Arid Zone Research ›› 2019, Vol. 36 ›› Issue (3): 630-638.doi: 10.13866/j.azr.2019.03.13

• Plant and Plant Physiology • Previous Articles     Next Articles

Grape Recognition Model Based on GF-1/WFV Time Series: A Case Study in Hongsibu District of Ningxia

ZHAO Xi-ni1,2, WANG Lei1,2,3, LIU Ya-qing1,2, QU Xiang-ning1,2, XU Xing1,2, WANG Rui1,2   

  1. 1. Breeding Base for State Key Laboratory of Land Degradation and Ecosystem Restoration in Northwest China,Ningxia University,Yinchuan 750021,Ningxia,China;
    2. Key Laboratory for Restoration and Reconstruction of the Degraded Ecosystem in Northwest China,Ningxia University, Yinchuan 750021,Ningxia,China;
    3. Institute of International Earth System Science,Nanjing University,Nanjing 210093,Jiangsu,China
  • Received:2018-08-28 Revised:2019-01-15 Published:2025-10-18

Abstract: Grape is one of the most widely distributed fruit tree species,and its accurate spatial distribution is of great significance for the management and development of grape planting and wine industry.In this study,the Hongsibao in Ningxia was taken as the study area to obtain the time series of spectrum data in grape growing season based on the Gaofen-1 satellite (GF-1/WFV),the J-M distance analysis was used to analyze the normalized difference vegetation index (NDVI) of the grape plots,and the best recognition phase was determined.The NDVI values of the best phase,adjacent to the value difference rate and curvilinear integral training samples were put into the Clementine data mining software.The extract grape model of decision tree was developed by using C5.0 decision tree classification algorithm and combining with the expert experience method.The results suggested that the established recognition model could meet the needs of grape recognition,but the accuracy was different from different grape plots due to their different coverage.The overall accuracy based on the decision tree classification was 93.71%,and the Kappa coefficient was 0.91.In which the production precision and users' precision of the grape plots with moderate or low coverage were 90.82% and 88.56%,and those with high coverage were 92.44% and 91.18%,respectively.

Key words: grape forest, remote sensing extraction, GF-1/WFV timing data, recognition model, curve integral, decision tree, Ningxia