基于EMDs阵列和BP网络的视觉运动感知过程

VISUAL MOTION PERCEPTION BASED ON EMDs ARRAY AND BP NETWORK

以Ｒｅｉｃｈａｒｄｔ的相关型初级运动检测器阵列和Ｒｕｍｅｌｈａｒｔ的误差反传学习（ｌｅａｒｎｉｎｇｂｙｂａｃｋ－ｐｒｏｐａｇａｔｉｎｇｅｒｒｏｒｓ，ＢＰ）网络相结合构成了一个视觉运动感知神经网络，探讨了视觉运动信息的感知过程。试图从计算神经科学的观点来阐明从一推运动分量的检测到二维模式运动感知的神经原理，从而回答运动矢量在脑内如何表征。计算机仿真表明，在有监督学习的条件下，网络可以学会解决局城运动检测所带来的多义性问题，给出模式的真实朝向、运动方向和运动速度。

In this paper, a visual motinn perception neural network is presented to explore the processing of winal motion information. This model emplops Reichardt's elementary motion detectors array and Rumelhart's BP (learming by back-propagating errors) neural network. In the viewpoint of computational neuroscience, we try to elaborate the nearel mechanism of perception of two dimensional pattern motion from the detection of one dimensional motion component, and answer how the motion vector is represented in the brain. By using computer simulation through a supervised learning process, it is proved that this neural network can solve the 'ambiguity' resultal from local inchon deteCtion and give the aCtual orientation, motion direction and motion sped of the pattern.