AUTHOR=Yang Li , Hao Zhongyu , Hu Bo , Shan Chaoyang , Wei Dehong , He Dixuan 

TITLE=Improved YOLOX-based detection of condition of road manhole covers

JOURNAL=Frontiers in Built Environment

VOLUME=Volume 10 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/built-environment/articles/10.3389/fbuil.2024.1337984

DOI=10.3389/fbuil.2024.1337984

ISSN=2297-3362

ABSTRACT=The denting and breaking of manhole covers in daily use is a common factor affecting road condition. If this damage is not repaired in time owing to a lack of timely information, it will not only affect the operation of related equipment but also cause great harm to vehicles and pedestrians on the road, which directly threatens the safety of people's lives and property. Therefore, to address these problems, a YOLOX-based improved model for detection of manhole cover status based on car recorders is proposed. The proposed method subdivides manhole cover states into normal, broken, and down. Integration of ECA(High Efficiency Channel Attention) modules into the YOLOX-s architecture. Inserting ECA-Net before the decoupling head of the YOLOX model enhances the model's ability to extract channel features. Experimental results show that this model improves the average accuracy metric (mAP) to 93.91%; moreover, the greatest improvement is seen in the most difficult to detect down class, with an AP value reaching 92.2%. The detection speed is only about five images per second less than the YOLOX model on average. The improved YOLOX model is compared with several mainstream detection models such as Faster R-CNN, SSD and CenterNet. The improved model demonstrated excellent performance in terms of detection accuracy and speed, especially in accurately identifying the challenging "down" state of manhole covers. Therefore, the improved model provides a new means of efficiently determining the location of road manhole covers and identifying their status.