神经网络算法在特色农产品品质分类中的应用  

The application of neural network algorithm in agricultural product quality classification

作　　者：王静[1] 杜勇[1] 赵忠华[1] 

机构地区：[1]新疆师范大学物理与电子工程学院电子系,乌鲁木齐830054

出　　处：《四川大学学报（自然科学版）》2016年第4期805-808,共4页Journal of Sichuan University(Natural Science Edition)

基　　金：新疆师范大学优秀青年教师科研启动基金(XJNU201319)

摘　　要：针对新疆核桃品质分类问题,本文选取特征参数数据,采用神经网络的最速下降BP网络算法、自组织竞争算法、概率神经网络算法建立分类模型,做了训练和测试分类的工作.实验结果表明,三种算法中动量BP网络算法在实现中较为简易直观,相比三种算法,其网络收敛速度较慢,在动量因子的合理选取下,误差在一定范围可以达到收敛的最小震荡;自组织竞争网络在预先设定好的类别范围内,可将分布比较密集的样本进行更加细化的聚类分类,使得分类问题达到更优结果;概率神经网络具有较好的网络收敛速度.实验结果可为实现核桃类坚果的自动化分类、提高工作效率提供一定的理论依据.For walnut quality classification problem in Xinjiang, the author selects feature parameters data, and research classification of the training and testing work using BP network of steepest descent al- gorithm of neural network, the self-organizing competitive algorithm based classification model and probabilistic neural network algorithm. The experimental results show that momentum BP network al- gorithm is relatively simple and intuitive in the implementation, meanwhile the network convergence speed is slow compared with three kinds of algorithm. Under the appropriate selection of the momentum factor, error convergence can achieve a minimum turbulence in a certain range; self-organizing competitive network can achieve better results about detailed classification of clustering in classification problems of dense distribution samples within the scope of the predefined categories; probabilistic neural net- work has better network convergence speed. Experimental results may provide certain theoretical basis to achieve the automation of walnut kinds of nuts and improve the work efficiency.

关 键 词：神经网络 动量最速下降BP法 自组织竞争网络 概率神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]