Abstract: In view of the characteristics of multiple red edge bands of GF-6 wide field view （WFV） multispectral imagery, a method for extracting major food crops from GF-6 WFV image based on feature optimization was proposed. Firstly, characteristic variables, including spectral feature, vegetation index, water index and red edge index, were extracted from preprocessed GF-6 WFV image. Then, the optimal feature set was generated by using a recursive feature elimination algorithm with permutation importance. Finally, machine learning methods and the optimal feature combination were utilized to extract major food crops. Taking Rudong County, Jiangsu Province as the study area, six experiments were used to extract grain crops, and the effects of different characteristics and different classification models on the extraction accuracy of wheat, rice and corn were discussed. The results indicated that the GF-6 WFV image was suitable for extracting major food crops, and the two red-edge bands and red edge indexes of GF-6 WFV data played an important role in distinguishing three main food crops and other objects. Among the six experiments, the overall accuracy of the classification result based on the optimal feature combination and XGBoost algorithm was the highest, improving 3.08 and 5.58 percentage point respectively compared with the classification result without using red edge bands and indexes.

Key words: GF-6, food crop, red-edge band, feature selection, XGBoost