Multi-Geometry Atmospheric Correction and Target Spectra Retrieval from Hyperspectral Images Via Deep Learning PDF Download

Author: Fangcao Xu
Publisher:
ISBN:
Category :
Languages : en
Pages :

Get Book Here

Book Description
Atmospheric correction is a fundamental task in remote sensing because observations are taken either of the atmosphere or looking through the atmosphere. Atmospheric correction errors can significantly alter the spectral signature of the observations, and lead to invalid classifications or target detection. This is even more crucial when working with hyperspectral data, where a precise measurement of spectral properties is required. State-of-the-art physical approaches for atmospheric correction require extensive prior knowledge about sensor characteristics, collection geometry, and environmental characteristics of the scene being collected. These approaches are computationally expensive, prone to inaccuracy due to lack of sufficient environmental and collection information, and often impossible for real-time applications. Recently, artificial intelligence (AI) and advanced deep learning (DL) techniques have obtained great achievements in many research areas, such as target detection, image classification and segmentation, and spatiotemporal analysis. To take full advantage of remote sensing observation in quickly and reliably acquiring data for a large area, integrating AI with remote sensing and GIScience could provide an automatic and efficient processing tool and discover knowledge that has never been revealed from massive datasets. In this dissertation, I propose three major research topics to expand the solution of current remote sensing image analysis for full geometric diversity to exploit multi-scans hyperspectral images simultaneously and incorporate deep neural networks. Three studies are conducted with simulated and real-world collected hyperspectral images for a full spectrum analysis, ranging from (0.4 - 13.5 um). The first study investigates the longwave infrared spectrum on the simulated data to understand the impact of different solar and atmospheric radiative components on the at-sensor signature under various geometries. The goal is to develop and test a general deep learning solution for atmospheric correction and target detection using multiple hyperspectral scenes. The second study proposes a geometry-dependent hybrid neural network that implements the causality of different geometric factors into the network structure. This network is trained on two different longwave hyperspectral dataset, one simulated using MODTRAN, and the second observed using the Blue Heron instrument in a dedicated field study. The third study focuses on the visible, near infrared and shortwave infrared spectrum, to improve the time-dependency of the network and represent the seasonal and diurnal characteristics of atmosphere and solar radiance. The main contributions of this dissertation are: 1) it makes use of the computer ability with new innovative AI methods and multi-scan hyperspectral data, which can better learn the non-linear relationship and complex interactions between atmosphere and different radiative components passing through it, and 2) it enhances the current state-of-the-science in hyperspectral remote sensing research and drives future hyperspectral sensor performance requirements and concepts of atmospheric characterization and target detection operations.