Lightweight Network Research Based on Deep Learning: A Review

Abstract

Deep learning is a field that has attracted a great concern in recent years, and plays an important role in computer vision. Traditional object detection methods failed to adapt to the increasingly complex application environment. While, deep learning, because of the powerful feature extraction capabilities, shows strong ability in object detection tasks in recent years. However, intensive and complex calculations of the deep network are very demanding for the hardware, which makes it will be difficult to deploy on the common hardware devices. In this case, lightweight network technology comes into being. Firstly, this paper analyzes the limitations of deep learning and the necessity of lightweight network technology. Then, According to the existing technology, the methods of lightweight network are summarized and analyzed. In addition, lightweight network methods are compared and analyzed, and the advantages and disadvantages of these methods are pointed out. Finally, we summarize the problems to be faced by the lightweight network approach and the direction of deep learning technology development.