Recurrent Neural Networks to Correct Satellite Image Classification Maps

Abstract

While initially devised for image categorization, convolutional neural networks (CNNs) are being increasingly used for the pixelwise semantic labeling of images. However, the proper nature of the most common CNN architectures makes them good at recognizing but poor at localizing objects precisely. This problem is magnified in the context of aerial and satellite image labeling, where a spatially fine object outlining is of paramount importance. Different iterative enhancement algorithms have been presented in the literature to progressively improve the coarse CNN outputs, seeking to sharpen object boundaries around real image edges. However, one must carefully design, choose, and tune such algorithms. Instead, our goal is to directly learn the iterative process itself. For this, we formulate a generic iterative enhancement process inspired from partial differential equations, and observe that it can be expressed as a recurrent neural network (RNN). Consequently, we train such a network from manually labeled data for our enhancement task.

Existing System

Kernel-Based Methods, Artificial Neural Networks.

Proposed System

In this paper, we presented an RNN that learns how to refine the coarse output of another neural network, in the context of pixelwise image labeling. The inputs are both the coarse classification maps to be corrected and the original color image. The output at every RNN iteration is an update to the classification map of the previous iteration, using the color image for guidance. Little human intervention is required, since the specifics of the refinement algorithm are not provided by the user but learned by the network itself. For this, we analyzed different iterative alternatives and devised a general formulation that can be interpreted as a stack of common neuron layers. At training time, the RNN discovers the relevant features to be taken both from the classification map and the input image, as well as the function that combines them.

Architecture

BLOCK DIAGRAM