基于YOLOv12-DRNet的排水管道树根识别研究

打开文本图片集

中图分类号：TH122 文献标识码：A

Abstract：The traditional methods for tree root detection are ineficient and subject to subjective factors. Therefore，a tree root recognition model for drainage pipes based on YOLOvl2 is proposed in this paper. The model adopts the YOLOvl2 algorithm，which integrates the attentionenhanced convolution module，multi-scale feature fusionand distribution focal loss to improve the accuracy and real-time performance of tree root recognition， providing an efficient means for the maintenance and management of drainage pipes. A self-built database containing 1 Ooo tree root images was constructed using CCTV images and camera shots of pipes，and data augmentation techniques are used to expand it to a database of 5 Ooo tree root images. Experiments show that the recognition accuracy rate has increased to O.806，the recall rate has increased to O.697，mAP ④50 has increased to O. 929，and mAP@50-95 has increased to O.50o. This indicates that the model has significantly improved the recognition accuracy and real-time performance in the recognition task.

Keywords： drainage pipes; root recognition； YOLOvl2； target recognition; deep learning

作为农业基础设施的重要组成部分，排水管道的正常运行对于农田水资源的利用处理至关重要[1-2]。（剩余10483字）