Application of Deep Learning in Vehicle Classification: Boosted by AugMix Image Enhancement

For decades, car classification research has attracted much attention, mainly classifying cars based on factors like model, logo, and size. This paper explores car classification using deep-learning architectures such as convolutional neural network (CNN), Transformer, large model Clip, and model transfer-learning techniques. It adopts the AugMix image-enhancement method to boost data diversity and model generalization ability. It also analyzes the performance of different profound - learning architectures in the car - classification task combined with this strategy. The study shows that VIT and ResNet perform best on the vehicle-image dataset with an accuracy of 96.5%. After AugMix image enhancement, the accuracy of all models improves, indicating that the AugMix technique effectively enhances model performance.