Authors: Andong Ma*, Texas A&M University, Anthony Filippi, Texas A&M University
Topics: Remote Sensing
Keywords: Hyperspectral remote sensing, image classification, deep learning, convolutional neural networks
Session Type: Poster
Start / End Time: 1:20 PM / 3:00 PM
Room: Napoleon Foyer/Common St. Corridor, Sheraton, 3rd Floor
Presentation File: No File Uploaded
Hyperspectral imaging (HSI) has the potential to discriminate materials better than multispectral. Using HSI, workers can obtain abundant spectral and spatial information simultaneously, which can improve the characterization of the inherent physical and chemical properties of land cover. In this study, we propose a 3D convolutional neural network (3D-CNN) framework for HSI classification. This framework allows for joint analysis of spectral and spatial information for each pixel when performing 3D convolutional processing. In addition, no preprocessing, including feature extraction and feature selection, is required to perform the classification since the original spectral features are employed in this 3D-CNN framework, and high-level hierarchical features will be extracted from the raw input data without any loss. For comparison, we also applied two other classification algorithms in our experiments: support vector machines (SVMs) and a 2D-CNN. Experimental results illustrate that the proposed 3D-CNN method can produce more accurate results than the other two algorithms tested and achieve state-of-the-art performance.