Crop Field Classification using Data Fusion of Unmanned Aerial Vehicle (UAV) and Sentinel 2A satellite: the case of Oda Dhawata Kebele Cluster farmland, Oromia Region, Ethiopia

Accurate crop classification using remote sensing based satellite imageries approach remains challenging due to mix in spectral signatures. Employing Unmanned Aerial Vehicle (UAV) together with satellite imageries is believed in improving crop classification at field. Accordingly, this study aims to evaluate the potential of UAV images by blending with Sentinel 2A satellite images for crop field classification in Ethiopian agricultural context. The main purpose of the blending is to upgrade and or improve the lower resolution of the data source that is the sentinel 2A data which was 10m resolution. In the study, UAV data was used and preprocessed. The preprocessing includes camera calibration, photo alignment, dense point cloud generation based on the estimated camera positioning of scouting crop types. Then, orthomosaic UAV image was generated from single dense point cloud. Then, the processed UAV data was fused with Sentinel 2A (medium resolution) satellite data using Gram Schmidt pan sharpening method.this method is the most approach that it can run large data sets of spatial resultions. For crop classification, the Random forest (RF) machine-learning algorithm and Maximum likelihood methods were applied. Apart from the UAV and S2A data, field data was collected for training the crop classification. The point field data was collected from Teff, Wheat, Faba bean, Barley and Sorghum crop fields The results show that RF classifier algorithm classifies the crop types with 94% overall accuracy whereas the Maximum likelihood classifier with 90% overall accuracy. This implies that fused image has a potential to be used for crop type classification together with relatively better classification technique with high accuracy level