RSG-YOLO:用于检测道路坑洼的高效神经网络  

RSG-YOLO:An Efficient Neural Network for Road Pothole Detection

作　　者：吴铁钰 杨光[1] 邹丽[1,2] WU Tie-yu;YANG Guang;ZOU Li(School of Railway Intelligent Engineering,Dalian Jiaotong University,Dalian 116028,China;Dalian Key Laboratory of Blockchain Technology and Application,Dalian 116028,China)

机构地区：[1]大连交通大学轨道智能工程学院,辽宁大连116028 [2]大连市区块链技术与应用重点实验室,辽宁大连116028

出　　处：《计算机技术与发展》2025年第2期199-206,共8页Computer Technology and Development

基　　金：辽宁省科学技术基金(2023JH2/101300236);辽宁省教育厅高校科研基金(LJKMZ20220844)。

摘　　要：道路在人类社会的发展中至关重要。坑洼为道路缺陷最严重的一类,容易导致轮胎爆胎、车辆失控,严重影响行车安全。道路坑洼检测有助于及时修复道路缺陷,提高行车安全性。为了解决坑洼大小不同和形状复杂导致的多尺度特征提取困难、检测精度差等问题,该文以深度学习算法中YOLOv8n为基础进行改进,提出了RSG-YOLO模型。模型中使用RFCAConv,利用其动态感受野和避免卷积核参数共享的特性,使模型能够更好地捕捉不同尺度和位置的特征。设计C2f_Sequential Polarized self-attention模块,帮助模型更好地捕获图像中的长程依赖关系和全局上下文信息,有效检测不同大小坑洼。使用GIoU损失函数,其计算过程中对目标边界框的外接矩形的面积的考虑,可帮助模型更好地检测形状和尺寸变化较大的坑洼。最后,经过多组实验结果验证,RSG-YOLO模型相比于原YOLOv8n模型在Precision、Recall、mAP50、mAP50-95、F1-score指标上分别提高了5.0百分点、2.7百分点、3.0百分点、4.2百分点和3.8百分点。Roads are crucial in the development of human society.Potholes are the most serious type of road defects,which can easily lead to tire blowouts and vehicle loss of control,seriously affecting driving safety.Road pothole detection helps to repair road defects in time to improve driving safety.In order to solve the problems of difficult multi-scale feature extraction and poor detection accuracy caused by potholes with different sizes and complex shapes,the RSG-YOLO model is proposed based on the improvement of YOLOv8n in the deep learning algorithm.The use of RFCAConv in the model takes advantage of its dynamic sense field and avoidance of convolutional kernel parameter sharing,allowing the model to better capture features at different scales and locations.A C2f_Sequential Polarized self-attention module is designed to enhance the model's capability in capturing long-range dependencies and global contextual information for effective detection of potholes of varying sizes.The use of the GIoU loss function,which considers the area of the enclosing rectangle around the target bounding box during its computation,can help the model achieve better detection results when the shape and size of pits vary significantly.Finally,after verification through multiple sets of experimental results,the RSG-YOLO model outperforms the original YOLOv8n model by an increase of 5.0 percentage points,2.7 percentage points,3.0 percentage points,4.2 percentage points,and 3.8 percentage points respectively in the Precision,Recall,mAP50,mAP50-95,and F1-score.

关 键 词：道路坑洼检测 多尺度特征 深度学习 YOLOv8n 动态感受野 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]