Study on Image Recognition Based on Stacked Sparse Auto-encoder
Gui-Ming CAO, Xiang Qian Ding, Hui -Li GONG · 2017
Image recognition has the characteristics of large sample size, high complexity and redundant information, which has become a hot and difficult topic in the present study.To solve this problem, a feature extraction and image classification model based on the sparse auto-encoder deep neural network is proposed.By using the Greedy layer-wise training, the internal features of the data are learned from the unlabeled data, and the features of the learning are taken as inputs to the softmax classifier.Then, the sparse auto-encoder is tuned by the back propagation algorithm using the data of the label.Finally,the whole model was tested using the test sets data, and compared with the traditional PCA , BP neural network and auto-encoder deep neural network.And the accuracy could reach 91%, which is better than the other methods in the experiment.It has certain practical value for image recognition.