关键词: 高分六号, 红边波段, 春季作物, 分类精度, 随机森林, Jeffries-Matusita距离

Abstract: The objective of this study is to explore the application of the red edge bands of GF6 WFV image in the identification of spring crops. Based on analysis of the spectral characteristics of single image,the identification and acreage extraction of major spring crops can be effectively achieved by random forest algorithm,taking Qixian,Henan Province as a study area,and employing basic image with 8 bands,which was collected in March 25th,2019.Combined with the ground samples and sample points data,the overall classification accuracy of four schemes, J-M distance and spectral reflectance among different training samples were calculated and analyzed.The result showed that,compared with the scheme without red-edge,the overall identification accuracy of 3 types of ground objects(winter wheat,garlic and others) with one or more red-edge was enhanced,and the separability was improved by calculating the JM distance of different features. Compared with the scheme with red-edge band 1,the overall classification accuracy of red-edge band 2 was improved by 1.98 percentage points.The overall identification accuracy of 3 types of ground objects with all red-edge bands was 86.19%,the Kappa coefficient was 0.79,while the overall identification accuracy of 3 types of ground objects without red-edge was 81.56%,and the Kappa coefficient was 0.72.By introducing all red-edge bands,the overall identification accuracy of 3 ground objects was improved by 4.63 percentage points,the separabilities of winter wheat-garlic,winter wheat-other crops and garlic-other crops were increased by 0.085 6,0.076 1 and 0.025 1 based on J-M distance,respectively.Therefore,by introducing red-edge band,the rate of wrong classification and miss classification,and “Pepper salt” effect were reduced. It could improve the overall identification accuracy of crop planting area.The result of this paper will provide a reference for the application of domesticallyproduced red-edge satellite data in agriculture.

Key words: GF6, Red edge band, Spring crops, Classification accuracy, Random forest, Jeffries-Matusita distance