An Investigation in the Use of Hyperspectral Imagery Using Machine Learning For Vision Aided Navigation
Abstract
The goal of this research is to investigate the possible use of hyperspectral imagery in vision-aided navigation using machine learning. This paper is to show that hyperspectral can and should be used in vision-aided navigation and that it is more effective than traditional RGB images. The database of hyperspectral imagery used was NASAs AVIRIS, due to the availability of large amounts of data across the United States. The neural network that was trained and tested was convolutional neural network that uses a multi-scale filter bank and residual learning to improve the network. A second Network created, trained, and tested on RGB images was used to create comparison data to the results of the hyperspectral network. These neural networks were tested in two ways, the first being a full data test even though each of the classes had a considerable difference in data, and the second being a test that had an even size data between classes test. The results investigated the best hyper parameters that could have been used in both networks. The best hyperspectral network out performed the best RGB network on every test. This shows a strong potential of using hyperspectral images future research for vision aided navigation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 29, 2024
- Accession Number
- AD1219875
Entities
People
- Isaac T. Ege
Organizations
- University of Dayton