Hyperspectral and LiDAR data fusion using extinction profiles and deep convolutional neural network

Pedram Ghamisi, Bernhard Höfle, Xiao Xiang Zhu

Research output: Contribution to journalArticlepeer-review

196 Scopus citations

Abstract

This paper proposes a novel framework for the fusion of hyperspectral and light detection and ranging-derived rasterized data using extinction profiles (EPs) and deep learning. In order to extract spatial and elevation information from both the sources, EPs that include different attributes (e.g., height, area, volume, diagonal of the bounding box, and standard deviation) are taken into account. Then, the derived features are fused via either feature stacking or graph-based feature fusion. Finally, the fused features are fed to a deep learning-based classifier (convolutional neural network with logistic regression) to ultimately produce the classification map. The proposed approach is applied to two datasets acquired in Houston, TX, USA, and Trento, Italy. Results indicate that the proposed approach can achieve accurate classification results compared to other approaches. It should be noted that, in this paper, the concept of deep learning has been used for the first time to fuse LiDAR and hyperspectral features, which provides new opportunities for further research.

Original languageEnglish
Article number7786851
Pages (from-to)3011-3024
Number of pages14
JournalIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Volume10
Issue number6
DOIs
StatePublished - Jun 2017

Keywords

  • Convolutional neural network (CNN)
  • Deep learning
  • Extinction profile (EP)
  • Graph-based feature fusion (GBFF)
  • Hyperspectral
  • Light detection and ranging (LiDAR)
  • Random forest (RF)
  • Support vector machines (SVMs)

Fingerprint

Dive into the research topics of 'Hyperspectral and LiDAR data fusion using extinction profiles and deep convolutional neural network'. Together they form a unique fingerprint.

Cite this