Abstract:

The red−edge band and red−edge vegetation index have been proven to be related to fine−scale identification of crop types, and evaluating the impact of red−edge information from satellite remote sensing images on the extraction accuracy of specific crop cultivation can provide an effective reference for the deeper application of related satellite data in agriculture, and at present there are fewer assessment studies on the impact of red−edge information of the new generation of environmental satellites, HJ−2. In order to explore the impact of red−edge information from HJ−2 satellite remote sensing images on the extraction accuracy of winter wheat planting areas, Xiao county of Anhui province was taken as the study area, and four planting extraction schemes with different bands and derived vegetation indices were constructed from single image of winter wheat in the jointing period, on the basis of which, the Jeffries−Matusita（J−M）distance was used to calculate the separability between sample categories, and Random forest algorithm was applied for the classification of the land covers, realized the comparative analysis for category separability between winter wheat and other vegetation, recognition accuracy and the distribution map of winter wheat planting area under different schemes. The results showed that under the condition of all red−edge information participation, compared with the condition of no red−edge information and no derived vegetation index participation, the category separability between winter wheat and other vegetation improved from 1.9337 to 1.9988, and the classification accuracies of the recognition accuracy, UA（user accuracy）, PA（producer accuracy）, OA（overall accuracy） and Kappa, improved from 86.92%, 90.37%, 85.11% and 0.77 to 98.45%, 95.67%, 94.95% and 0.92, respectively, the AA（area accuracy） improved from 80.38% to 98.94%, the RE（relative error）decreased from 19.62% to 1.06%, the misclassification phenomenon and the “phenomenon of salt and pepper” in the distribution map of winter wheat planting areas were significantly reduced, and the identified winter wheat planting plots were complete, smooth and had good boundary continuity. This study demonstrates that the introduction of HJ−2 satellite remote sensing imagery red−edge information to participate in the classification could effectively improve the accuracy of winter wheat planting extraction and mapping effect, which was of important reference significance for the popularization and application of domestic red−edge satellite and the accurate monitoring of winter wheat.

Key words: HJ?2 satellite, Red?edge information, Winter wheat, Planting extraction, Random forest, Vegetation index