Land-Cover Classification Technology Based on Deep Learning Algorithms for Predicting Fire Spread in Wildland-Urban Interface

High-precision wildland and urban feature segmentation in the Wildland-Urban Interface (WUI) is crucial for the accurate prediction of fire spread in these areas according to different fuel condition. This study proposes an object recognition algorithm suitable for predicting fire spread in the WUI area based on remote sensing images. The algorithm utilizes an improved YOLOv8-based semantic segmentation model, trained and tested on the Gaofen Image Dataset. It demonstrates good segmentation performance on the Gaofen-2 remote sensing images and is practically applied to the Foshan city dataset. Experimental results indicate that this method improves the accuracy of semantic segmentation in remote sensing images, providing a solid foundation for constructing multi-model fire spread predictions in the WUI.