脐橙叶片黄龙病鉴别的激光诱导击穿光谱检测研究  

Identification of Huanglong Disease in Navel Orange by Laser-Induced Breakdown Spectroscopy

作　　者：丁琪萍 姚明印[1,2] 吴书佳 薛乃豪 万奇 曾敏 徐将 DING Qiping;YAO Mingyin;WU Shujia;XUE Naihao;WAN Qi;ZENG Min;XU Jiang(College of Engineering,Jiangxi Agricultural University,Nanchang 330045,China;Jiangxi Key Laboratory of Modern Agricultural Equipment,Nanchang 330045,China)

机构地区：[1]江西农业大学工学院,江西南昌330045 [2]江西省现代农业装备重点实验室,江西南昌330045

出　　处：《江西农业大学学报》2022年第4期1015-1022,共8页Acta Agriculturae Universitatis Jiangxiensis

基　　金：国家自然科学基金项目(31772072,31560482)。

摘　　要：【目的】脐橙黄龙病(HLB)是由一种由柑橘木虱传播的革兰氏阴性细菌引起的毁灭性病害,因其传播迅速、破坏力巨大、不可治愈等特点,被认为是世界上最具破坏性的柑橘病害,严重影响着脐橙产量及品质。激光诱导击穿光谱技术(LIBS)是一种有效的材料成分快速检测技术,由于LIBS具备无需复杂预处理、非接触、损伤小等优点,可以实现赣南脐橙叶片中黄龙病的快速绿色鉴别研究。【方法】试验以赣南地区黄龙病与健康脐橙叶片作为研究对象,利用LIBS技术采集叶片在200~900 nm波长范围的全谱段数据。首先运用九点平滑(9 SM)进行数据预处理,将样本数据按照3∶1的比例分为训练集和预测集。而后采用两种分类模型输入方式,第一种是全谱段数据直接输入至支持向量机(SVM)分类模型;第二种是将光谱数据经主成分分析(PCA)方法提取特征,分别输入至支持向量机(SVM)、线性判别分析(FDA)、径向基函数(RBF)和多层感知(MLP)等4种分类模型,最后对比分析不同分类模型的建模效率以及黄龙病脐橙叶片的判别准确率。【结果】采用PCA方法结合MLP分类模型对黄龙病与健康脐橙叶片的分类效果最佳,训练集及预测集准确率分别为99.43%,98.48%。其次SVM分类模型训练集与预测集的准确率效果也较高,达到98%以上,具有较好的脐橙黄龙病鉴别能力。而PCA-SVM相较于SVM方法提高建模效率,建模时间缩短102 s,但预测集的分类准确率却从98.33%降至96.67%。PCA-FDA和PCA-RBF模型的判别精度和分类效果相对不佳,预测集准确率分别为90.83%及94.94%,这可能是RBF相较于MLP来说,它的隐藏层数量过少,无法对数据维度过多、较为繁杂的多分类问题进行较好的分类,FDA方法对非线性函数的辨认率较低。【结论】将LIBS光谱经九点平滑预处理后,利用PCA方法结合MLP分类模型,建立激光诱导击穿光谱全光学诊断方法,可以进�[Objective]Navel orange Huanglong disease is a devastating disease caused by gram-negative bacteria transmitted by the citrus psyllid. As the most destructive citrus disease in the world due to its rapid spreading speed,huge destructive force and incurable characteristics,Huanglong disease affects seriously the yield and quality of navel orange.As an effective method for rapid detection of material composition,laser-induced breakdown spectroscopy(LIBS)has the advantages of freeing from complex pretreatment and contact and causing little damage,which can realize the rapid identification of HLB in navel orange leaves in southern Jiangxi province.[Method]The leaves of HLB and healthy navel orange in southern Jiangxi were selected as the research objects.LIBS technology was used to collect the whole spectrum data of leaves in the wavelength range of 200-900 nm.Firstly,nine-point smoothing(9 SM)was used to preprocess the data,then the sample data were divided into the training set and the prediction set in a ratio of 3∶1.Two classification-model input methods were adopted. The first method was to input the full spectrum data directly into the support vector Machine(SVM)classification model.The second method was to extract features from spectral data by principal component analysis(PCA)and input them into four classification models including support vector machine(SVM),linear discriminant analysis(FDA),radial basis function(RBF)and multi-layer perception(MLP),respectively.Finally,the efficiency of different classification models and the discrimination accuracy of navel orange leaves were compared and analyzed.[Result]PCA combined with the MLP classification model had the best classification effect on Huanglong disease and healthy navel orange leaves,and the accuracy rates of the training set and the prediction set were 99.43% and 98.48%,respectively.Secondly,the accuracy rate of the training set and the prediction set based on SVM classification model were as high as more than 98%,which showed a high identification a

关 键 词：激光诱导击穿技术 主成分分析 支持向量机 脐橙叶片 绿色鉴别 

分 类 号：S436.669[农业科学—农业昆虫与害虫防治]