[1]吴铁钰,杨光,邹丽.RSG-YOLO:用于检测道路坑洼的高效神经网络[J].计算机技术与发展,2025,(02):199-206.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0330]
　WU Tie-yu,YANG Guang,ZOU Li.RSG-YOLO:An Efficient Neural Network for Road Pothole Detection[J].,2025,(02):199-206.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0330]
点击复制

RSG-YOLO:用于检测道路坑洼的高效神经网络()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年02期
页码:
199-206
栏目:
新型计算应用系统
出版日期:
2025-02-10

文章信息/Info

Title:
RSG-YOLO:An Efficient Neural Network for Road Pothole Detection
文章编号:
1673-629X(2025)02-0199-08
作者:
吴铁钰1; 杨光1; 邹丽1; 2
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
Author(s):
WU Tie-yu1; YANG Guang1; ZOU Li1; 2
1. School of Railway Intelligent Engineering,Dalian Jiaotong University,Dalian 116028,China;
2. Dalian Key Laboratory of Blockchain Technology and Application,Dalian 116028,China
关键词:
道路坑洼检测; 多尺度特征; 深度学习; YOLOv8n; 动态感受野
Keywords:
road pothole detection; multi-scale features; deep learning; YOLOv8n; dynamic feeling wild
分类号:
TP181
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0330
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。
Abstract:
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 per-centage points,and 3. 8 percentage points respectively in the Precision,Recall,mAP50,mAP50-95,and F1-score.

相似文献/References:

[1]朱 鹏,陈 虎*,李 科,等.一种轻量级的多尺度特征人脸检测方法[J].计算机技术与发展,2020,30(04):1.[doi:10. 3969 / j. issn. 1673-629X. 2020. 04. 001]
　ZHU Peng,CHEN Hu*,LI Ke,et al.A Face Detection Method with Lightweight and Multi-scale Feature[J].,2020,30(02):1.[doi:10. 3969 / j. issn. 1673-629X. 2020. 04. 001]
[2]殷 齐,丁 飞,朱 跃,等.基于 CNN 与多尺度特征融合的城市交通流预测模型[J].计算机技术与发展,2022,32(10):175.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 029]
　YIN Qi,DING Fei,ZHU Yue,et al.An Urban Traffic Flow Prediction Model Based on CNN and Multi-scale Feature Fusion[J].,2022,32(02):175.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 029]
[3]蔡俊敏,孙 涵.基于注意力机制和多尺度特征的伪装目标检测[J].计算机技术与发展,2023,33(08):131.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 019]
　CAI Jun-min,SUN Han.Camouflaged Object Detection Based on Attention Mechanism andMulti-scale Features[J].,2023,33(02):131.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 019]
[4]郭 锐,熊风光*,谢剑斌,等.基于改进残差池化层的纹理识别[J].计算机技术与发展,2023,33(09):37.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 006]
　GUO Rui,XIONG Feng-guang*,XIE Jian-bin,et al.Texture Recognition Algorithm Based on Improved Deep Residual Pooling Layer[J].,2023,33(02):37.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 006]
[5]王新武,陈春雨.基于 U 形多尺度注意力方法的真实图像去噪[J].计算机技术与发展,2024,34(04):48.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 008]
　WANG Xin-wu,CHEN Chun-yu.Real-world Image Denoising Based on U-shaped Multi-scale Attention Method[J].,2024,34(02):48.[doi:10. 3969 / j. issn. 1673-629X. 2024. 04. 008]
[6]张鑫雯,邹瑞,宋成,等.基于多尺度特征提取的网络入侵检测方法[J].计算机技术与发展,2025,(06):87.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0053]
　ZHANG Xin-wen,ZOU Rui,SONG Cheng,et al.A Network Intrusion Detection Method Based on Multi-scale Feature Extraction[J].,2025,(02):87.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0053]
[7]尹春勇,沈子宁.基于交互式特征与多尺度特征的文本相似度研究[J].计算机技术与发展,2024,34(08):86.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0140]
　YIN Chun-yong,SHEN Zi-ning.Research on Text Similarity Based on Interactive Features and Multi-scale Features[J].,2024,34(02):86.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0140]
[8]张国有,崔永强.基于双分支注意力机制的图像自动标注研究[J].计算机技术与发展,2024,34(09):167.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0172]
　ZHANG Guo-you,CUI Yong-qiang.Research on Automatic Image Annotation Based on Dual-branch Attention Mechanism[J].,2024,34(02):167.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0172]
[9]于海庆,郑廷帅,史伟*.基于改进PSENet的西夏文检测研究[J].计算机技术与发展,2025,(05):16.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0402]
　YU Hai-qing,ZHENG Ting-shuai,SHI Wei*.Research on Xixia Script Detection Based on Improved PSENet[J].,2025,(02):16.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0402]
[10]严亚滔,卜鹏辉,王航,等.基于全局注意力机制的单目深度估计算法[J].计算机技术与发展,2025,(06):34.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0015]
　YAN Ya-tao,BU Peng-hui,WANG Hang,et al.Monocular Depth Estimation Algorithm Based on Global Attention Mechanism[J].,2025,(02):34.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0015]

更新日期/Last Update: 2025-02-10