Satellite image classification with Tensorflow and Keras. The dataset used is available here. The network architecture is based on the PatreoNet, however pretrained weights were not used.
Flowchart of network architecture.
Credit: Towards better exploiting convolutional neural networks for remote sensing scene classification - Keiller Nogueira, Otávio A.B. Penatti b, Jefersson A. dos Santos
Accuracy: 72.5%
Loss: 0.562
There is still a lot of room for improvement, the presented version is the first one.