Answers are grounded in this paper and cite their sources.
Regional inventories and local condition assessment of waterway regulation structures require complementary observations from satellite and unmanned aerial vehicle (UAV) imagery. Automated interpretation must preserve elongated structures with weak boundaries while separating targets from complex river backgrounds. We propose semantic-guided multiscale fusion and anisotropic refinement YOLO (SGFAR-YOLO), a novel pre-head feature design for independently trained satellite instance segmentation and UAV-oriented detection models. A semantic-guided feature pyramid network (SGFPN) coordinates coarse semantics and fine detail, while the residual strip enhancement module (RSEM) provides gated local and axial context at high-resolution pyramid levels. On the satellite validation split, SGFAR-YOLO achieves 57.31% horizontal-bounding-box AP and 43.14% mask AP, exceeding the YOLO11 baseline by 3.11 and 2.80 percentage points. UAV-oriented box AP reaches 34.75%, a gain of 5.71 percentage points. Ablation studies and qualitative comparisons examine the design across both tasks and illustrate boundary continuity, small-target recovery, and localization of surface damage conditions. The proposed method provides a practical framework for regional mapping of waterway regulation structures and local screening of their surface condition within the Jingjiang reach.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: