Deep learning-based algorithm for canopy detection in complex scenes

Forest resources play a crucial role in maintaining national ecological security. Rapid and accurate detection of individual tree crowns under multiple tree species in diverse scenarios, as well as extraction of relevant parameter information from UAV remote sensing data, have become key research focuses in forest resource surveying. However, detection accuracy often declines in complex background environments with dense and mixed tree stands. Specifically, under-detection issues frequently occur when tree crowns have small gaps or overlaps, while missing detection problems arise when the internal texture of tree crowns shows low contrast with the background.To address these challenges, this paper proposes an algorithm integrating YOLOv8 and Mamba models, featuring edge information enhancement modules, global information extraction modules, shared convolutional detection heads, and WIMIoU. Compared to the original model, the proposed method achieves a 2.7% improvement in [email protected] and reduces the parameter count by 10.8%, effectively meeting the requirements for practical detection applications.