Fourier描绘子和人工神经网络应用于大样本集分类

Fourier Descriptors and Neural Network for Classification of Huge Quangtity of Patterns

模式识别常要求识别结果不受目标的平移、旋转和大小等变化的影响，因此通常先对模式进行规一化预处理，得到与图形形状有关但与目标的平移、旋转和放大缩小无关的图象数据，然后对预处理后的图象数据进行识别分类，这样就减少了原始数据量，且有助于区别带有不同特征的区域。为了识别有变形等失真的图形，要求识别系统有很强的容错能力，因此用神经网络是比较适合的。本文介绍了目前最常用的Ｆｏｕｒｉｅｒ描绘子和反向传播算法（ＢＰ算法），并针对飞机轮廓图的特点对这两种方法分别做了一些必要的讨论、比较和改进。成功地把这两个算法应用到大样本集飞机的分类和识别，网络训练后能识别三维空间中任意姿态的飞机在两维平面上的投影图。

The Fourier descriptor method is a well-known method of preprocessing to eliminate the effects of scale, rotation and translation while recognize two-dimensional Patterns. To recognize the deformed Patterns, neural networks are often used.In this paper we discussed the Fourier descriptor and backpropagation algoritm, and we applied these methods to the problem of identifying the huge quantity of aircrafts from their silhouettes or boundalles. The effect of error due to the digital represntation of image such as chain code error and sampling error on the coefficient of Fourier discriptor are also dboussed in details. And we compared the performance of some normalizizng methods. To improve the Performance of BP, we proposed a simple way to reduce learning time and avoid oscillation by modifying step size during each iteration. The advantage of this method is that it is extremely easy to implement and can lead to significan improvements in the convergence rate.