Research on crop classification in Northeast China based on deep learning for Sentinel-2 data

Abstract

With the development of space satellite technology, a large number of high resolution remote sensing images have emerged. Deep learning has become an effective way to handle big data. Crop classification can estimate crop planting area and structure, and classification result is an important input parameter for crop yield model. Crop classification based on deep learning can further improve the estimation accuracy of production. In this paper, multi-temporal Sentinel-2 data and GF2 data are used as data sources. Sentinel-2 data is used as training data, and GF-2 data is used as validation data. Jilin Province in Northeast of China is selected as the experimental area. The experimental area is classified as rice, towns, corn and soy. Firstly, the multi-temporal Sentinel-2 data and GF-2 data is preprocessed. Then, the Sentinel-2 data is used to classify crops based on convolutional neural network (CNN) and visual geometry group (VGG). The red edge band, multiple indexes including normalization difference vegetation index (NDVI), normalized difference water index (NDWI) and difference vegetation index (DVI) are added respectively to compare with the classification results of original multitemporal Sentinel-2 data. The final classification results using CNN and VGG are compared with the other two machine learning algorithms including support vector machine (SVM) and random forest (RF). The experimental results show that the VGG performs best in crop classification accuracy. The overall classification accuracy of the crop can reach 94.8878%, and the Kappa coefficient can reach 0.9253, which is superior to the two traditional machine learning algorithms.