Road Pothole Defect Detection Based on Improved YOLOv8xPothole Road Recognition with New Improved VOLOv8x AlgorithmTo solve the problem of recognizing road potholes, we use the improved VOLOv8x model to achieve high-precision recognition results.
Year: 2023
Type: conference-paper
Abstract: With the vigorous development of autonomous driving vehicles, pothole road detection has become an increasingly prominent issue. Pothole road detection is a fundamental prerequisite for safe autonomous driving. In this paper, aiming at the problems of low accuracy and slow response in traditional pothole road detection methods, we propose an improved YOLOv8x algorithm based on the Global Attention Mechanism (GAM) and SCConv module. This method introduces the GAM attention mechanism to enhance the network's focus on important features and incorporates the SCConv module to reduce feature redundancy, thus improving the performance and flexibility of the algorithm. The model is trained using pothole and normal road image data. Experimental results show that the improved YOLOv8x algorithm achieves an accuracy (Acc) of 99.12%, a loss of 0.0777, and a detection time of 7.1ms. Compared to YOLOv8x, the Acc is improved by 0.64%, the Loss is reduced by 0.0015, and the Time is shortened by 0.8ms. This algorithm is worth considering for pothole road image classification tasks.
Authors: Jianwei Yu, Ying Zhang, Zihao Chen, Yuhao Li, Chao Deng
Institution: Guangdong Ocean University
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
Funder: Guangdong Ocean University
OpenAlex ID: W4396648564