Wheel classification using convolutional neural networks
Yuncong Nie, Siyu Xia, Yu Wu · 2018
With the fast development of automobile wheel industry, many methods of wheel classification have been proposed. The appearance of wheel changes a lot during the process of production, but few of these conventional methods is designed to handle the many challenges of wheel classification for different appearances. This paper studies on wheel classification for different appearances during the process of production, and proposes a wheel classification method using convolutional neural network. Firstly, we implement circle detection, gray value analysis and size normalization on wheel images and collect ten common types of wheel to build a dataset. Based on the features of the wheel images, a convolutional neural network is proposed. Then data augmentation is implemented on the dataset and local response normalization layer is added in the convolutional neural network. The experiment results show that the presented method has a better performance both in classification accuracy and real-time requirement than conventional methods and achieves 95.6% accuracy on wheel classification in the dataset.