Recurrent Neural Networks to Enhance Satellite Image Classification Maps

The automatic pixelwise labeling of satellite images is of paramount importance in remote sensing. Convolutional neural networks represent a competitive means to learn the contextual features required to distinguish an object class from the rest. However, spatial precision is usually lost in the process, leading to coarse classification maps that do not accurately outline the objects. While specific enhancement algorithms have been designed in the literature to improve such coarse neural network outputs, it requires a decision-making process to choose and tune the right enhancement algorithm. Instead, we aim at learning the enhancement algorithm itself. We consider the class of partial differential equations, see them as iterative enhancement processes, and observe that they can be expressed as recurrent neural networks. Consequently, we train a recurrent neural network from manually labeled data for our enhancement task. In a series of experiments we show that our network effectively learns an iterative process that significantly improves the quality of satellite image classification maps.