Visual Perception Inspired Deep Learning Models: A Study of Visual Attention and Information Interaction

Visual perception inspired deep learning (VPI-DL) is a subset of machine learning, which aims to build new models by imitating visual hierarchy, information-processing mechanism, multi-mechasnism integration, visual specificity and transferability in humans and animals. In this paper, the latest work in the intersection with deep learning, visual perception, and computational neuroscience is reviewed to provide knowledge about the correlation between biological visual perception system and deep learning. First, the cutting-edge researches of VPI-DL models are reviewed and analyzed in the areas of visual attention mechanism and information-processing mechanism. Then, based on the analysis of the above results, some points of view about the prospects of the VPI-DL are presented. Although there are various visual perception inspired deep learning methods (VPI-DLMs), most of them only focus on the local, single mechanism or structure of visual perception, the study of VPI-DLMs remains fragmented. This survey will provide a comprehensive reference for future research in this promising direction. It aims to provide knowledge and insights into visual perception inspired intelligent models and attract more attention to this field from the scientific communities of artificial intelligence and computational neurosciences.