Natural Video Encoding And Semantic Network Analysis Based On Functional Magnetic Resonance Imaging

Human brain is the most important information processing.The main way to perceive and recognize the outside world of human beings is the visual system.Therefore,interpret the visual information has become an important way to explore the information processing mechanism of brain to the outside world,and it is a research hotspot in the field of neural information science.Furthermore,exploring brain understanding mechanism of semantic information provides a new research idea and method for visual information interpretation.Due to the high spatial resolution,functional magnetic resonance imaging(fMRI)is become one of the main way to interpret visual information of the human brain.In view of the semantic information comprehension mechanism of brain,this paper mainly explores the brain semantic network under the condition of natural video tasks by means of fMRI.This paper focuses on the visual encoding model of natural video semantic features and predictive response of brain voxels to word categories,and a brain semantic network was constructed by binding fMRI feature selection method based on the visual encoding model.After that,using the ways of complex network analysis to analyze brain semantic network and WordNet word connection networks.The main work includes:1.A visual encoding model based on the natural video semantic features was constructed,and the encoding model was used to analyze the predictive response of brain voxels to semantic information.The semantic features of video were extracted by annotating objects and actions in video,the hierarchical “is a” relationships defined by WordNet were used to infer the high-order categories.And the methods of model building in machine learning was used to obtain the visual encoding model.The results show that the semantic features extracted from the video can effectively represent the video stimulus information.Furthermore,the performance of the visual encoding model and the significantly different predictive response of brain voxels to different semantic information indicate the effectiveness of brain semantic network construction.2.The brain semantic network was constructed by binding fMRI feature selection method based on the visual encoding model.Similarities and differences of brain semantic network and WordNet word connection network was analyzed from the perspective of complex network statistic features.The results show that the similarity between different subjects' brain semantic networks is very high,which is quite different from the WordNet word connection network.3.From the perspective of complex network community detection,Similarities and differences of brain semantic network and WordNet word connection network was analyzed.The results of the analysis of the relationship between the community structure of the network and the semantic information of the word nodes in the same community show that the way of human to classify the words is based on some related features of the words,and it is different to the way of WordNet to define the words according to their hierarchical relationship.