Name
Title | ||
|---|---|---|
On the Impact of Tensor Completion in the Classification of Undersampled Hyperspectral Imagery. |
Abstract | ||
|---|---|---|
Typical HSI sensors employ scanning along certain dimensions in order to acquire the hyperspectral data cube. Snapshot Spectral Imaging architectures associate a particular spectral band with each pixel, achieving high temporal sampling rates at a lower spatial resolution. In this work, we study the problem of efficient estimation of missing hyperspectral measurements and we evaluate the impact of the reconstruction quality on the subsequent task of classification. We explore two cutting edge techniques for undersampled signal recovery, namely matrix and tensor completion, and we evaluate their performance on hyperspectral data recovery. Furthermore, we quantify the effects of the reconstruction error on state-of-the-art machine learning algorithms via metrics such as classification accuracy and F1-score. The results demonstrate that robust and efficient classification is feasible, even from a substantially reduced number of measurements being available, especially when emerging deep learning approaches are adopted. Moreover, significant gains are obtained when exploring higher order structural information via tensor modelling, as compared to low order matrix-based methods. |
Year | DOI | Venue |
|---|---|---|
2018 | 10.23919/EUSIPCO.2018.8552934 | European Signal Processing Conference |
Field | DocType | ISSN |
Iterative reconstruction,Spectral imaging,Tensor,Pattern recognition,Computer science,Hyperspectral imaging,Pixel,Artificial intelligence,Deep learning,Spectral bands,Data cube | Conference | 2076-1465 |
Citations | PageRank | References |
0 | 0.34 | 0 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Michalis Giannopoulos | 1 | 0 | 0.34 |
| Grigorios Tsagkatakis | 2 | 122 | 21.53 |
| P. Tsakalides | 3 | 954 | 120.69 |