Crop Classification Based on Multitemporal Sentinel-2 Satellite Imagery

【Objective】The study was to obtain the surface cover and clarify the planting distribution and scope of major crops.【Method】Based on the multitemporal sentinel-2 satellite imagery, 3 remote sensing indexes of NDVI, MNDWI and CI were selected and support vector machine, decision tree auto threshold and random forest were used to classify the single-view data and to evaluate accuracy of every sentinal-2 image with ground field data.【Result】The classification result in July was the best, with the overall accuracy reaching 91.05% and the Kappa coefficient reaching 0.8518. By comparing the classification accuracy of different combinations of time-phase data, the classification effect was better after overlaying the NDVI data from March to October, with the overall accuracy reaching 92.25% and the Kappa coefficient reaching 0.8736. Compare with the three different classification methods, the classification accuracy of SVM was the highest, with the overall accuracy reaching 94.19% and the Kappa coefficient reaching 0.9024.【Conclusion】July is the best time for crop classification in the study area.The classification accuracy of multi-temporal data is significantly higher than that of single scene data. Classification combined with multi-temporal NDVI, MNDWI and CI can effectively identify the crop planting distribution in the study area.