基于半监督学习的黄曲条跳甲预警方法  

The Forecast of Flea Beetle Based on Semi-supervised Learning Method

作　　者：彭晓琴[1] 杨敬锋[1] 胡月明[1] 李健华[1] 贾正雷[1] 

机构地区：[1]华南农业大学,广州510642

出　　处：《农机化研究》2008年第3期150-152,156,共4页Journal of Agricultural Mechanization Research

基　　金：国家星火计划项目(2004EA780023);广东省自然科学基金项目(04300504;980150);广东省科技计划项目(2005B20701008;2005B101028;2004B20701006);广州市科技计划项目(2004Z2-E0171)

摘　　要：蔬菜病虫害的预警通常依靠植保专家知识来进行,较少采用数学建模方法来进行定量分析。为此,利用部分已知类别的训练样本抽取其关联规则作为监督信息,结合非监督学习的K-mean聚类算法,建立蔬菜黄曲条跳甲的预警模型。半监督学习算法既能发挥有监督学习准确率高的优点,又能充分地利用无监督学习的灵活性,具有一定的研究意义和实际意义。通过对广东省蔬菜黄曲条跳甲数据实验表明,半监督学习算法预警准确率比同条件下K-mean聚类算法的准确率高出24.31%。The forecast of vegetable plant diseases and insect pests commonly bases on experts＇ knowledge of plant protection while math modeling methods are used to analyze the associated data quantitatively scarcely. This paper takes advantage of some classified samples to extract association rules as the supervised information, and combining with the use of the unsupervised learning method K-mean algorithm, the paper establishes the forecast model for vegetable pest flea beetle. Semi-supervised learning method not only obtains a high accuracy of supervised learning method, but also takes advantage of the flexibility of unsupervised learning method, which is meaningful for both research and practice. The experimental results of Guangdong vegetable pest flea beetle show the forecast accuracy of semi-supervised learning method provides a higher accuracy rate of 24.31% than that of K-mean cluster in the same condition.

关 键 词：植物保护 黄曲条跳甲 试验 预警 关联规则 聚类 K-mean算法 

分 类 号：S431[农业科学—农业昆虫与害虫防治]