Remote sensing detection of forest fires by integrating remote sensing and spatial relationship features

By comprehensively analyzing the effects of short-wave infrared bands, terrain correction, and spatial relationship features on burned area recognition, a forest fire classification method that integrates multi-source features is proposed. The terrain correction method and the application of short-wave infrared bands as feature inputs are used to reduce the effects of mountain shadows and smoke on burned area recognition, and spatial relationship features are introduced to correct the burn area recognition results. The forest fire in Chongqing Municipality in 2022 is studied based on Sentinel 2 images and support vector machine model. The study shows that: 1) the application of short-wave infrared bands can reduce the interference of smoke on the recognition of burned areas, and the recognition accuracy will be improved by 3%~8%; 2) terrain correction can reduce the interference of the shadow of the mountain, but it has little effect on the recognition accuracy of the burned areas; 3) the introduction of spatial relationship features can significantly reduce the interference of the bare ground, hard paved ground, etc., and the accuracy will be improved by 3%~4%. The physical mechanism of this method is clear and can effectively extract forest fires (especially burning forest fires).