基于高光谱成像的番茄叶霉病的无损检测  

Non-destructive Detection of Tomato Leaf Mold Based on Hyperspectral Imaging

作　　者：崔江南 付芸[1] 赵森 邓泽宇 王天枢 CUI Jiangnan;FU Yun;ZHAO Sen;DENG Zeyu;WANG Tianshu(School of Opto-Electronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学光电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2022年第4期65-71,共7页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金项目(61975021);吉林省重点科技计划项目(20170204015GX,20180201049YY)。

摘　　要：为了实现番茄叶霉病的无损检测,利用高光谱成像系统在370~930 nm波长范围内分别采集了健康、轻微和严重病变三类叶片样本的高光谱数据。首先采用主成分分析法和连续投影法提取光谱数据的特征信息,然后运用网格搜索算法、粒子群算和遗传算法对支持向量机分类器中使用的惩罚因子c和核参数g进行参数寻优,最后分别以全谱数据、PCA提取的2个特征变量、SPA提取的14个特征变量、SPA-PCA提取的6个特征变量作为SVM模型的输入,建立番茄叶霉病的全谱-SVM、PCA-SVM、SPA-SVM和SPA-PCA-SVM分类模型。结果表明,SPA-PCA-SVM模型的分类效果最优,建模输入变量少,检测精度较高,运行速度较快。In order to realize the non-destructive detection of tomato leaf mold,the hyperspectral imaging system was used to collect the hyperspectral data of healthy,mild and severely diseased leaf samples in the wavelength range of 370~930 nm.Firstly,t he principal component analysis and the continuous projection methods were used to extract the characteristic information of the spectral data.Then grid search,particle swarm and genetic algorithms were used to optimize the parame-ters of the penalty factor c and the kernel parameter g used in the support vector machine classifier.Finally,t he full spec-trum data,t he 2 feature variables extracted by PCA,t he 14 feature variables extracted by SPA,and the 6 feature variables extracted by SPA-PCA were used as the input of the SVM model to establish the full spectrum of tomato leaf mold SVM and PCA-SVM,SPA-SVM and SPA-PCA-SVM classification models.The results showed that the SPA-PCA SVM model has the best classification effect,l ess modeling input variables,higher detection accuracy,and faster running speed.

关 键 词：高光谱 番茄叶霉病 特征提取 支持向量机 参数寻优 

分 类 号：O433.4[机械工程—光学工程]