摘要
农作物分类是精准农业中的重要技术之一。为探究多品类作物分类的有效方法,基于无人机高分辨率遥感影像,分别应用基于像元和面向对象分类方法建立了研究区内28类典型农作物分类模型,并采用总体精度、Kappa系数、用户精度、生产者精度对分析结果进行了评价。结果表明:基于像元的最小距离法、马氏距离法、最大似然法、神经网络法和支持向量机的农作物分类结果均存在严重的“椒盐现象”,分类总体精度均低于90%,Kappa系数低于0.9,而面向对象分类法有效解决了地块破碎及作物交织混杂等问题,分类后多数地块完整性良好,更符合实际情况,分类总体精度达91.73%,Kappa系数达0.87。同时,对比分析了2种方法下各类作物的分类结果,发现面向对象分类方法改善了多数作物的分类效果及精度,为基于无人机高分辨率影像的农作物分类提供了参考。
Crop classification is one of the important techniques in precision agriculture.In order to explore an effective method for multi-category crop classification,the image-oriented and object-oriented methods were applied to establish 28 types of typical crop classification models in the study area based on the unmanned aerial vehicle(UAV)high-resolution remote sensing image.The overall accuracy,Kappa coefficient,user accuracy and producer accuracy were used to evaluate the analysis results.The results showed that results of crop classification existed serious“salt-and-pepper noise”based on the minimum distance method,markov distance method,maximum likelihood method,neural network and support vector machine(SVM),which belonged to pixel oriented classification.Overall accuracy of classification was less than 90%.The Kappa coefficient was less than 0.9,while the object-oriented classification method effectively solved the problems of land fragmentation and crop intermixing.After classification,most land plots had good integrity,which was more in line with the actual situation.The overall classification accuracy reached 91.73%,and the Kappa coefficient reached 0.87.At the same time,the results of crop classification under the two methods were compared and analyzed,and it was found that the object-oriented classification method improved the classification effect and accuracy of most crops,which provided a reference for crop classification based on unmanned aerial vehicle(UAV)high-resolution images.
作者
张澜
王妮
朱冰雪
李丹
谢巴图
隋智钟
陈圣波
ZHANG Lan;WANG Ni;ZHU Bing-xue;LI Dan;XIE Ba-tu;SUI Zhi-zhong;CHEN Sheng-bo(College of GeoExploration Science and Technology,Jilin University,Changchun 130026,China)
出处
《江西农业学报》
CAS
2022年第2期178-187,共10页
Acta Agriculturae Jiangxi
基金
吉林大学“大学生创新创业训练计划”创新训练项目(202010183695)。
关键词
无人机
高分辨率影像
面向对象分类
基于像元分类
农作物分类
Unmanned aerial vehicle(UAV)
High resolution image
Object-oriented classification
Pixel-oriented classification
Crop classification
