A frequency domain enhanced lightweight oriented object detector for floating raft aquaculture mapping in high-resolution coastal imagery

Accurate mapping of floating raft aquaculture (FRA) from high-resolution remote sensing imagery is essential for long-term coastal aquaculture monitoring and sustainable marine management. However, existing generic object detectors often struggle to achieve a satisfactory balance among target–background discrimination, orientation aware localization, and computational efficiency in large scale multi-temporal applications. To address these challenges, we propose FALOD, a frequency domain enhanced lightweight oriented detector for FRA mapping from high-resolution remote sensing imagery. FALOD introduces a discrete Fourier transform (DFT) based spectral attention module to enhance FRA-related edge, texture, and structural responses while suppressing wave-induced clutter and other sea-surface interference. In addition, an enhanced bidirectional multi-scale feature fusion strategy is designed to strengthen cross-scale feature interaction and improve the representation of dense, elongated, and scale-varying FRA targets. To evaluate the proposed method, we constructed an oriented FRA dataset using high-resolution Google Earth imagery from the Maowei Sea and Qinzhou Bay regions in Guangxi, China. Experimental results show that FALOD achieves a precision of 0.928 and a recall of 0.895, outperforming several representative oriented object detectors. Furthermore, FALOD was applied to large-scale multi-temporal imagery from 2010 to 2025 to investigate the spatiotemporal evolution of FRA activities. The results reveal a marked long-term expansion of FRA, with rapid growth during 2015–2020 followed by a relatively stable but intensified development stage. These findings indicate that FALOD provides an accurate, efficient, and scalable solution for orientation-aware FRA mapping and long-term coastal aquaculture monitoring.