视觉动目标辨认计算模型与自适应时延神经网络

We studied time-delay neural network, the possible neural mechanisms of visual asymmetric perception phenomenon on visual moving image-background (i.e., "head/tail preference") in some lower level animals, the uncertainty problem in visual motion information processing system and proposed a fuzzy processing algorithm model. According to recent discovery that there is an asymmetric time window on synaptic modification, we modified Hebb learning rule and constructed a time delay neural network model which could fairly explain the neural principle underlying visual "head/tail preference" and is more widely applicable than Linsker network and Winmbauer time-delay network; We also proposed a fuzzy processing algorithm model which is proved reasonable by simulations of our improved real-time processing neural network setup.Hence we proposed that the detection of motion direction can be transformed into fuzzy classification of it. we suggest that one of the main functions of hypercolumn may be to solve the problem of " based on uncertain visual motion direction information to control motion correctly" and a set of orientation column in hypercolumn may be a fuzzy classifier of visual motion direction. Eight papers have been published in Chinese Science Bulletin etc

应用神经信息时间编码思想，建立视觉动目标辨认计算模型，发展自适应多通道时延神经网络，并研制相应的实时模拟装置和技术对提出的模型加以检验；开展视错觉以及某些低等动物视觉动态图象－背景感知的不对称现象的研究以便从不同的角度加以比较。有助于阐明视觉信息在脑内的编码、表达和加工。在计算机虚拟现实控制等信息高科技领域有应用前景。