Research On Small Sample Image Recognition Based On Brain Network

Throughout the research and development of science and technology,researching technologies with similar human brain capabilities has always been a source of innovation.Especially in today's intelligent development,AI is in full swing,and neuroscience has always played a positive role,and the research and development of visual cognitive models based on human brain networks cannot be separated from the Spiking neural network(Spiking Neural Network,referred to as SNN).SNN originated from biologists' exploration of how the biological nervous system works.Slowly with the rapid development and rapid update of artificial intelligence technology,SNNs are also increasingly used in tasks that artificial neural networks(Artificial Neural Network,referred to as ANN)excel at.Compared with artificial neural networks,impulsive neural networks have unique biological advantages due to their origin.They have higher bio-explanability,low energy consumption,and the potential to better simulate biological visual cognitive processes,therefore,it has attracted the attention of many researchers and scientists.In this thesis,from the three aspects of dynamic image preprocessing,network structure model and network learning and memory,a computational model based on visual cortex image processing mechanism and its application in small sample image recognition are researched.(1)With reference to the pulse discharge mode and synaptic integration of synaptic currents in the brain network,a single static image data is converted into a dynamic input signal based on the pulse sequence to achieve expansion in the time dimension,This method reduces the number of training samples while ensuring accuracy.(2)A neural network is constructed by using the orientation selectivity of V1 / V2 neurons in the primary visual cortex of the brain to achieve a high degree of spatial expansion of image information and use it for image classification and recognition.(3)Draw on visually driven grid cells to capture the layout of composite stimuli in a specific coordinate system by encoding motion vectors between features to support recognition memory,thereby achieving image classification recognition based on a single sample.