Cotton Area Extraction Based on High-Resolution Sentinel-2 Satellite Images

Cotton is the main economic crop in Xinjiang, China. Timely and accurate acquisition of the spatial distribution of cotton planting areas is of vital importance for cotton yield prediction, policy formulation and the sustainable development of the cotton industry. Remote sensing technology has been widely applied in the identification of crop planting areas by virtue of its advantages of wide coverage and fast update speed. However, there is a relative lack of research on the high-precision extraction of cotton planting areas in this region. This study focuses on the cotton planting areas in Xinjiang. It uses Sentinel-2 satellite images to extract the cotton planting areas during the key growth stages. Three classification methods, namely support vector machine (SVM), random forest (RF), and maximum likelihood classification, were adopted. Among them, the overall accuracy of SVM in single-time-phase images reached 89.52%. Although the classification results of SVM were better than those of the other two methods, it was still necessary to further improve the accuracy. Therefore, for multitemporal remote sensing data, the summation operation of the Normalized Difference Vegetation Index was used to synthesize spectral features, which made the overall accuracy reach 91.96% and achieved a relatively ideal effect. The multitemporal method used in this study can effectively utilize the spectral-temporal characteristics and can provide high-precision technical support for the monitoring of cotton planting areas and agricultural resource management in other regions.