Hyperspectral imaging is a highly viable method in which rock delineation can be performed with minimal classification error. Though previous studies based on this method yield high accuracy results, it is often reported that the computational requirements necessary to process data of this magnitude are high due to the hundreds of spectral bands, hence making field applicability rather challenging. To counter this problem with field applicability in mind, this study proposes a method of employing Neighbourhood Component Analysis (NCA) as a dimensional reduction tool, which, based on feature weights, determines the most important spectral bands which can in turn be used in delineating rocks based on their spectral signatures. From our hyperspectral images database, we attempted to reduce each spectral anomaly from their original resolution of 204 bands captured using our hyperspectral camera; which has a spatial range within the Visible-Near-Infrared Range (VNIR) of 400-1000nm. To achieve this, we performed NCA which assigns feature weights to each one of the 204 spectral bands, thereby allowing the user to select the number of spectral bands they would subsequently prefer to employ in field applicable classification and identification problems, which in our case was 5 bands. Tied into one task, our algorithm then automatically performs classification of these rocks based on the 5 chosen spectral bands with the highest feature weights. Our model was able to achieve a validation accuracy of 70.9%, and an average per-class precision of 72%, hence, proving that high accuracy rock classifications can indeed be performed on dimensionally reduced spectral bands. It can therefore be said that advantages of using NCA include; low computational requirements, shorter classification times, and most importantly, the ability to reduce the dimensions of hyperspectral layers, thereby making it possible to accurately classify rocks with fewer spectral bands.
Skip Nav Destination
ISRM International Workshop on Rock Mechanics and Engineering Geology in Volcanic Fields
September 9–11, 2021
Fukuoka, Japan
Employing NCA as a Band Reduction Tool in Rock Identification from Hyperspectral Processing
Elisha Shemang;
Elisha Shemang
Botswana International University of Science and Technology
Search for other works by this author on:
Paper presented at the ISRM International Workshop on Rock Mechanics and Engineering Geology in Volcanic Fields, Fukuoka, Japan, September 2021.
Paper Number:
ISRM-IWRMEGV-2021-36
Published:
September 09 2021
Citation
Sinaice, Brian B., Owada, Narihiro, Utsuki, Shinji, Bagai, Zibisani, Shemang, Elisha, and Youhei Kawamura. "Employing NCA as a Band Reduction Tool in Rock Identification from Hyperspectral Processing." Paper presented at the ISRM International Workshop on Rock Mechanics and Engineering Geology in Volcanic Fields, Fukuoka, Japan, September 2021.
Download citation file:
Sign in
Don't already have an account? Register
Personal Account
You could not be signed in. Please check your username and password and try again.
Could not validate captcha. Please try again.
Pay-Per-View Access
$20.00
Advertisement
1
Views
Advertisement
Suggested Reading
Advertisement