LBAM神经网络及其在图像识别中的应用

Liked-Bam Neural Network for Image Recognition

本文提出了一种用于图像识别的神经网络．它由映射网络ＭＮ（ＭａｐｐｉｎｇＮｅｔｗｏｒｋ）和ＬＢＡＭ（ＬｉｋｅｄＢｉｄｉｒｅｃｔｉｏｎａｌＡｓｓｏｃｉａｔｅｄＭｅｍｏｒｙ）网络组成．ＭＮ中使用了不变性变换法，降低了图像样本的维数且保持分类距离不变．在设计ＬＢＡＭ网络时，通过全局考虑，使得网络的吸引点和吸引区域满足实际全局最优之需要．ＬＢＡＭ具有网络结构简单和收敛速度快的优点，计算机模拟证实。

A neural network model and its application to image recognition are proposed in this paper. This model consists of Mapping Network (MN) and Liked Bidirectional Associative Memory (LBAM). Invariant mapping is used in MN in order to decrease the number of dimensions of image samples and not to change the distance between them. LBAM'S structure is simple and its convergence speed is fast. Several computer simulations given to prove that the model is capable of recognizing the corrupted targets and the incomplete targets.