摘要： 保健酒中可见异物个体微小、形状复杂多变，不利于自动分拣。为此，提出一种基于异物几何特征和不变矩特征的神经网络复合分类方法。通过单层感知器进行一级分类以检测毛发类异物，利用BP网络对非毛发类异物进行二级分类。为提高BP网络训练速度，设计动量因子和学习速率可自适应调整的改进学习算法。实验结果表明，该分类方法识别准确度高，识别速度快。

关键词: 复合分类器, BP算法, 神经网络, 不变矩, 几何特征

Abstract: As for the difficulty of classifying visible foreign body with little size and complex shapes in healthy wine, a hybrid classifying method using neural network based on visible foreign body characteristics is proposed. The hairs are detected by the single-layer perception. Other kinds of particles are classified by BP neural network. In order to improve the BP network training speed, an improved learning algorithm with adaptive momentum factor and learning rate is presented. Experiment results show that this hybrid classified method has high accuracy and fast identification speed .

Key words: composite classifier, BP algorithm, neural network, invariant moment, geometric feature

中图分类号: 

• TP391