Deakin University
Browse

File(s) under embargo

MC-NET: Spectral-Spatial Feature Reduction for Hyperspectral Image Classification with Optimized Technique Series

conference contribution
posted on 2024-03-13, 01:01 authored by MT Islam, M Kumar, MR Islam
The use of hyperspectral images (HSI) in a variety of practical applications has become widespread. High correlations between and among classes, the curse of dimensionality, and overfitting have drawn attention. Deep learning-based algorithms with dimensionality reduction methods have been proposed as a feasible alternative for hyperspectral image analysis owing to their effective feature extraction and high overall performance. Principal component analysis (PCA) and Sparse PCA (SPCA) are frequently employed in the fields of dimensionality reduction. Since principal component analysis (PCA) is hard to interpret because each principal component is a linear mixture of the original variables and SPCA cannot select the most important features. Dimensionality reduction via SPCA-MIFS (mutual information feature selection) is utilized to address the problem. The 2D and 3D Convolutional Neural Network (CNN) methods provide the backbone of many of the deep learning combination techniques. A fused 3D-2D CNN makes the model less complicated than a 3D-CNN alone, and it can also work well in noisy environments and when there aren't many training samples. Overfitting is addressed and classification accuracy is enhanced by using several optimization techniques. To evaluate the efficacy of this proposed strategy performance results compared highly reputed models with the less numbers of characteristics.

History

Volume

00

Pagination

1-4

Location

Rajshahi, Bangladesh

Start date

2022-12-29

End date

2022-12-31

ISBN-13

9798350320541

Language

eng

Publication classification

E1.1 Full written paper - refereed

Title of proceedings

4th International Conference on Electrical, Computer and Telecommunication Engineering, ICECTE 2022

Event

2022 4th International Conference on Electrical, Computer & Telecommunication Engineering (ICECTE)

Publisher

IEEE

Place of publication

Piscataway, NJ.

Usage metrics

    Research Publications

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC