DDVC-YOLOv5: An Improved YOLOv5 Model for Road Defect Detection

Abstract

Road defect detection is crucial for enhancing traffic safety, optimizing urban management efficiency, and promoting sustainable urban development. Traditional manual detection methods are inefficient and costly, and most deep learning-based road defect detection models lack superior feature extraction capabilities in complex environments. To address this challenge, this paper proposes an innovative detection framework based on an improved YOLOv5 network. To reduce the processing required for feature extraction and improve detection speed, this study introduces the C3ghost module in both the backbone and neck networks. Furthermore, to enhance the model’s feature extraction capability, this research incorporates the Explicit Visual Center (EVC) module to optimize the feature pyramid layer, thereby improving the model’s detection performance. Additionally, the adaptive feature augmentation dynamic detection head (DyHead) module is introduced to enhance the model’s ability to capture target features at different scales. To validate the performance of the proposed algorithm, it was tested using the RDD2022 dataset. The experimental results demonstrate that the enhanced algorithm achieved an [email protected] of 81.6%, with a precision of 83.1% and a recall of 79.8%. These results indicate improvements of 2.9%, 3.7%, and 7.2% in comparison to the original YOLOv5s algorithm. Moreover, there was a 4.4% decrease in FLOPs. This further illustrates the effectiveness and superiority of the proposed algorithm, providing valuable insights for advancing real-time road defect detection methods.

Keywords

• YOLOv5
• object detection
• deep learning
• dynamic detection head
• explicit visual center