Application of MobileNet Variant Model in Road Classification

This paper reviews the applications and innovative approaches of deep learning based models for road classification in recent years, based on which a MobileNet variant model is proposed and designed, introducing the Squeeze Excitation (SE) module and cross-validation and as well as weight balancing on imbalanced datasets for the identification and analysis of road classification. The design of the MobileNet variant model is then explained in detail and the effectiveness of the introduced novel modules in improving the model performance and generalisation is experimentally verified, with conclusions and suggestions for improvement in future research directions. Finally, we developed a model-based web application.