ETCNext: Multiscale Dual-view Feature Fusion Network for highway vehicle image classification

Accurate vehicle classification is essential for efficient toll collection on highways. According to current highway tolling standards, vehicles are classified into ten detailed categories. However, there has been limited researches on this ten- class classification task. To address this issue, we propose a multiscale dual-view feature fusion network ETCNext and a dual dimensional attention mechanism (DDAM). The proposed ETCNext achieved 99.21% accuracy on the Yunnan highway vehicle dataset, demonstrating outstanding performance. The effectiveness of the proposed DDAM is validated through ablation experiments. To the best of our knowledge, this is the first study to adopt dual-view feature fusion for ten-class highway vehicle classification, offering significant research and practical value.