Advances in Deep Learning for Hyperspectral Image Analysis—Addressing Challenges Arising in Practical Imaging Scenarios

Abstract

Deep neural networks have proven to be very effective for computer vision tasks, such as image classification, object detection, and semantic segmentation-these are primarily applied to color imagery and video. In recent years, there has been an emergence of deep learning algorithms being applied to hyperspectral and multispectral imagery for remote sensing and biomedicine tasks. These multichannel images come with their own unique set of challenges that must be addressed for effective image analysis. Challenges include limited ground truth (annotation is expensive and extensive labeling is often not feasible), and high dimensional nature of the data (each pixel is represented by hundreds of spectral bands), despite being presented by a large amount of unlabeled data and the potential to leverage multiple sensors/sources that observe the same scene. In this chapter, we will review recent advances in the community that leverage deep learning for robust hyperspectral image analysis despite these unique challenges-specifically, we will review unsupervised, semi-supervised and active learning approaches to image analysis, as well as transfer learning approaches for multi-source (e.g. multi-sensor, or multi-temporal) image analysis.

Adversarial Machine Learning in Wireless Communications Using RF Data: A Review

“…These advancements have been described in various surveys as detailed in Table II. For instance in the computer vision domain, surveys have focused on applications of DL, including object detection [140]- [143], 3D Data processing [170]- [173], Hyperspectral Image Analysis [174]- [178], Dialogue Systems [144]- [147] and speech processing [148]- [151] in Natural Language Processing (NLP) and other areas such as Bioinformatics and Computational Biology [156]- [159], Electronic Health Record (EHR) Analysis [159]- [163] and Cancer Diagnosis [164]- [169].…”

Towards On-Board Hyperspectral Satellite Image Segmentation: Understanding Robustness of Deep Learning through Simulating Acquisition Conditions

“…In their recent review, Zhou and Prasad [45] focus on the challenges concerned with the lack of labeled data which they identify as a major obstacle in deploying hyperspectral image analysis based on deep learning. Among the solutions that can help deal with limited amount of ground-truth data, are the unsupervised [46] and semi-supervised [47,48] approaches, including active learning [49].…”

Recent Advances in Multi- and Hyperspectral Image Analysis

“…Importantly, generating ground-truth image data containing manually-delineated objects of interest is not only user-dependent and error-prone, but also cumbersome and costly, as it requires transferring raw image data for further analysis; e.g., from an imaging satellite or other remote imagers [15]. This issue negatively affects our ability to train well-performing supervised learners for HSI analysis that could benefit from large training samples [16,17]. Additionally, the thorough and fair validation of developed approaches is challenging, as their generalization abilities must be investigated with care in order not to infer overoptimistic (or over-pessimistic) conclusions about their performance [18].…”