Classification Endmember Selection with Multi-Temporal Hyperspectral Data

Abstract

In hyperspectral image classification, so-called spectral endmembers are used as reference data. These endmembers are either extracted from an image or taken from another source. Research has shown that endmembers extracted from an image usually perform best when classifying a single image. However, it is unclear if this also holds when classifying multi-temporal hyperspectral datasets. In this paper, we use spectral angle mapper, which is a frequently used classifier for hyperspectral datasets to classify multi-temporal airborne visible/infrared imaging spectrometer (AVIRIS) hyperspectral imagery. Three classifications are done on each of the images with endmembers being extracted from the corresponding image, and three more classifications are done on the three images while using averaged endmembers. We apply image-to-image registration and change detection to analyze the consistency of the classification results. We show that the consistency of classification accuracy using the averaged endmembers (around 65%) outperforms the classification results generated using endmembers that are extracted from each image separately (around 40%). We conclude that, for multi-temporal datasets, it is better to have an endmember collection that is not directly from the image, but is processed to a representative average.

Examining long-term natural vegetation dynamics in the Aral Sea Basin applying the linear spectral mixture model

“…Generally speaking, all kinds of endmember extraction algorithms are still in the exploratory stage, and each algorithm has its own advantages and disadvantages, which needs further improvement. Here, we tended to choose the final elements with a basis to improve the accuracy of the classification results, and visual interpretation is also widely used to verify the accuracy of machine classification ( Yu et al, 2017 ), so visual interpretation was selected for endmember selection ( Feng et al, 2019 ; Jiang, Van der Werff & Van der Meer, 2020 ). The specific experimental steps were (1) MOD13A1 was overlapped with Landsat 7 and Google Earth images; (2) the pure ground object pixel of Landsat 7 and Google Earth images were interpreted and recognized visually; and (3) the pixel values in MOD13A1 at the same position as the pure ground object pixel were taken as the end element values.…”

Multi-Temporal Hyperspectral Classification of Grassland Using Transformer Network

“…Many studies of grassland classification have been carried out worldwide using HSI technology [ 3 , 20 , 21 ]. Multi-temporal hyperspectral data refer to the time series of HSI, which allow for the monitoring of vegetation evolution over the course of a growing season or between years [ 10 , 22 ].…”

The effects of changing laboratory illumination zenith angle on spectral feature parameters and mineral classification

“…Longhi et al, 2001;Guha et al, 2012a) and has become aware of the advantage of using time series data (e.g. Dlamini and Xulu, 2019;Jiang et al, 2020;Gao et al, 2021), where illumination conditions may vary between scenes.…”