An Efficient Ontology Based Machine Learning For Classifying High Definition Satellite Images

Abstract

The analysis of satellite images to determine the kind of land, forest cover, plant type, and
other aspects of a particular image has become a standard practice. Object-based image
classification for land-cover mapping with remote-sensing data has received a lot of interest
in recent years. Several researches have looked into a range of sensors, feature extraction,
classifiers, and other important elements throughout the last decade. These data, however,
have yet to be gathered into a complete reference on the implications of different guided
object-based land-cover categorization methods. To solve this issue, we provide an ontologybased
conceptual UKNN classification algorithm in this study. We create an 18-field
database from the rit18 database's qualitative and quantitative data in this research. Instead of
unexplainable characteristics provided by a CNN, this proposed study uses an ontology-based
methodology framework to enable picture classification using features extracted from hyper
spectral image data. The image is initially given a Gaussian filter. The suggested UNet K
nearest neighbours (UKNN) classifier is utilised to translate the ontology components into
image-based parameters, which are then used to improve the classification process. When a
typical support vector machine (SVM) is compared to UKNN, we find that UKNN
outperforms SVM in classification accuracy.