07/19/2022
By Dalila Megherbi

Chris Good will defend his master's of science thesis in Computer Engineering, titled “Analysis of the Effects of Wavelength Band Selection and Data Fusion Techniques on Multiple-modality Homeland Security Airborne Scenes Via Deep learning Models,” on Friday, July 29, 2022, 2022 at 1 p.m.

Location: This will be a virtual defense via Zoom. Those interested in attending should contact Committee Chair Dalila_Megherbi@uml.edu and chris_good@student.uml.edu at least 24 hours before the defense to request access to the meeting.

Committee Chair: D. B. Megherbi, ECE Department, (CMINDS), UML, (Thesis Advisor, Committee Chair)

Committee Members:

  • Kanti Prasad, ECE Department, UML
  • Xuejun Lu, ECE Department, UML

Abstract:

Remote sensing is an extremely useful field that allows individuals and organizations to collect information about objects or phenomena from afar, helping them improve their understanding of the world around them. Taken by means of satellite or aircraft, remote data collection often includes a suite of sensors such as optical cameras, LiDAR, hyperspectral imagers, and radar. Hyperspectral imagers are particularly interesting, with tens to hundreds of spectra across the electromagnetic (EM) spectrum captured in a contiguous range of narrow EM wavelength 'bands'. Hyperspectral sensing has useful applications in mineralogy, biology, environmental studies, and military defense and security. Due to the complexity of identifying land coverage or objects in remote sensing scenes, fusing data between different sets of sensors to improve the performance and interpretation of data have become an increasing area of interest. Hyperspectral imaging and data fusion are both areas of high importance to the military, with hyperspectral imagery providing the high discriminative ability to military targets of interest and data fusion improving classification performance and tactical decision making.

In this thesis, we study the problem of hyperspectral band selection in multimodal remote sensing scenes. Using a deep, dense convolutional network based on a three-dimensional variant of the DenseNet architecture, we present an analysis of improved sparse subspace clustering (ISSC) band selection algorithm and two data fusion strategies (early and late fusion), on the IEEE Geoscience and Remote Sensing Society (GRSS) 2018 data fusion contest urban land usage and land cover (LULC) dataset. Using this combination of dense network, band selection, data fusion, and training on a Google Colab Pro+ instance with an NVIDIA Tesla V100-SMX2 GPU, we demonstrate that we can exceed classification performance on the IEEE GRSS 2018 dataset over the national challenge contest first winners when ignoring ad-hoc post-processing methods. Furthermore, in our analysis, our work deeply explores the effects of data modality, band selection, and fusion methodology on model performance, with an aim to improve the starting point of future research using the challenging urban remote sensing datasets and the IEEE GRSS 2018 data fusion contest dataset or similar urban remote sensing airborne scenes, in particular, to show the potential value of the proposed methodology for remote sensing and homeland security applications.

All interested students, faculty members, and staff are invited to attend the online defense via remote access.