Study on Vehicle Recognition Algorithm Based on Convolutional Neural Network

As the application of vehicles expands, the types of vehicles have also become increasingly diverse. While vehicles bring convenience to daily life, they also pose certain traffic hazards due to their different types, such as large transport vehicles entering main roads and violations in prohibited areas and dedicated lanes. Therefore, it has become important to classify and manage actual vehicles. In the field of image processing, the key to vehicle classification is to design classifiers and classify images accordingly. Against this backdrop, relevant classifier algorithms are studied, and classifier design and optimization are completed. Currently, convolutional neural networks (CNN) have been widely applied in image processing and achieved good classification results. To address the issue of actual vehicle classification, this paper selects classic convolutional models, including LeNet5, VGG16, and ResNet50, for transfer learning. The image enhancement processing and optimizer selection are conducted to achieve the final classification. After multiple parameter iterations and comparative experiments, the optimal algorithm combination of VGG16 model, cross-entropy function, and Adam optimizer is selected. The validation accuracy rate of the last iteration under this algorithm combination is 93%, and the optimal accuracy rate throughout the entire training process is 94%.