Comparison and Performance Analysis of Machine Learning Models in Vehicle Recognition

Vehicles have greatly facilitated people's lives, and they have become an indispensable part of modern life. However, the widespread use of vehicles has significantly increased the complexity of traffic management and the number of traffic accidents. The growing variety and complexity of vehicle types further complicate traffic management from related organizations. At present, many computer-based detection models have been developed and used for vehicle recognition and management. However, due to the lack of accuracy and stability in detection, these models are still limited to basic driver assistance functions, and it might lead to safety issues. This paper aims to apply machine learning methods to vehicle type recognition by constructing classifiers using deep learning models to identify common vehicle types. By doing so, it can assist traffic management and autonomous driving technologies. In this research, images of eight common vehicle types are collected as research samples, which are divided into training and validation sets. Using this dataset, the study investigates the performance of various machine learning models (including DT, GBDT, XGBOOST, LR, SVM, RF, CNN and Densenet) in multi-class classification. Meanwhile, techniques such as data augmentation and pre-training are used to overcome the accuracy declining caused by issues like overfitting. We implement our research methodology and evaluate it via experimental studies. The experimental results show that the classifier trained with the DenseNet model achieves a high accuracy of 91% on the validation set, significantly outperforming other models. This demonstrates the superiority of neural networks in image recognition and their potential applications in the field of vehicle type recognition.