Tien‐Wen Sung, Yixin Chen, Chao-Yang Lee, Qingjun Fang · Journal of King Saud University - Computer and Information Sciences 2026 · 2026
DOI: 10.1007/s44443-026-01416-9
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
To address the inherent limitations of the YOLOv12n model in complex underwater environments, such as insufficient global context capture, restricted detection accuracy, and feature information loss during downsampling, this paper proposes a lightweight enhanced model named WDF-YOLO. By analyzing the baseline model’s constraints through experimental results, we optimize the structural configuration and introduce three core modules to improve detection performance for underwater objects. The Wavelet-Transform Enhanced Channel-Expansion Parallel Model (WTECEP) strengthens the model’s ability to capture low-frequency shape features, global contexts, and deep semantic information. The Multi-Strategy Enhanced Downsampling Module (DS-MSE) reduces feature loss during downsampling while maintaining a balance in model parameters. Additionally, the Fusion Spatial Pyramid Pooling Residual module (FSPPR) integrates multi-path features with the spatial pyramid pooling structure to balance target scale adaptability and computational efficiency. Experimental results demonstrate that WDF-YOLO achieves an m A P 50 of 83.5% on the URPC2021 dataset, representing an improvement of 2.4 percentage points (pp) over the baseline YOLOv12n. With only 1.94M parameters and a computational complexity of 5.9 GFLOPs, the model improves the overall detection accuracy and recall while reducing model size. Furthermore, multi-dataset evaluation on the RUOD and WSODD datasets demonstrates that WDF-YOLO consistently outperforms the YOLOv12n baseline on both additional benchmarks.
Intelligent Marine Technology and Systems 2026 · 0 citations
No comments yet — start the discussion below.