Hyper-parameter Comparison on Convolutional Neural Network for Visual Aerial Localization

J. Mark Berhold, Robert C. Leishman, Brett Borghetti, Donald Venable

Abstract: Global Positioning System (GPS) is the default modern application used for geospatial location determination. Unfortunately, the signals from all Global Navigation Satellite Systems (GNSS) to include GPS can be jammed or spoofed. Visual aerial localization can be used to augment navigation in these instances. This paper will discuss the performance of various Convolutional Neural Network (CNN) hyper-parameters to address visual aerial localization. The localization CNNs are trained and tested through satellite imagery of a localized area of 150 square kilometers. Three hyper-parameters of focus are: initializations, optimizers, and finishing layers. This paper shows that specific initializations, optimizations and finishing layers can have significant effects on the training and performance of a CNN. Selecting best combinations based of the study decreased mean test set error by 358 meters.
Published in: Proceedings of the ION 2019 Pacific PNT Meeting
April 8 - 11, 2019
Hilton Waikiki Beach
Honolulu, Hawaii
Pages: 875 - 885
Cite this article: Berhold, J. Mark, Leishman, Robert C., Borghetti, Brett, Venable, Donald, "Hyper-parameter Comparison on Convolutional Neural Network for Visual Aerial Localization," Proceedings of the ION 2019 Pacific PNT Meeting, Honolulu, Hawaii, April 2019, pp. 875-885.
https://doi.org/10.33012/2019.16846
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In