Machine Learning on Remote Sensing data: Land Use Classification using Statistical Features Raster Representation

The difficulties in monitoring on public investments and large-scale projects are largely due to the complexity of comprehensive monitoring, both in physical terms and from an economic-financial and administrative perspective. This results in a general inefficiency in surveillance. In the context of the Common Agricultural Policy, specifically regarding public subsidies for agricultural producers, ensuring that funds meet their intended impact translates into verifying if the declared crop type is coherent with the species that are actually been planted. The introduction of remotely sensed data and machine learning models today represent the most efficient way to address such challenges. This work aims to introduce a new approach in assessing the compliance of CAP beneficiaries. The project provides the implementation of different tree ensemble classifiers integrated with data from GeorRaster Service, that process satellitar Sentinel-2 data returning a statistical representation for a desired area. The goal of the analysis is, indeed, to assess the potential of the service and evaluate the various methods that can be applied to the data produced.