Sentinel SAR-optical fusion for crop type mapping using deep learning and Google Earth Engine

Accurate crop type mapping provides numerous benefits for a deeper understanding of food systems and yield prediction. Ever-increasing big data, easy access to high-resolution imagery, cloud-based analytics platforms like Google Earth Engine have drastically improved the ability scientists advance data-driven agriculture with algorithms using remote sensing, computer vision, machine learning. Crop techniques mainly relied on standalone SAR optical few studies investigated potential SAR-optical data fusion, coupled virtual constellation, 3-dimensional (3D) deep learning networks. To this extent, we use approach that utilizes denoised backscatter texture information from multi-temporal Sentinel-1 spectral Sentinel-2 ten different types, as well water, soil urban area. Multi-temporal was fused in an effort improve classification accuracies types. We compared results 3D U-Net state-of-the-art networks, including SegNet 2D U-Net, commonly used method such Random Forest. The showed (1) fusing yields higher training overall (OA) (3D 0.992, 0.943, 0.871) testing OA 0.941, 0.847, 0.643) or (2) via denoising convolution neural network (OA 0.912) performed better boxcar 0.880), Lee 0.881), median 0.887) filtered (3) convolutional networks perform than (SAR 0.912, 0.937, 0.992).