瞳孔光反应系统的空间分布式神经网络模型

为模拟刺激光空间分布变化引起瞳孔反应的实验现象，本文建立了空间分布式神经网络瞳孔模型。它是在瞳孔双通道模型基础上，借鉴Ｃａｎｎｏｎ－Ｒｏｂｉｎｓｏｎ的Ｏｃｕｌｏｍｏｔｏｒ模型的双层网络结构和视网膜的镶嵌式特点，经空间延括而成。空间各部位信号经第一层神经元处理得到对应各部位的线性ＤＣ和非线性ＡＣ输出，在第二层神经元进行空间综合，再经第三层神经元复合去控制效应器官虹膜肌的反应。该分布式部位机制模型能解释多种瞳孔实验现象。

A SPATIAL DISTRIBUTIVE NEURAL NETWORK MODEL FOR PUPILLARY CONTROL SYSTEM

A spatial distributive neural network model is presented to explain the experimental phenomenon that transient pupillary responses can be evoked by shifts of light spot constant in luminance. The model was developed from the dual - path pupil model. We modified the structure referring to the double - layer network of Cannon - Robinson' s oculomotor model and extended it into a spatial distributive neural network according to the mosaic retina.The model contains principally three layers. The neurons in the first layer, which form distributive parallel DC and AC paths, process separately the input signals from different positions. The corresponding linear DC outputs and nonlinear AC outputs are integrated respectively by the second layer with logarithmic activation functions. The integrated DC and AC signals are then summed up by the linear neuron in the third layer to control the plant dynamics - - the counterpart of the iris muscle.The parameters of this neural network model were determined by the learning algorithm of the neural network. The training samples were chosen from a set of experimental pupillary responses to a step light stimulus. After training, the simulation results of this model fit the pupillary experiment data very well.