基于PCA-PSO-LSSVM的茶叶品质计算机视觉分级研究  

Study on computer vision classification of tea quality based on PCA-PSO-LSSVM

作　　者：余洪[1] 吴瑞梅[1] 艾施荣[1] 范苑[1] 吴彦红[1] 刘木华[1] 

机构地区：[1]江西农业大学工学院,南昌330045

出　　处：《激光杂志》2017年第1期51-54,共4页Laser Journal

基　　金：国家自然科学基金项目(31460315);江西农业大学大学生创新创业训练项目

摘　　要：国内外茶叶品质评价主要以感官审评方法评定。基于茶叶外形品质的评茶师审评结果,将72个茶样分成4个等级,建立茶叶品质的计算机视觉最小二乘支持向量机(LSSVM)分级模型。对茶叶外形特征参数进行主成分分析,采用粒子群算法(PSO)对LSSVM算法的惩罚系数(C)和核参数(σ2)进行优化。当主成分因子数为5,优化得到的惩罚系数C为65.6085,核参数σ2为35.7213时,建立的LSSVM模型识别精度最高。该模型对校正集的总体回判率为93.75%,测试集总体识别率为91.67%。结果表明,采用PCA-PSO-LSSVM建立的茶叶品质计算机视觉分级模型,比PSO-LSSVM、传统LSSVM、SVM模型具有更高的识别精度。可为茶叶品质的实时快速检测提供方法支持。The evaluation of tea quality is mainly assessed by sensory organs at home and abroad. The 72 tea samples are divided into 4 grades through the evaluation results of tea experts. The classification model of tea quality based on the computer vision was developed by least squares support vector machine （LSSVM） method. Principal component analysis method was used to analyze the tea appearance characteristic. And particle swarm optimization （PSO） algorithm was used to optimize the punishment coefficient （c） and kernel parameters （σ^2） of LSSVM algorithm. When PCs is 5, the optimal penalty factor C is 65. 6085 and kernel parameter σ^2 is 35.7213. Based on the optimal parameters, the recognition accuracy of the developed LSSVM classification model is the highest. The overall recognition rate of the calibration set was 93.75%, and the overall recognition rate of the test set was 91.67%. The results showed that the recognition accuracy of the computer vision classification model of tea quality by PCA-PSO-LSSVM method was higher than that of PSO-LSSVM, traditional LSSVM and SVM model. The method can provide the support for the real-time and rapid detection of tea quality.

关 键 词：茶叶品质 LSSVM 粒子群算法 感官评价 计算机视觉 

分 类 号：TN209[电子电信—物理电子学]