Application of Image Feature Classification Method Based on Deep Learning

Abstract

This article mainly takes feature learning and classification as the main theme of the subject, introduces the method of single-level feature learning and the method of multi-level feature learning and classification. This paper mainly proposes the characteristics of learning discriminative layer by layer using a technical method based on manifold subspace learning, and combines this technical feature with convolutional networks to form a discriminative local registration network (DLANet). In this paper, DLANet feature is selected as a low-level feature and applied in the traditional machine learning framework. In the scene filter classification experiment based on the DLANet feature, it is found that the best scene filtering result is 85.14% when the number of scene filters is 9. The DLANet feature not only goes beyond the SIFT feature, but also goes beyond the PCANet and LDANet features. DLANet feature is an effective and robust feature on RGBD data.