Dense Semantic Labeling of Very-High-Resolution Aerial Imagery and LiDAR With Fully-Convolutional Neural Networks and Higher-Order CRFs

Yansong Liu, Sankaranarayanan Piramanayagam, Sildomar T. Monteiro, Eli Saber; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2017, pp. 76-85

Abstract


Efficient and effective multisensor fusion techniques are demanded in order to fully exploit two complementary data modalities, e.g aerial optical imagery, and the LiDAR data. Recent efforts have been mostly devoted to exploring how to properly combine both sensor data using pre-trained deep convolutional neural networks (DCNNs) at the feature level. In this paper, we propose a decision-level fusion approach with a simpler architecture for the task of dense semantic labeling. Our proposed method first obtains two initial probabilistic labeling results from a fully-convolutional neural network and a simple classifier, e.g. logistic regression exploiting spectral channels and LiDAR data, respectively. These two outcomes are then combined within a higher-order conditional random field (CRF). The CRF inference will estimate the final dense semantic labeling results. The proposed method generates the state-of-the-art semantic labeling results.

Related Material


[pdf]
[bibtex]
@InProceedings{Liu_2017_CVPR_Workshops,
author = {Liu, Yansong and Piramanayagam, Sankaranarayanan and Monteiro, Sildomar T. and Saber, Eli},
title = {Dense Semantic Labeling of Very-High-Resolution Aerial Imagery and LiDAR With Fully-Convolutional Neural Networks and Higher-Order CRFs},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {July},
year = {2017}
}