为提高现有害虫分类方法的分类效果,该文分析了现有害虫分类方法的局限性,在此基础上,提出一种基于压缩感知理论的害虫分类新方法.该方法利用害虫训练样本构造训练样本矩阵,通过求解l1范数意义下的最优化问题实现害虫测试样本的稀疏分解,由于稀疏分解结果中包含了明确的分类信息,可直接用于害虫分类.利用该方法对12类储粮害虫和110类常见害虫进行分类,在4种不同试验条件下,分类准确率分别达到92.9418％、98.2877％、78.8651％和61.5938％,证实了压缩感知理论用于害虫分类是合理可行的.%In order to improve the effectiveness of the existing classification methods of pests, a novel classification method of pests was presented by using compressed sensing theory. In the proposed method, a large number of the representative training samples of pests were used to construct the training samples matrix, and then the sparse decomposition representation of the testing samples of pests was obtained by solving the //-norm optimization problem, which had distinct class information and could be used for the different species of pest classification directly. The 12 species of stored-grain pests and the 110 species of common pests were separately classified by the proposed method, and the classification precision reached around 92.9418%、 98.2877%、 78.8651% and 61.5938% respectively under 4 kinds of different experimental conditions. The experimental results indicated that the application of compressed sensing theory in the classification of pests was practical and feasible.
展开▼
