data with spatial resolutions of 400 m and 50m respectively; step 1-2, calculating the bright temperature of a sensor end: infrared band radiation brightness value using GF-4 satellite And the Planck radiation law to calculate the brightness temperature of each pixel at the sensor end with the spatial resolution of 400 m : ; The radiation brightness value of the infrared band in the GF-4 satellite is 150158.90 and 3785.45 for K 1 and K 2 respectively; And 1-3, carrying out fire point identification by adopting a self-adaptive threshold method based on the sensor end brightness temperature data with the spatial resolution of 400 m, taking the fire point as a seed point, and capturing the high Wen Xiangyuan around the fire point by utilizing an area growth algorithm to form a closed fire field range.
4. The method for reducing the temperature of a forest fire based on machine learning and energy conservation constraint according to claim 1, wherein the auxiliary factor screening method in the step one is as follows: The method comprises the steps of taking a sensor end brightness temperature and auxiliary data set with 400m spatial resolution as a basis, evaluating the correlation degree of each factor and the sensor end brightness temperature through calculating a Pearson correlation coefficient to obtain a strong correlation factor, detecting multiple collinearity among each factor through calculating a variance expansion factor, removing multiple collinearity variables to finally obtain 7 effective factors including an altitude, a gradient, a slope direction, a normalized vegetation index NDVI, an earth surface coverage type, a GF-4 satellite red light wave band reflectivity and a near infrared wave band reflectivity, wherein the auxiliary data set comprises optical wave band data, near infrared wave band data, middle infrared wave band data of GF-4 satellite images, slope direction and slope data of a forest fire field, a normalized vegetation index NDVI and an earth surface coverage type of GF-4 satellite images.
5. The method for reducing the forest fire temperature based on machine learning and energy conservation constraint of claim 1, wherein in the second step, auxiliary data with spatial resolution of 400 m is taken as an independent variable, corresponding sensor end brightness temperature data is taken as a dependent variable, a sample set is obtained through spatial registration and extraction, and the sample set is divided into a training set and a test set.
6. The method for downscaling the temperature of a forest fire based on machine learning and energy conservation constraint according to claim 1, wherein in the third step, the energy conservation constraint factor The calculation method of (2) is as follows: ; Wherein, the , Indicating the sensor tip light temperature (K), subscripts L, H represent low and high resolution respectively, To the power of 4 for the bright temperature of the low resolution pixel sensor end, To the power of 4 for the bright temperature of the i-th high-resolution pixel sensor end before correction, 。
7. The forest fire temperature downscaling system based on machine learning and energy conservation constraint is characterized in that the forest fire satellite remote sensing image reconstruction system is used for running the forest fire temperature downscaling method based on machine learning and energy conservation constraint according to any one of claims 1-5.

Description

Forest fire temperature downscaling method and system based on machine learning and energy conservation constraint Technical Field The invention belongs to the field of satellite remote sensing images, and particularly relates to a forest fire temperature downscaling method and system based on machine learning and energy conservation constraint. Background With the global warming and the aggravation of extreme drought events, the occurrence frequency and intensity of forest fires continuously rise, and the forest fires become a great natural disaster which threatens the safety of an ecological system and the sustainable development of forestry resources. The satellite remote sensing technology is used for acquiring forest fire temperature information, and is important for carrying out fire monitoring and accurate suppression. At present, the mid-infrared and thermal infrared remote sensing data are widely applied in the forest fire monitoring field. However, the spatial resolution of mid-infrared and thermal infrared sensors is generally low, making it difficult to detect complex changes in the internal temperature of a fire scene. The downscaling algorithm is a key technology for improving the spatial resolution of remote sensing data, is widely applied to the field of temperature downscaling, and the most commonly used temperature downscaling method comprises temperature downscaling based on an image fusion method and temperature downscaling based on a machine learning algorithm. The principle of ground surface temperature downscaling based on the image fusion method is to fuse the abundant space detail information of the high-spatial-resolution optical data with the thermal information of the low-spatial-resolution temperature data, so as to generate a temperature product with high spatial resolution and high thermal information precision. However, the temperature downscaling algorithm based on the image fusion method is suitable for scenes with small temperature space change and insignificant boundary effects, and the application effect in the forest fire temperature downscaling with severe temperature change and significant boundary effects is not ideal. The principle of ground surface temperature downscaling based on the machine learning algorithm is that the complex nonlinear mapping relation from the low-spatial-resolution temperature data and the high-spatial-resolution auxiliary data to the high-spatial-resolution temperature data is automatically learned through the machine learning algorithm, and the high-spatial-resolution temperature is estimated by combining the high-spatial-resolution auxiliary data. However, the automatic learning process does not consider the conservation rule of radiant energy, which is to be observed by the spatial resolution conversion, namely the sum of radiant energy of the pixels with high spatial resolution after downscaling is equal to the radiant energy of the corresponding pixels with low spatial resolution, so that the problem of result distortion still exists in the downscaling of the temperature based on the machine learning algorithm. Therefore, the temperature downscaling algorithm based on the image fusion method is suitable for scenes with small temperature space change and insignificant boundary effect, but the application effect in the forest fire temperature downscaling with severe temperature change and significant boundary effect is not ideal. The temperature downscaling based on the machine learning algorithm can establish a complex nonlinear mapping relation through automatic learning so as to improve the accuracy of the temperature downscaling, but the problem of conservation of radiant energy during spatial resolution conversion is not considered in the automatic learning process, so that the temperature data after downscaling still has a distortion problem. In summary, the existing temperature downscaling algorithm cannot be directly used for forest fire temperature downscaling. Therefore, according to the temperature characteristics of the forest fire scene, a temperature downscaling method suitable for the forest fire scene is designed, so that fine temperature information in the fire scene is obtained, and the problems of forest fire monitoring and accurate extinguishment are still needed to be solved. Disclosure of Invention In order to solve the problems, the invention discloses a forest fire temperature downscaling method and system based on machine learning and energy conservation constraint. In order to achieve the above purpose, the technical scheme of the invention is as follows: A forest fire temperature downscaling method based on machine learning and energy conservation constraint comprises the following steps: Step one, preparing remote sensing images and auxiliary data: collecting data of a forest fire scene, wherein the data of the forest fire scene comprises optical wave band data, near infrared wave band data and middle infrared