Utilizing Improved YOLOv8 to Identify Soil Erosion Based on Aerial Imagery

Traditional soil erosion monitoring methods are accurate but limited by efficiency, scalability, and environmental factors, making them challenging for widespread or complex terrain applications. Unmanned aerial vehicle remote sensing, combined with deep learning models, enables cost-effective, high-resolution, and efficient detection of soil erosion distribution and intensity in small to medium-sized areas. In this study, an improved EBP-YOLOv8s-seg model is proposed for object detection and instance segmentation of aerial images. The experimental results show significant performance gains in precision (89.5%), recall (90.9%), and mAP metrics (93.8% at mAP50), demonstrating the model's capability to effectively handle complex aerial imagery scenarios for environmental monitoring while maintaining practical real-time performance (66 FPS).