Deep Learning for Multi-Channel Image Analysis with Applications to Remote Sensing and Biomedicine

Deep neural networks are emerging as a popular choice for multi-channel image analysis – compared with other machine learning approaches, they have been shown to be more effective for a variety of applications in hyperspectral (HSI) and multispectral imaging. We focus on application specific nuances and design choices with respect to deploying such networks for robust analysis of hyperspectral and multispectral images. We provide quantitative and qualitative results with a variety of deep learning architectures in remote sensing, biomedical FTIR and multiplex rat brain images. In this work, not only are traditional deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Convolutional-Recurrent Neural Networks (CRNNs) investigated, we also design and develop Graph based convolutional neural networks (GCNs) with applications to hyperspectral images. To optimize GCNs for HSI data, we proposed a new method of adjacency matrix construction (a semi-supervised adjacency matrix) which leverages class specific and cluster specific properties of the underlying imagery data. Finally, we designed a semi-automatic pipeline that utilizes registration and deep semantic segmentation for aligning (fitting) a brain atlas on multiplex rat brain images.