Abstract: In order to increase the learning speed and enhance the classification accuracy of SOM network，modify existing methods of determining the initial connection weights and the numbers of competitive layer nodes.Clustering method is proposed to determine the initial connection weights.A new method is proposed to determine the numbers of competitive layer nodes by adding up the cluster numbers and the numbers of the neighborhood neurons.And then the improved classification algorithm based on SOM network is presented.Apply the improved SOM network to classify stored-grain pests，and use leave-one-out method to train and test the network.The experimental results show that the modified SOM network has been markedly improved in learning speed and classification accuracy，which can prove the validity of the proposed methods.

Key words: self-organizing map, neural network, classification, clustering, stored-grain pests

摘要： 为了提高自组织特征映射（SOM）神经网络学习速度及分类精度，对初始连接权值及竞争层神经元数的确定方法进行改进。提出用聚类方法确定初始权值的新方法，还提出了采用聚类数与邻域之和确定竞争层神经元数的方法，并给出了改进后的SOM分类算法。将改进的SOM网络用于储粮害虫分类，采用留一方法进行分类验证实验。仿真结果表明，改进后的SOM网络在学习速度和分类精度方面都有明显提高，证明了该方法的有效性。

关键词: 自组织映射, 神经网络, 分类, 聚类, 储粮害虫