Fine-grained vehicle recognition method based on improved ResNet
Ailing Qi, Tian Ning · 2020
Traditional image recognition algorithms generally extract features manually and use classifiers for classification. Aiming at the problems of general recognition effect and limited generalization ability, an improved ResNet-based fine-grained vehicle recognition method is proposed. The basic network adopts the residual structure, which improves the flow of information by introducing jump connections and can effectively alleviate over-fitting. And add the Inception module to the ResNet model, It uses different sizes of convolution kernels and has different sizes of receptive fields, which can further improve the recognition accuracy. The optimization algorithm uses Adam optimizer, the experiment uses Stanford University model database stanford cars as the data source, and compares experiments with other methods. The results show that the accuracy of this method on the test set reaches 91%, which effectively improves the recognition accuracy.