基于支持向量机的花生荚果品种识别模型优化研究  

Model Optimization of Peanut Varieties Recognition Based on Support Vector Machine

作　　者：于仁师[1] 孙华丽[1] 宋欣欣 韩仲志[1] 

机构地区：[1]青岛农业大学理学与信息科学学院,山东青岛266109 [2]青岛出入境检验检疫局,山东青岛266001

出　　处：《河南农业科学》2016年第6期157-160,共4页Journal of Henan Agricultural Sciences

基　　金：国家自然科学基金项目(31201133);青岛市科技发展计划项目(14-2-3-52-nsh);青岛市民生计划项目(13-1-3-107-nsh)

摘　　要：为实现通过自动化手段进行花生品种真伪的鉴定,通过扫描仪采集了花生荚果侧面的图像,花生共20个品种,每个品种50个花生荚果,对采集的每幅图像提取形态、颜色、纹理方面的50个特征,首先通过主分量分析(PCA)对这些特征进行组合优化,然后采用RBF核函数搭建了支持向量机模型,最后通过网格搜索法、基因算法和粒子群方法优化支持向量机模型的惩罚参数c与gamma参数。优化结果表明,在主成分累积贡献率为95%时,PCA是10个主分量,3种参数优化方案中20个品种的5折交叉验证识别率分别为78.6%、77.6%、78.0%,识别效果相当,花生品种真伪的二分类识别率最高达到95%。优化后该模型对品种真伪的识别已经基本可以推广到实际生产中使用。Abstract ： In order to realize the identification of peanut varieties automatically, using scanner wecollected the si（ie images of peanut pods. Here were 20 varieties and each variety had 50 pods. For eachimage we extracted 50 characters of shape, color, and texture. First by principal component analysis （PCA ） we did the combinatorial optimization on these characteristics, then using the R B Fbuilt a recognition model based on support vector machine, and finally, usingthe grid search, geneticalgorithm and particle swarm methods optimized the penalty parameter C and gamma parameters of thesupport vector machine model. Optimization results showed that, when the principal component percentage was 95% , the number of principal components was 10. By the three parameter optimization methods, the recognition rates of five-fold cross validation were 78. 6% , 77. 6% , 78. 0% separately for 20 varieties. If there were only 2 kinds of peanut cultivars, the highest classification recognition method of identifying the authenticity of peanut varieties can be used in actual production.

关 键 词：花生荚果 品种识别 支持向量机 模型优化 

分 类 号：S126[农业科学—农业基础科学] S565.2