Accurate Oracle Classification Based on Deep Convolutional Neural Network
Zhen Yang, Qiqi Wang, Xiuying He, Yang Liu, Fan Yang, Zhijian Yin, Chen Yao · 2018
The primary issue in the study of oracle is how to recognize and understand oracle characters. Therefore, the recognition of oracle is an significant research in the oracle field. In this paper, we firstly exploy different convolutional neural network (CNN) architectures, which are aim at implementing automatic feature extraction and classification to improve the performance of oracle recognition. We also show that the Gabor feature extraction method, which is demonstrated effectively to improve the precision of classification. To reduce overfitting caused by insufficient training samples, we generate more training samples from original training data sets by using a variety of data argument techniques. Compared to the raw image data, traditionally directional feature is able to better performance in network. Experimental results show that AlexNet models achieve recognition accuracy of 97.82%.