A Study on Pavement Classification and Recognition Based on VGGNet-16 Transfer Learning

Abstract

The types of road surfaces on which intelligent connected cars operate are complicated and varied, and current research lacks the achievement of real-time and reasonably high accuracy for road surface categorization. In this research, we provide a deep learning-based technique for classifying and identifying road surfaces that makes use of an improved (VGGNet-16) model, in conjunction with a transfer learning strategy, to gather data from the road surface in front of the car using an on-board camera. To accurately classify data based on obtained road surface photos, the dataset is first preprocessed, then pretrained weights are frozen, and the network is initialized using transfer learning parameters. In order to explore the accuracy analysis of the various models regarding the identification of six types of road surfaces, comparisons were made via the VGG16, AlexNet, InceptionV3, and ResNet50 models, using the same parameter values. The experimental findings demonstrate that the improved VGGNet-16 model, combined with the transfer learning approach, achieves 96.87% accuracy for the classification and recognition of pavements, demonstrating the improved network model’s superior accuracy for these tasks. Additionally, the driving recorder of the vehicle may be used as the sensor to complete pavement detection, which has significant financial advantages.