Research Advanced in Deep Learning Based Image Recognition

Abstract

Image recognition has already been a popular discussion widely studied in computer vision community, which aims to predict the category of each image to distinguish different images. The traditional image recognition methods struggle to process images of large order of magnitude, taking less time and yielding unsatisfactory results. In recent years, benifiting from the powerful feature expression ability of deep convolutional neural networks, both the recognization accuracy and speed of deep learning based mage recognition methods have made a breakthrough, which replacing the traditional methods and becoming the mainstream method of image recognition. In this paper, based on detailed literature research and analysis, we conduct a comprehensive survey on existing methods of image recognition based on deep learning. Specifically, we first introduce the preferential neural networks including self-encoder, convolutional neural network, etc. We then introduce the application of image recognition in various fields and give a series of experiments to analyze the achievement of different detection algorithms on some common datasets. Finally, we try to speculate on the possible future directions of image recognition to make it practical.