Extending the Unmixing methods to Multispectral Images

Abstract

In the past few decades, there has been intensive research concerning the Unmixing of hyperspectral images. Some methods such as NMF, VCA, and N-FINDR have become standards since they show robustness in dealing with the unmixing of hyperspectral images. However, the research concerning the unmixing of multispectral images is relatively scarce. Thus, we extend some unmixing methods to the multispectral images. In this paper, we have created two simulated multispectral datasets from two hyperspectral datasets whose ground truths are given. Then we apply the unmixing methods (VCA, NMF, N-FINDR) to these two datasets. By comparing and analyzing the results, we have been able to demonstrate some interesting result for the utilization of VCA, NMF, and N-FINDR with multispectral datasets. Besides, this also demonstrates the possibilities in extending these unmixing methods to the field of multispectral imaging.

Endmember Extraction for Pigment Identification Pre- and Post-intervention: A Case Study from a XVIth Century Copper Plate Painting

“…The endmembers (EM) sets differ for the three endmember extraction methods, and are not directly similar to the reference pigments (obtained with the auxiliary copper plate) in all cases, similarly to the findings in [7]. In Figure 3, the endmembers extracted from the painting pre-and post-intervention are shown.…”

“…Pigment identification from spectral imaging data has been extensively explored in recent years [1][2][3][4][5]. From simple approaches like identifying features of the first and secondderivative spectra [3,6] to using optimal spectral bands selected for a particular set of pigments [7] or applying Deep Learning algorithms [8][9][10][11], many perspectives and solutions have been proposed. In most cases, a reference collection of prepared patches is needed, and often they must be prepared specifically for the substrate used in the painting to be analyzed.…”

“…Then, spectral unmixing and endmember extraction techniques borrowed from the field of satellite imaging can be potentially helpful. They have been used in some previous studies [14,15] with variable degrees of success.…”

Bidirectional Recurrent Imputation and Abundance Estimation of LULC Classes With MODIS Multispectral Time-Series and Geo-Topographic and Climatic Data

“…Most previous works focus on HS or MS data and do not exploit temporal information to estimate the abundance of mixed pixels. Obtaining MS time series of large territories can be prohibitive due to the cost and time required to acquire them [85], [86]. In addition, no previous work has explored the possibility of adding ancillary data to enhance the spectral unmixing results, which are used successfully in other computer vision tasks [11], [12], [13].…”

Color and Gloss Measurements in Cultural Heritage Conservation Science: Recent Advances in France

“…Identification often needs other techniques like XRF or Raman to confirm the hypothesis [6]. Research on hyperspectral pigment identification actually focuses on accuracy of classification algorithms for complex cases (non flat objects, mixed pigments…) [7,8].…”