Application of Transfer Learning Methods in Vehicle Classification Tasks

With the acceleration of urbanization and the surge in vehicle ownership, vehicle classification technology plays a vital role in intelligent traffic management systems. Although deep learning has significantly advanced vehicle classification research, its performance heavily relies on large-scale annotated datasets, while manual labeling is time-consuming and costly. Moreover, the growing demand for data quality and diversity further intensifies the complexity and time consumption of annotation tasks. This study aims to reduce reliance on labeled data by proposing two transfer learning-based solutions. We conduct systematic experiments using both Convolutional Neural Networks (CNNs) and Transformer frameworks, exploring model optimization strategies from input and output perspectives. Experimental results demonstrate that both approaches yield notable performance gains in improving classification accuracy on the target dataset.