Super-resolution Image via Hyperspectral and Multispectral Data Fusion using Big-data Convex Optimization
讲座名称 | Super-resolution Image via Hyperspectral and Multispectral Data Fusion using Big-data Convex Optimization |
讲座时间 | 2018-11-01 10:00:00 |
讲座地点 | 北校区新科技楼 1012会议室 |
讲座人 | Prof. Chong-Yung Chi |
讲座人介绍 | ![]() |
讲座内容 | Direct acquisition of hyperspectral and high-spatial-resolution (HSR) image data (target images) with remote-sensing sensors is expensive, while such images play an essential role in accurate identification/classification of the underlying materials in the scene of interest. In this talk, we present a newly invented convex optimization based coupled non-negative matrix factorization (CO-CNMF) algorithm to fuse two different datasets, one captured by HSR multispectral sensor and the other by low-spatial-resolution (LSR) hyperspectral sensor. With sparsity-promoting l1-norm regularization and simplex (of materials’ spectral signatures) volume-demoting regularization, this algorithm solves a large-scale bi-convex problem, by the alternating direction method of multipliers (ADMM), where two non-negative matrix factorizations are intrinsically coupled by a low-rank model of the target image, one for extracting the spectral information from the LSR hyperspectral image and the other for extracting the spatial information from the HSR multispectral image. The CO-CNMF algorithm can be shown to converge to a stationary-point solution together with a carefully designed ADMM that are practically applicable to fusion of million-scale datasets. Finally, experiments designed based on the Wald’s protocol are presented to demonstrate that the CO-CNMF algorithm yields much superior fusion performance over six benchmark super-resolution methods on ROSIS, HYDICE and AVIRIS datasets, in terms of several performance measures including peak signal-to-noise ratio (PSNR) and root mean square error (RMSE), with lower or comparable computational load. |
转载请注明出处:西安电子科技大学学术信息网
如果您有学术信息或学术动态,欢迎投稿。我们将在第一时间确认并收录,投稿邮箱: meeting@xidian.edu.cn
如果您有学术信息或学术动态,欢迎投稿。我们将在第一时间确认并收录,投稿邮箱: meeting@xidian.edu.cn