3Detection of Urban Underground Sewage Pipeline Based on YOLOv5-Ghostnet Lightweight Model
2023 IEEE 6th International Conference on Pattern Recognition and Artificial Intelligence (PRAI)(2023)
摘要
To detect anomalies in the sewer environment, we identified five common faults: pipe breakage, deformation displacement, accumulation, corrosion, and detachment. However, traditional target detection methods are often cumbersome, slow, and not efficient. To address these challenges, we proposed a new YOLOv5 lightweight detection technology that introduced feature extraction mechanisms from GhostNet, ShuffleNet, and MobileNet. These mechanisms significantly reduce the computation cost of the model and are suitable for comparative experiments. After introducing the GhostNet feature extraction mechanism, it can be observed that the amount of computation and the substantial decrease in the number of model parameters. The mean average precision also increased from 87.4% to 88.5%. This improvement maintains the detection accuracy while reducing the model's weight. Our research provides an essential reference for lightweight research in this field.
更多查看译文
关键词
Target detection,lightweight Yolov5,GhostNet,ShuffleNet,MobileNet
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要