具有旋转不变性的二值化的互连权重研究

采用线性组合Ｃｌｉｐｐｅｄ方法得到初始的具有旋转不变性的ＷＴＡ模型输入层与中间层间的二值互连权重，并利用局部ＭｏｎｔｅＣａｒｌｏ方法对权值进行优化提高系统的正确识别率，以４类飞行目标进行了计算机模拟，实验结果表明此模型用于识别旋转目标是可行的，还提出一个实现此模型的光电混合系统。

STUDY ON BINARY INTERCONNECTION WEIGHT WITH ROTATION INVARIANCE

The linear combination Clipped method was used to obtain an original binary interconnection weight between the input layer and the middle layer of the WTA neural network model.This interconnection weight makes the WTA model with the rotation invariance.To improve the rate of correct recognition of the system the Local Monte Carlo was used to obtain the optimum interconnection weight.A computer simulation was completed for four kinds of flyers. The results of the experiment demonstrated that the WTA model is suitable for recognizing the rotating objects. A hybrid electro optic system was presented to implement the WTA model.