关键词: Sentinel-2A, 冬小麦, 面向对象, 随机森林, 深度学习, 面积提取

Abstract:  To extract winter wheat planting areas accurately,  efficiently,  and automatically over a large area,  experiments were conducted using Sentinel-2A satellite images. The remote sensing winter wheat extraction method is proposed based on objectoriented and deep learning for mediumresolution images. The area and planting area of winter wheat in Weifang city in 2021 were extracted and compared by using objectoriented classification method and random forest classification algorithm, which proved the feasibility and effectiveness of objectoriented classification method in extracting planting area of winter wheat. In addition,  using objectoriented binary classification images as label images,  UNet is used to build a deep learning neural network model based on the TensorFlow framework. The optimal model obtained by training is used to extract the winter wheat planting area in Weifang City from 2017 to 2021. Field survey data was used to verify the accuracy of the classification results and to analyze the interannual changes in the winter wheat planting area in Weifang City in the past five years. The overall classification accuracy of this classification method reached 93.1%,  and the Kappa coefficient was 0.91. This research method can provide a reliable scientific basis for largescale winter wheat planting guidance and agricultural structure adjustment.

Key words: Sentinel-2A, winter wheat, objectoriented, random forest, deep learning, area extraction