Image Classification Based on Deep Belief Network and YELM

Abstract

For a large amount of image data, an image classification method based on improved deep belief network (IDBN) and improved extreme learning machine (YELM) was proposed. Firstly, IDBN was used to simplify the complex high-dimensional data to lower dimensions space, and get the main inherent feature of the images with lower dimension. Then YELM was used to classify data after the dimension reduction. The proposed IDBN-YELM method has a significant improvement in classification accuracy. Extensive experiments were performed using challenging dataset and results were compared against the models such as DBN, YELM, IDBN-ELM. Though a lot of comparative experiments on the street view house numbers (SVHN) dataset, the results show that the IDBN-YELM has the high classification accuracy for the problem of large-scale image classification.