Program for Satellite Imagery Feature Classification using Deep Learning

Program will be a web-service which goal is to help small and medium companies who face problems of preprocessing and analyzing satellite images. It will be capable of predicting objects on user’s datasets as well as on publicly available sentinel 2 photos. Interactive map will be developed to access sentinel 2 objects predictions. Value of our solution is its ease of use. There are no similar services right now and companies have to use manual labor or hire a high-cost machine learning specialist to get value from satellite data. We will use convolutional neural networks, trained on various datasets to predict hoses, cars, roads and lakes on satellite images. As an input we will use sentinel satellite photos and as a teacher we will use polygons downloaded from openstreetmap. Keywords—machine learning, deep learning, convolutional neural networks, satellite images, geo data, classification problem.