基于改进YOLOv8算法的井盖检测研究

Research on Manhole Cover Detection Based on Improved YOLOv8 Algorithm

摘要: 针对现有井盖检测中存在的速度不足、因数量庞大和环境侵蚀导致误检漏检等问题，本研究提出改进YOLOv8检测算法。其创新点包括：结合井盖侵蚀特征处理数据集并强化关键特征；融合Faster Neural Networks技术构建C2f-faster模块，引入部分卷积(PConv)以减少冗余计算与内存访问，提升空间特征提取效率；设计FasterNet网络结构，在保证精度的同时加快运行速度。实验显示，改进模型较原始YOLOv8的平均精度(mAP)提高5.4%，召回率提升8.2%，计算量减少5.3%，且相较Faster R-CNN、SSD等传统算法，在精度和速度上均具优势，为城市井盖智能检测提供了有效方案。

Abstract: To address issues in existing manhole cover detection, such as insufficient speed and false/missed detections caused by large quantities and environmental erosion, this study proposes an improved YOLOv8 detection algorithm. Its innovations include: processing datasets and enhancing key features based on manhole cover erosion characteristics; integrating Faster Neural Networks technology to construct a C2f-Faster module, and introducing partial convolution (PConv) to reduce redundant computations and memory access while improving spatial feature extraction efficiency; designing a FasterNet network structure to accelerate operation speed while ensuring accuracy. Experiments show that compared with the original YOLOv8, the improved model increases mean average precision (mAP) by 5.4%, recall by 8.2%, and reduces computational load by 5.3%. It also outperforms traditional algorithms like Faster R-CNN and SSD in both accuracy and speed, providing an effective solution for intelligent urban manhole cover detection.