Optimal trained ensemble of classification model for satellite image classification
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Satellite image classification is the most significant remote sensing method for computerized
analysis and pattern detection of satellite data. This method relies on the image’s diversity struc
tures and necessitates a thorough validation of the training data using the chosen classification
algorithm. The classifying of remotely sensed images refers to the process of classifying pixels
into a fixed set of distinct classes based on their data values. The pixel is categorized into that
class if it complies with a specific set of requirements. A new SIC scheme is suggested in this
work, where, Contrast limited adaptive histogram equalization (CLAHE) is deployed for pre
processing the image. Further, statistical features, Vegetation Indices (VI) and Improved Hybrid
Vegetation Index (IHVI) are extracted from the preprocessed image. Then, we deploy 2 phase
classification process for classifying the image. In the first phase, Bi-Directional Gated Recurrent
Unit (BI-GRU), Convolutional Neural Network (CNN) and Deep Max Out (DMO) are used. In
2nd phase, optimized RNN is used for determining the final classification result. Also, the train-
ing of the model is done by the new Chaotic Opposition Behavior Improvised Coot Algorithm
(COBI-CA) optimization via tuning the optimal weights. The COOT method has been enhanced
and is now the suggested algorithm. Lastly, we conduct an analysis to compare the suggested
Satellite Image Classification (SIC) model’s performance to the traditional approaches in terms
of statistical analysis, precision, accuracy, and error measure. In comparison to other conven
tional methods, the accuracy of the suggested model is 97%.