关中地区李树遥感辨识的最佳时相与方法

The best phase and the optimal identification method for identifying plum trees in Guanzhong area

关键词:  关中地区  李树  GF1-WFV  遥感辨识  最佳时相

Key words:  Guanzhong area  plum trees  GF1-WFV  remote sensing identification  optimum phase

摘要:

[目的] 探寻李树遥感辨识的最佳时相与方法，为关中地区以及其他果区的李树遥感监测提供理论与方法支撑。[方法] 文章以关中地区为研究区域，基于16种地物样地的感兴趣区数据，采用反射光谱及其差分序列对比与分析、光谱距离法、图像增强处理与分析法、图像差值与比值法、光谱指数法、光谱指数变化分析法和辨识方法优化组合7类方法，探究李树遥感识别并对辨识精度进行了验证。[结果] （1）李树遥感辨识的最佳时相为盛花期；（2）R₆₆₀/R₅₅₅阈值法对盛花期的李树具有较强的辨识效能；（3）两指数（NDVI_3-19与R₄₈₅+R₅₅₅）阈值联用法可以较高精度将盛花期的李树与同时期的梨树、冬小麦、冬油菜、撂荒地予以区分，但是该方法难以将李树与其他10种果树精确区分；（4）三指数（R₆₆₀/R₅₅₅、NDVI_3-19与R₄₈₅+R₅₅₅）阈值联用法可将盛花期的李树与同时期除撂荒地以外的其他地物予以精确区分，但是该方法对李树与撂荒地的区分精度依然不够理想；（5）NDVI_10-19阈值法可将10月中旬的李树与撂荒地精确区分；（6）四指数（R₆₆₀/R₅₅₅、NDVI_10-19、R₄₈₅+R₅₅₅与NDVI_3-19）阈值联用法可高精度识别当年的李树，李树类的分类精度可达95.49%，非李地物类的分类精度可达96.02%，总体分类精度可达95.92%。[结论] 开展李树遥感监测时，融合李树盛花期与10月中旬两期影像，采用四指数阈值联用方法可获得较高的监测精度。

Abstract:

[Objective] Exploring the optimal phase and methods for identify plum trees by using remote sensing images，providing theoretical and methodological support for remote sensing monitoring of plum trees in the Guanzhong region and other fruit regions.[Methods] The Guanzhong region was taken as the research area，and seven kinds of exploratory methods（comparison and analysis for reflection spectra and their differential sequences，spectral distance method，image enhancement processing and analysis method，image difference and ratio method，spectral index method，analysis method of spectral index change，optimal combination of identification methods）were used to identify plum trees based on the ROI（Region of interest）data of sample plots of 16 surface features.[Results] （1）The best identification phase of plum tree was full flowering stage；（2）R₆₆₀/R₅₅₅ threshold method had strong identification efficiency for plum trees during their blooming period. （3）2-index（NDVI_3-19 and R₄₈₅+R₅₅₅）thresholds method could accurately distinguish plum trees during their blooming period from pear trees，winter wheat，winter rape and abandoned lands during the same period；（4）3-index（R₆₆₀/R_555，NDVI_3-19 and R₄₈₅+R₅₅₅）thresholds method could accurately distinguish plum trees during their blooming period from other land features except abandoned lands during the same period，but the accuracy of the method for distinguishing plum trees from abandoned lands was still not ideal；（5）NDVI_10-19 threshold method could accurately distinguish plum trees from abandoned lands in mid-October；（6）4-index（R₆₆₀/R_555，NDVI_3-19，R₄₈₅+R₅₅₅ and NDVI_10-19）thresholds method could accurately identify plum trees of the current year，the classification accuracy of plum trees could reach 95.49%，the classification accuracy of non plum ground objects could reach 96.02%，the overall classification accuracy could reach 95.92%.[Conclusion] When using remote sensing images to monitor plum trees in the research area，composite two images which are collected respectively during the blooming period of plum trees and mid-October，and using 4-index thresholds method can achieve high monitoring accuracy.