DGP-DETR improves underwater target detection accuracy by enhancing feature fusion and detection capabilities, indicating significant benefits in precision and recall metrics.
Abstract The exploration of contemporary underwater visual information collection faces issues of poor image fidelity and the complexity of heterogeneous backgrounds. To address these challenges, this paper proposes the DGP-DETR target detection architecture. By introducing Partical Reparameter Convolution (RPConv) of the backbone, we reducing parameter complexity. The Deformable attention mechanism significantly enhances the model's detection capability by dynamically and adaptively selecting crucial area features. We also designed a Global-Local Spatial Attention Bidirectional Feature Pyramid Network (G-Bifpn), which strengthens feature interaction through multi-scale feature fusion, thereby improving target detection accuracy while optimizing computational efficiency. Experimental results indicate that Precision and Recall improvements of 5.3% and 9.2%, respectively. Moreover, the mAP50 and mAP50-95 metrics show enhancements of 4.2% and 8.6%, respectively. The empirical findings substantiate that the proposed architecture constitutes an effective solution to the inherent challenges encountered in underwater target detection.
No takes yet. Share an insight, caveat, or question.
Lv et al. (2025) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: