GCD‐Net: A Gated and Cross‐Dynamically Enhanced Network for Accurate Extraction of Offshore Raft Aquaculture

ABSTRACT Raft aquaculture is an important form of offshore mariculture. While its rapid development has brought significant economic benefits to coastal areas, it also poses potential ecological and environmental challenges. Therefore, efficient and accurate extraction of offshore raft aquaculture is crucial for marine resource management and ecological monitoring. To address the issues of insufficient feature utilisation and blurred boundary recognition in raft aquaculture extraction under complex marine environments, this paper proposes a Gated and Cross‐Dynamically Enhanced Network (GCD‐Net). The model incorporates three key innovations: Gated Residual Blocks (GRB) employ multi‐dimensional gating to adaptively calibrate feature flow and mitigate detail loss; Cross‐Guided Attention (CGA) modules achieve mutual enhancement between channel and spatial dimensions to improve focus on discriminative features; and Dynamic Attention ASPP (DA‐ASPP) modules utilise attention‐guided multi‐scale fusion to enhance contextual perception. Evaluated on two offshore raft aquaculture datasets from China, GCD‐Net achieved outstanding performance, with precision, recall, F1‐score, Kappa, and IoU reaching 98.23%, 93.65%, 95.82%, 95.26% and 91.49%, respectively. Comparative studies confirm our method significantly outperforms existing approaches, providing a reliable, automated tool for supporting sustainable coastal management and long‐term environmental impact assessments of aquaculture activities.