Road Disease Detection Algorithm Based on YOLOv5s-DSG

Abstract

Automatic detection and classification of road conditions are critical for the timely maintenance and repair of road surfaces. Aiming at the problems of single detection type, low detection efficiency, the low resolution of detection objects, and difficult detection of small object features in the road disease detection scene, this paper proposes an improved YOLOv5s road disease detection algorithm, YOLOv5s-DSG. First, optimize the depth and width of the network structure to reduce the impact on road damage image detection performance. Secondly, the Ghost module replaces the traditional convolution to reduce the number of model parameters, making the model lightweight and improving the detection rate. Finally, the Space-to-depth-Conv module is introduced to adapt to low-resolution and small object detection tasks. A large number of experiments on datasets such as Road Damage Dataset 2022 show that the average accuracy of the improved model at low resolution 640×640 is 1.1% higher than that of the original model, and the FPS is increased from 85 to 90; when the resolution is reduced to 480×480, the average accuracy remains the same and the FPS increases to 93. Compared with existing algorithms, it has apparent advantages in road comprehensive disease detection and classification.

Review of Intelligent Road Defects Detection Technology

“…However, its performance was influenced by the transmitting frequencies of the GPR due to the detail loss of the pavement distresses in the GPR images. Tian Yang et al [120] adopted the fast regional convolutional neural network (Fast-RCNN) method to organically combine image recognition technology, a global positioning system (GPS) position information, and vehicle-mounted signal information. Road pictures collected by road inspection vehicles were automatically identified.…”