Synergy of Sentinel-1 and Sentinel-2 Imagery for Crop Classification Based on DC-CNN

Over the years, remote sensing technology has become an important means to obtain accurate agricultural production information, such as crop type distribution, due its advantages of large coverage and a short observation period. Nowadays, cooperative use multi-source imagery new development trend in field classification. In this paper, polarimetric components Sentinel-1 (S-1) decomposed by model-based decomposition method adapted dual-polarized SAR data were introduced into classification for first time. Furthermore, Dual-Channel Convolutional Neural Network (DC-CNN) with feature extraction, fusion, encoder-decoder modules based on S-1 Sentinel-2 (S-2) was constructed. The two branches can learn from each other sharing parameters so effectively integrate features extracted high-precision map. proposed method, firstly, backscattering (VV, VH) (volume scattering, remaining scattering) obtained S-1, multispectral S-2. Four candidate combinations formed above features. Following that, optimal one found trial. Next, characteristics input corresponding network branches. extraction module, strong collaboration ability learned parameter sharing, they deeply fused fusion module more results. experimental results showed that components, which increased difference between categories reduced misclassification rate, played role Among four combinations, combination S-2 had higher accuracy than using single source, highest when utilized simultaneously. On basis features, effectiveness verified. DC-CNN reached 98.40%, Kappa scoring 0.98 Macro-F1 0.98, compared 2D-CNN (OA 94.87%, scored 0.92, 0.95), FCN 96.27%, 0.94, 0.96), SegNet 96.90%, 0.95, 0.97). study demonstrated significant potential