基于高光谱成像的甘蔗叶片早期轮斑病与锈病识别技术  

Identification of Early Wheel Spot and Rust on Sugarcane Leaves Based on Spectral Analysis

作　　者：黄亦其[1] 刘祥焕 黄震宇 钱万强 刘双印 乔曦[1,2] HUANG Yiqi;LIU Xianghuan;HUANG Zhenyu;QIAN Wanqiang;LIU Shuangyin;QIAO Xi(College of Mechanical Engineering,Guangxi University,Nanning 530004,China;Agricultural Genomics Institute at Shenzhen,Chinese Academy of Agricultural Sciences,Shenzhen 518120,China;College of Information Science and Technology,Zhongkai University of Agriculture and Engineering,Guangzhou 510225,China)

机构地区：[1]广西大学机械工程学院,南宁530004 [2]中国农业科学院(深圳)农业基因组研究所,深圳518120 [3]仲恺农业工程学院信息科学与技术学院,广州510225

出　　处：《农业机械学报》2023年第4期259-267,共9页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家重点研发计划项目(2021YFD1400100、2021YFD1400101);广西自然科学基金项目(2021JJA130221);国家自然科学基金项目(61871475)。

摘　　要：针对甘蔗叶片早期轮斑病与锈病发病症状相似,难以区分,导致在实际生产中不便对症施药的问题,以甘蔗早期轮斑病和锈病叶片为研究对象,探究利用高光谱成像技术来识别甘蔗叶片早期轮斑病与锈病的可行性。首先,利用高光谱成像系统在406~1014 nm光谱范围内采集甘蔗健康叶片、早期轮斑病叶片和锈病叶片的高光谱图像,提取图像的感兴趣区域(Region of interest,ROI)并计算其平均光谱作为原始光谱数据,采用一阶导数(First derivative,FD)、Savitzky Golay卷积平滑(Savitzky Golay convolutional smoothing,SG)和标准正态变换(Standard normal variate,SNV)分别对原始光谱数据进行预处理。然后,在预处理的基础上采用主成分分析(Principal component analysis,PCA)算法、蚁群优化(Ant colony optimization,ACO)算法进行特征降维,并将降维后的特征作为后期建模的输入变量。最后,结合降维和不降维2种方式使用支持向量机(SVM)和随机森林(RF)进行识别。为了确定最优的识别模型,对不同的预处理方法、降维方法和分类器共18个组合模型进行了试验。经对比发现,SG SVM识别模型效果最佳,测试集准确率为99.65%。试验结果表明,利用高光谱成像技术进行甘蔗叶片早期轮斑病和锈病的识别可行且有效,可为植保无人机超低空遥感病害监测提供参考。Aiming at the problem that the symptoms of early wheelspot disease and rust disease on sugarcane leaves are similar and difficult to distinguish,which leads to the inconvenience of prescribing the right medicine to the disease in actual production.The feasibility of using hyperspectral imaging technology to identify early wheel spot disease and rust disease on sugarcane leaves was explored.Firstly,hyperspectral images of healthy sugarcane leaves,early wheel spot leaves and rust leaves were collected by hyperspectral imaging system in the spectral range of 406~1014 nm.The average spectral reflectance of region of interest(ROI)was extracted and its average spectrum was calculated as the raw spectral data.The first derivative(FD),Savitzky Golay convolution smoothing(SG)and standard normal variate(SNV)were used to preprocess the original spectral data.Then on the basis of preprocessing,principal component analysis(PCA)and ant colony optimization(ACO)were used to reduce the feature dimension,and the feature after dimensionality reduction were used as the input variables in the later modeling.Finally,the support vector machine(SVM)and random forest(RF)were used for recognition by combining dimensionality reduction and non⁃dimensionality reduction.In order to determine the optimal recognition model,totally 18 combined models with different preprocessing methods,dimensionality reduction methods and classifiers were tested.By comparison,it was found that the SG SVM recognition model had the best effect,and the accuracy of the test set was 99.65%.It was feasible and effective to use hyperspectral imaging technology to identify early wheel spot and rust on sugarcane leaves,which can provide reference for ultra⁃low altitude remote sensing disease monitoring of plant protection UAV.

关 键 词：甘蔗叶片 病害识别 高光谱成像 数据预处理 光谱降维 

分 类 号：S435.661[农业科学—农业昆虫与害虫防治]