<p>Aiming at the problems such as the coexistence of large-scale and small-scale targets, mutual occlusion of targets, and poor image quality in the underwater environment, LUW-DETR, a lightweight end-to-end model for underwater target detection, is proposed. While achieving a relatively high detection accuracy, this model significantly reduces the computational load and the number of parameters, and can be well applied to underwater edge computing devices with low computing power. First of all, the SCFE module is proposed to reduce computational redundancy and simultaneously enhance the feature extraction ability for underwater targets. Then, a brand-new Dual-Branch ACSF Fusion Net is proposed as the backbone network, which reduces the computational load while realizing the interactive fusion of multi-level features. Secondly, the HiLo self-attention mechanism is introduced to improve the encoder, enhancing the interaction effect between high-dimensional and low-dimensional features. Experiments on the underwater dataset RUOD show that the detection accuracy of LUW-DETR reaches 84.1%, which is almost the same as that of the benchmark model. However, its computational load (GFLOPs) is reduced by 60.5%, and the number of parameters is decreased by 49%, demonstrating an excellent lightweight effect. This provides a reference for underwater detection devices with limited resources.</p>