ResearchBib Share Your Research, Maximize Your Social Impacts
Sign for Notice Everyday Sign up >> Login

A HYBRID DEEP LEARNING APPROACH FOR GROUND OBJECT INFORMATION EXTRACTION FROM HYPERSPECTRAL IMAGES

Journal: International Journal of Advanced Research (Vol.12, No. 08)

Publication Date:

Authors : ; ;

Page : 359-381

Keywords : Ground Object Information Hyperspectral Images Wingsuit Spotted-Hyena Optimizer Modified Bi-LSTM;

Source : Downloadexternal Find it from : Google Scholarexternal

Abstract

Ground object information extraction out of hyperspectral images refers to the process of identifying and characterizing different materials or objects present on the earths exterior using hyperspectral imaging technology. Hyperspectral imaging captures images in hundreds of small and adjacent spectral bands, providing more detailed and accurate information about the materials present in the scene. Hyperspectral data is high-dimensional, which can make it challenging to visualize and analyze. This can make it difficult to identify patterns or trends in the data, and can also make it challenging to interpret the results of classification algorithms. In this research work, collected raw images (GIS images) are pre-processed via Augmented median filtering and CLAHE approach. From the pre-processed images, the features like Haralick features, Normalized Difference Vegetation Index (NDVI), Standardized Vegetation Index (SVI), correlation coefficient, Green-NDVI (GNDVI), Simple Ratio (SR), and Red-edge Simple Ratio (SRre) based features are extracted. Among the extracted features, the optimal features are preferred utilizing a new hybrid optimization model stated to as Wingsuit Spotted Hyena Optimizer (WSHO) model. This WSHO model is the conceptual blend of standard Wingsuit Flying Search (WFS) and Spotted Hyena Optimizer (SHO) Mineral Exploration phase is modelled with a three-fold-deep-learning-classifiers framework (proposed). The threefold-deep-learning-classifiers framework encapsulates the Support Vector Machine (SVM), Random Forest (RF), and modified Bi-LSTM. SVM and RF are trained with the selected optimal features acquired from WSHO. The outcome from SVM and RF is fed as input to Modified Bi-LSTM. To further enhance the detection accuracy, the loss function of Bi-LSTM is modified. The proposed model is implemented by PYTHON and validated in terms of various performance metrics.

Last modified: 2024-09-14 21:06:31