Multiscale Two-Stream Fusion Network for Benggang Classification in Multi-Source Images

Sensors, Journal Year: 2025, Volume and Issue: 25(9), P. 2924 - 2924

Published: May 6, 2025

Benggangs, a type of soil erosion widely distributed in the hilly and mountainous regions South China, pose significant challenges to land management ecological conservation. Accurate identification assessment their location scale are essential for effective Benggang control. With advancements technology, deep learning has emerged as critical tool classification. However, selecting suitable feature extraction fusion methods multi-source image data remains challenge. This study proposes classification method based on multiscale features two-stream network (MS-TSFN). Key targeted areas, such slope, aspect, curvature, hill shade, edge, were extracted from Digital Orthophotography Map (DOM) Surface Model (DSM) collected by drones. The network, with ResNeSt backbone, images an attention-based block was developed explore complementary associations among achieve information across types. A decision employed global prediction classify areas or non-Benggang. Experimental comparisons different inputs models revealed that proposed outperformed current state-of-the-art approaches extracting spatial textures Benggangs. best results obtained using combination DOM data, Canny edge detection, DSM images. Specifically, model achieved accuracy 92.76%, precision 85.00%, recall 77.27%, F1-score 0.8059, demonstrating its adaptability high under complex terrain conditions.

Language: Английский

Assessing the influence of geo-environmental factors on discontinuous gully erosion at regional scales: A case study of spoon gullies on the loess plateau of China

Journal of Environmental Management, Journal Year: 2024, Volume and Issue: 373, P. 123940 - 123940

Published: Dec. 31, 2024

Language: Английский