A lightweight model RDMSegNet enabling UAV-Based high-resolution mapping of Sonneratia apetala for mangrove ecosystem ecological monitoring

Sonneratia apetala is a crucial mangrove species widely utilized in coastal wetland ecological restoration in southern China, making a substantial contribution to the recovery of mangrove ecosystems. Producing high-precision maps of Sonneratia apetala is therefore essential for environmental monitoring and mangrove management. However, the species exhibits highly diverse crown morphologies in high-resolution imagery, making it difficult for traditional methods to balance accuracy with inference efficiency, which limits their practical utility in real-world monitoring tasks. To address this challenge, this study introduces the Residual Depthwise Multi-scale Segmentation Network (RDMSegNet), a lightweight neural network designed for precise segmentation of Sonneratia apetala using UAV-based high-resolution imagery. RDMSegNet enables an efficient representation of the fine-grained structures of Sonneratia apetala in complex mangrove environments by leveraging lightweight multi-scale feature modelling and cross-layer feature fusion. Experimental results show that RDMSegNet achieves an F1 score of 0.8627 and an IoU of 0.7585 for the Sonneratia apetala class, demonstrating strong discriminative capability. Validation across multiple study regions indicates that the proposed method excels at identifying small-scale fragmented patches and exhibits robust cross-regional generalization and high-resolution mapping performance. Overall, RDMSegNet provides an effective solution for mapping in complex mangrove environments and shows promising potential for high-resolution ecological assessment, monitoring, and management.