Abstract: Constructive Morphology Neural Network（CMNN） is a kind of new nonlinear neural network which is the combination of mathematical morphology and the traditional neural network model and has a strong practicability. Its training algorithm is derived according to the morphological associative memory. Morphological operator is used to rank the test sample to the box in the process of testing. But it is unable to classify this sample which is out of the box into correct group. Someone presents a morphology neural network based on fuzzy lattice（FL-CMNN）. This algorithm can improve effect of classification by calculating the membership grade sample of sample and box. But the complexity of the algorithm is increased and the result is unstable. A kind of lifting algorithm based on morphological neural network（LCMNN） has been presented. This algorithm inherits the quick operation speed of original morphological operator and is able to classify this sample which is out of the box into correct group. Numerical experiments show that LCMNN has the best effect of classification and is simple, with less computation time compared with several other algorithms.

Key words: constructive morphology neural network, ascension algorithms, classification

摘要： 构造性形态学神经网络算法（CMNN）是一种数学形态学与传统的神经网络模型相结合的一种非线性神经网络，有较强的实用性。其训练算法根据形态学联想记忆而来，在测试过程中采用形态学算子将测试样本归类于训练得到的超盒之中。由于其测试过程无法正确地将落在超盒外的样本进行分类，后有人提出了一种基于模糊格的形态学神经网络（FL-CMNN），该算法用样本与超盒的隶属度判断提高了原CMNN算法的分类效果，但增加了算法的复杂程度且分类效果不稳定。这里提出一种基于构造性形态学神经网络算法的提升算法（LCMNN），该算法继承了原有的形态学算子运算速度快的优点且能够将落在超盒之外的样本进行准确地归类。数值实验表明，基于构造性形态学神经网络算法的提升算法（LCMNN）与其他几种算法相比，能够达到最好的分类效果，而且简单易行，计算时间少。

关键词: 构造性形态学神经网络, 提升算法, 分类