A Pre-processing framework for spectral classification of hyperspectral images
- PDF / 4,093,425 Bytes
- 19 Pages / 439.642 x 666.49 pts Page_size
- 23 Downloads / 250 Views
A Pre-processing framework for spectral classification of hyperspectral images Simranjit Singh1 · Singara Singh Kasana2 Received: 12 September 2018 / Revised: 21 April 2020 / Accepted: 4 June 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Classification of Hyperspectral images is mostly based on the spectral-spatial features in existing classification techniques. The captured Hyperspectral images from satellites may contain some noisy bands due to water absorption. The process of radiometric and atmospheric corrections leads to the removal of useful bands present in the acquired HSI. In this paper, a novel framework is proposed in which interpolation is used to accommodate the loss of noisy bands. Further, the extraction of hybrid features is performed using PCA and LPP to preserve spatial information, and these features are passed as input to the machine learning models. The proposed framework is compared with the existing spectralspatial and spectral based frameworks by using the standard datasets-Indian Pines, Salinas, Pavia University, and Kennedy Space Centre. The accuracy of the classification is increased significantly when the proposed framework is blended with state-of-art classifiers. Keywords Hyperspectral images · PCA · LPP · SVM · Classification · LSTM
1 Introduction Spectral imaging is a branch of spectroscopy in which a complete spectral feature is available in the image in the form of pixels. Various devices based on spectrometry can capture the spectral data. A hyperspectral image(HSI) contains hundreds of bands spread across the electromagnetic spectrum due to which it can provide more information in comparison to an RGB image. Hyperion (NASA), Bhuvan (ISRO) etc are satellites that can capture hyperspectral images by using hyperspectral cameras/sensors. Multiple bands are obtained over electromagnetic spectra or wavelength to form an HSI. Simranjit Singh
[email protected] Singara Singh Kasana [email protected] 1
Department of Computer Science and Engineering, Bennett University, Greater Noida, Uttar Pradesh, India
2
Computer Science and Engineering Department, Thapar Institute of Engineering and Technology Patiala, Patiala, Punjab, India
Multimedia Tools and Applications
HSIs have high spectral resolution due to which they are used in a range of applications like the mapping of mineral, forest and vegetation [8, 10, 14, 25, 29], chemical imaging [3], classification of land cover, habitat [17] and tree ages [31], anomaly detection [27] etc. However, the mapping of minerals using HSI is quite a complicated task due to spectral mixing. To properly differentiate the minerals from the spectrum of HSI, different spectral unmixing algorithms [13] are available in the literature. The most popular algorithms include the non-negative matrix factorization (NMF) model, sparse regression model, and deep matrix factorization [24]. Literature indicates earlier scientists have given preference to the spectral aspect of the Hyperspectral data for variou
Data Loading...
