A pavement distresses identification method optimized for YOLOv5s

Keyou Guo,Sudong Wang,Chengbo He,Min Yang

doi:10.1038/s41598-022-07527-3

Keyou Guo, Sudong Wang + Show 2 more

Open Access

https://doi.org/10.1038/s41598-022-07527-3

Copy DOI

Journal: Scientific Reports	Publication Date: Mar 3, 2022
Citations: 21	License type: open-access

Affiliation: Beijing Technology and Business University

Abstract

Automatic detection and recognition of pavement distresses is the key to timely repair of pavement. Repairing the pavement distresses in time can prevent the destruction of road structure and the occurrence of traffic accidents. However, some other factors, such as a single object category, shading and occlusion, make detection of pavement distresses very challenging. In order to solve these problems, we use the improved YOLOv5 model to detect various pavement distresses. We optimize the YOLOv5 model and introduce attention mechanism to enhance the robustness of the model. The improved model is more suitable for deployment in embedded devices. The optimized model is transplanted to the self-built intelligent mobile platform. Experimental results show that the improved network model proposed in this paper can effectively identify pavement distresses on the self-built intelligent mobile platform and datasets. The precision, recall and mAP are 95.5%, 94.3% and 95%. Compared with YOLOv5s and YOLOv4 models, the mAP of the improved YOLOv5s model is increased by 4.3% and 25.8%. This method can provide technical reference for pavement distresses detection robot.

Highlights

Automatic detection and recognition of pavement distresses is the key to timely repair of pavement
Liu et al.[5] proposed “An Attention-based category-aware gated recurrent unit (GRU) model for the POI Recommendation”, which uses an attention mechanism to focus on the relevant historical check-in traces in the check-in sequence
The optimization of the network framework for the feature extraction part is based on the basic requirements of the pavement distresses recognition algorithm

Summary

Introduction

Automatic detection and recognition of pavement distresses is the key to timely repair of pavement. Experimental results show that the improved network model proposed in this paper can effectively identify pavement distresses on the self-built intelligent mobile platform and datasets. Compared with YOLOv5s and YOLOv4 models, the mAP of the improved YOLOv5s model is increased by 4.3% and 25.8% This method can provide technical reference for pavement distresses detection robot. Traditional computer vision detection methods mainly include threshold-containing s egmentation[6], edge d etection[7], minimum path search[8] and wavelet transform[9] Machine learning methods such as handcrafted feature-based clustering[10], random forests[11] and support vector m achines[12] have obtained good results in detection tasks. In this study, we use deep learning algorithm to further explore the development of artificial intelligence to study the effect of pavement distresses detection in smart traffic. In the section of ‘Conclusions’, we summarize the whole paper and put forward the idea of improving our research in the step

Methods

Results

Conclusion