Abstract With the rapid development of industrialization and urbanization, the problem of sewage discharge is becoming more and more serious. Accurate detection of sewage discharge outlets is of great importance to environmental protection and water resources management. However, the existing target detection methods often have the problem of insufficient accuracy and low efficiency in the face of serious occlusion and small volume and shape of sewage discharge. To solve these problems, this study proposed a sewage outfall detection method based on the improved YoloV10n deep learning model. Based on the original YoloV10n model, this paper makes a number of improvements. Firstly, the spatial depth conversion Convolution (SPD-Conv) module is introduced to improve the efficiency and accuracy of feature extraction. Secondly, the weighted bidirectional feature pyramid network (BIFPN) is used to enhance the information fusion between features of different scales through learnable weights and normalization processing. Finally, the SEAttention module is introduced to improve the ability of the model to pay attention to important features by adaptively adjusting the channel weights of feature graphs. The experimental results show that the detection performance of the improved YoloV10n model is significantly improved. Specifically, the Precision of the model reached 86.8%, the recall rate (R) reached 84.9%, and the mean accuracy (mAP@50) reached 91.1%, while the mAP (mAP@50-95) covering different IoU thresholds reached 65.3%. Compared with the original model, each index has improved by 2.5%, 1.8%, 1.4% and 1% respectively, which makes the model perform better in the task of sewage outfall detection.
Tang et al. (Mon,) studied this question.