摘要
为提高枸杞分级正确率,建立一种基于枸杞图像的分级模型,以3个产地共6个等级的枸杞为研究对象,在提取枸杞图像颜色和纹理特征参量的基础上,运用Fisher判别分析(FDA)和核Fisher判别分析(KFDA)方法对6个等级的枸杞进行鉴别分析。在KFDA分析过程中,选取径向基函数(RBF)为核函数,采用基于距离测度的矩阵相似性度量方法,优化确定了RBF核函数的最优参数值为13.2436;选用前150个主成分分析时,基于WilksΛ准则的枸杞分类及验证正确率分别为100.00%和87.80%,而基于传统主成分贡献率的枸杞分类及验证正确率分别100.00%和81.70%。结果表明,运用基于WilksΛ准则的KFDA方法,选用150个最有利于分类的主成分,鉴别结果由FDA的91.7%提高到KFDA的100%。该研究方法不仅有效提高了枸杞图像的鉴别正确率,而且为其他农产品图像分级提供了理论依据。
In order to improve the accuracy of Chinese wolfberry classification,a classification model based on the image of Chinese wolfberry was established. Six kinds of Chinese wolfberry from three different origin areas were studied,Fisher discriminate analysis（ FDA） and kernel Fisher discriminate analysis（ KFDA） were used to identify the 6 kinds of Chinese wolfberry samples,based on the color and texture feature parameters of Chinese wolfberry image. In the method of KFDA,Radius basis function（ RBF） was selected as the kernel function,the measuring method of matrix similarity based on distance discrimination was taken to define the RBF characteristic parameter,and the optimum characteristic parameter was 13. 2436. Selects the first 150 principal components,based on the Wilks Λ criterion,the classification of Chinese wolfberry and verify accuracy were 100. 00% and 87. 80%,respectively,and based on the contribution rate,the classification of Chinese wolfberry and verify accuracy were 100. 00% and 81. 70%,respectively. The final results showed that 150 most conducive PC were selected by this Wilks Λ criterion screening method,and the identification correct rates were respectively from 91. 7%（ FDA） up to 100%（ KFDA）,this method not only improved the correct rate of Chinese wolfberry effectively but also would provide a theoretical guide for the application of the other agriculture products image classification.
出处
《核农学报》
CAS
CSCD
北大核心
2016年第1期96-102,共7页
Journal of Nuclear Agricultural Sciences
基金
河南省
科技创新杰出青年资助项目(624420017)
