Parcel-Level Crop Classification in Plain Fragmented Regions Based on Multi-Source Remote Sensing Images

Abstract

Accurately obtaining crop cultivation extent and estimating the cultivated area are significant for adjusting regional planting structure. This article proposes a parcel-level crop classification method using time-series, medium-resolution, remote sensing images and single-phase, high-spatial-resolution, remote sensing images. The deep learning semantic segmentation network feature pyramid network with squeeze-and-excitation network (FPN???SENet) and multi-scale segmentation were used to extract cultivated land parcels from Gaofen-2 imagery, while the pixel-level crop types were classified by using support vector machine algorithms from time-series Sentinel-2 images. Then, the parcel-level crop classification was obtained from the pixel-level crop types and land parcels.