Abstract:  Satellite remote sensing technology is a commonly used monitoring and classification technology in crops at present. In order to achieve accurate classification and area extraction of regional winter wheat, Yangzhou City, Jiangsu Province had been taken as an example in this study. The Sentinel-2 satellite data and Shuttle Radar Topography Mission (SRTM) elevation data were used as data sources. Four machine learning algorithms, including classification and regression decision tree (CART), Gradient Boosted Decision Tree (GBDT), support vector machine (SVM) and Random forests (RF), were used to establish the classification model. At the same time, the MSI multispectral image of March 22, 2021 in the study area was called and downloaded to extract parameters such as spectrum, texture and terrain features, and the winter wheat in the study area was classified and extracted, and the classification effect and accuracy of the four models were analyzed. The results showed that RF and GBDT classification methods had the best effect and the highest overall accuracy (OA), and both of the OA values were 0.967 and Kappa coefficient was 0.960. The OA of SVM classification method was the lowest (0.514), but the user accuracy (UA) was the highest (0.972). The method mentioned above could realize accurate classification and extraction of regional crops.

Key words: winter wheat, machine learning, single temporal, area extraction, remote sensing classification

摘要： 卫星遥感技术是目前较为常用的农作物监测与分类技术。为实现区域冬小麦精确分类和面积提取，以江苏省扬州市为例，以哨兵2号卫星数据及航天飞机雷达地形测量（SRTM）高程数据为数据源，利用分类与回归决策树（CART）、梯度提升决策树（GBDT）、支持向量机（SVM）和随机森林（RF）4种机器学习算法建立分类模型。同时下载并调用2021年3月22日研究区的MSI多光谱影像，提取光谱、纹理、地形特征等参数，对研究区冬小麦进行分类提取，并分析4种模型的分类效果和精度。结果表明，RF和GBDT分类方法效果最好，总体精度最高，均为0.967，Kappa系数达0.960；SVM分类方法总体精度最低，为0.514，但用户精度最高，为0.972。上述方法可以实现区域农作物的精确分类和提取。

关键词: 冬小麦, 机器学习, 单时相, 面积提取, 遥感分类