Confusion matrix.

Marine biological detection is critical to environmental conservation and the use of marine resources. In actual applications, detecting aquatic species quickly and accurately while using few resources remains a difficulty. To address this problem, this research proposes a novel fast and efficient lightweight target detection network (SCR-Net). First, a fast and lightweight Spatial Pyramid Pool ELAN (SPPE) module is proposed and implemented, which enhances the model’s performance by leveraging ELAN’s effective feature aggregation ability and SPPF’s spatial pyramid pooling capacity. Second, a cross-scale feature fusion pyramid (CFFP) structure is introduced, which significantly reduces the number of parameters and computational cost during feature fusion. Third, a lightweight feature extraction module named RGE is designed, utilizing low-cost processes to create duplicate feature maps and reparameterization to drastically accelerate model inference. Compared to the baseline model, SCR-Net has 57.4% fewer parameters, 37% less computation, and an mAP@0.5 of 83.2% on the DUO dataset. Ablation experiments validate the effectiveness of the proposed modules, and comparative experiments on DUO and UDD datasets demonstrate that SCR-Net achieves superior overall performance compared to existing lightweight state-of-the-art underwater target detection models.