模糊格构造型形态神经网络

摘要

A novel neural network model named fuzzy lattice constructive morphological neural network (FL-CMNN ) is presented to overcome the deficiency of the original constructive morphological neural network (CMNN ) ,which suffers for the prob-lem of decision function in classification phase .The fuzzy lattice inclusion measure function is introduced to calculate the member-ship of testing sample belong to the hyper-boxes trained by the CMNN .Three standard datasets are employed to evaluate and com-pare the presented FL-CMNN with the CMNN ,artificial neural network (ANN ) ,support vector machine (SVM )and K nearest neigh-bor(KNN)classifiers .Experimental results have revealed that the presented FL-CMNN yields better performance than the original CMNN model .It also achieved comparative classification accuracies with much lower computational cost than traditional ANN and SVM model .%针对构造型形态神经网络（CMNN ）决策函数的局限性，提出了一种模糊格构造型形态神经网络（FL-CMNN ）；该模型在利用训练好的CMNN进行分类时，引入模糊格包容性测度计算测试样本属于各超盒的隶属度值。采用仿真数据集对提出的FL-CMNN模型进行了评价，并与原始的CMNN和传统的人工神经网络、支持向量机、最近邻分类器进行了对比；试验结果表明，FL-CMNN在测试精度上明显优于原始的CMNN ，训练时间远远低于传统的神经网络和支持向量机，而分类精度丝毫不亚于传统的神经网络和支持向量机。