Image recognition is an important research field in artificial intelligence. Recently, with the rapid growth of the number of images on Internet, image recognition technology has been attracted more and more attention. In the image recognition’s research field, it is very important to do the research on classifier. The performance of a classifier often determines the success of a practical application.In the era of Internet and big data, the emergence of deep learning has aroused widespread concern in the academic and industrial circles. It becomes very popular to do research on the theory and application of deep learning. Deep learning can imitate the internal structure of human brain, learn different knowledge and solve the multiclass complex intelligence problem effectively. Many algorithms of deep learning have been widely applied to various research fields.Extreme Learning Machine is developed from the single hidden layer neural network. It is of easy to perform, efficient, requiring less training parameters and good generalization performance. The emergence of extreme learning machine solves the problem that the parameters learning processing of traditional feed-forward neural network is relatively complicated. Afterwards, Incremental Extreme Learning Machine, Regularized Extreme Learning Machine, Kernel Extreme Learning Machine and other improved extreme learning machine algorithms have been proposed one after another. More and more researchers in the world are transferring their research focus to the extreme learning machine’s theory and application currently.In this paper, we combine sparse auto-encoder with extreme learning machine and kernel extreme learning machine and propose two algorithms with deep architecture and apply them to the image recognition task. The two algorithm models are named as Stacked Sparse Auto-encoder-Extreme Learning Machine and Stacked Sparse Auto-encoder-Kernel Extreme Learning Machine. We train the sparse auto-encoders to constitute stacked sparse auto-encoder reference the feature learning method used in deep learning. Stacked sparse auto-encoder with deep architecture can be used to learn features from the original input data. Extreme learning machine and Kernel extreme learning machine are used to classify the input representation after features extracted.In order to examine the performance of two proposed models, we respectively carry out experiments on three different image data sets. The experimental results show that the proposed model’s performance can be not only superior to the shallow architecture models of Support vector machine, Extreme learning machine and Kernel extreme learning machine but also better than the deep architecture models, such as Stacked auto-encoder, Deep belief network and Stacked denoising auto-encoder. |