Semantic Labeling of Aerial and Satellite Imagery

Abstract

Inspired by the recent success of deep convolutional neural networks (CNNs) and feature aggregation in the field of computer vision and machine learning, we propose an effective approach to semantic pixel labeling of aerial and satellite imagery using both CNN features and hand-crafted features. Both CNN and hand-crafted features are applied to dense image patches to produce per-pixel class probabilities. Conditional random fields (CRFs) are applied as a postprocessing step. The CRF infers a labeling that smooths regions while respecting the edges present in the imagery. The combination of these factors leads to a semantic labeling framework which outperforms all existing algorithms on the International Society of Photogrammetry and Remote Sensing (ISPRS) two-dimensional Semantic Labeling Challenge dataset. We advance state-of-the-art results by improving the overall accuracy to $88\%$ on the ISPRS Semantic Labeling Contest. In this paper, we also explore the possibility of applying the proposed framework to other types of data. Our experimental results demonstrate the generalization capability of our approach and its ability to produce accurate results.