摘要
【目的】探究基于无人机可见光通道和支持向量机(SVM)模型的柑橘黄龙病个体识别方法,为生产上快速、高效发现柑橘黄龙病病株提供参考依据。【方法】构建2套基于SVM的识别模型,先通过柑橘黄龙病黄化识别模型确定具备黄龙病黄化特征的植株,再通过黄龙病斑驳特征识别模型对黄化植株的叶片进行斑驳特征分析确认黄龙病植株;对模型进行1次个体识别试验和2次普适性验证试验。【结果】对于黄龙病植株黄化特征识别,红(R)、绿(G)、蓝(B)、色调(H)和明度(V)在光谱分布上均具备特征性,可作为柑橘黄龙病黄化识别模型的输入变量;对于黄龙病叶片斑驳特征识别,通过叶片左、右部分平均反射率之差(D_(rl))和上、下部分平均反射率之差(D_(ub))及波形振幅的平均值(^(-)A)可有效排除缺素及其他非黄龙病黄化病害对识别结果的干扰,上述3个指标可作为柑橘黄龙病斑驳特征识别模型的输入变量。在个体识别试验中,对28株柑橘进行基于无人机可见光通道和SVM模型的柑橘黄龙病个体识别,识别准确率达100.00%。在中国南方柑橘黄龙病防治研究中心果场开展的普适性验证试验中,对4383株柑橘进行识别,非黄龙病个体识别准确率达100.00%,黄龙病个体识别准确率为89.47%;在广西南宁市武鸣区四明村果园开展的普适性试验中,非黄龙病个体识别准确率为97.30%,黄龙病个体识别准确率为86.67%。【结论】基于无人机可见光通道和SVM机模型的柑橘黄龙病个体识别方法能较好地识别柑橘种植区的黄龙病植株,且相较于高光谱识别方法成本更低,可在柑橘种植区域黄龙病防治中广泛应用。
【Objective】The present study aimed to explore individual identification method of citrus Huanglongbing based on UAV visible light channel and support vector machine(SVM) model, so as to provide reference for rapid and efficient detection of citrus Huanglongbing strain.【Method】Two sets of recognition models based on SVM were constructed. The identification method was to first identify plants with Huanglongbing yellowing characteristics through citrus Huanglongbing yellowing identification model, and then analyzed and confirmed Huanglongbing plants through the identification model of Huanglongbing mottling characteristics on the leaves of yellowing plants. One individual identification test and 2 universal validation tests on the model were conducted.【Result】For Huanglongbing plant yellowing recognition, red(R),green(G),blue(B),hue(H) and brightness(B) had characteristics in spectral distribution and could be used as input variables for citrus Huanglongbing yellowing recognition model. Regarding Huanglongbing leaf mottling feature recognition, the interference of other non-Huanglongbing yellowing diseases such as deficiency of vegetation on the recognition results could be effectively excluded by taking difference of mean reflectivity in the left part and right part(D_(rl)),difference of mean reflectivity in the upper part and below part(D_(ub)) and average value of waveform amplitude(^(-)A) as input indicators of citrus Huanglongbing mottling feature recognition model. In the Huanglongbing individual recognition experiment, citrus Huanglongbing individual recognition based on UAV visible channel and SVM model was performed on 28 citrus plants, and the recognition accuracy was 100.00%. In the universal verification test carried out in the fruit field of citrus Huanglongbing Prevention and Control Research Center in Southern China, 4383 citrus were identified with an accuracy of non-Huanglongbing individual identification was 100.00%, and the accuracy of Huanglongbing individual identification was 89.47%
作者
林奕桐
梁健
刘书田
贾书刚
玉建成
侯彦林
LIN Yi-tong;LIANG Jian;LIU Shu-tian;JIA Shu-gang;YU Jian-cheng;HOU Yan-lin(Nanning Meteorological Bureau,Nanning 530022,China;Nanning Nonnal University,Nanning 530022,China;Guangxi Haipei Intelligent Technology Co.,Ltd.,Nanning 530022,China)
出处
《西南农业学报》
CSCD
北大核心
2022年第11期2554-2563,共10页
Southwest China Journal of Agricultural Sciences
基金
广西科技重大专项(AA17204077)
广西科技基地和人才专项(AD18126012)
广西八桂学者专项(中共广西壮族自治区委员会办公厅发[2019]79)
北部湾环境演变与资源利用教育部重点实验开放课题(GTEU-KLOP-X1705)。
关键词
柑橘黄龙病
无人机
可见光通道
支持向量机(SVM)模型
个体识别
Citrus Huanglongbing
UAV(unmanned aerial vehicle)
Visible light channel
Support vector machine(SVM)model
Individual identification
