End-to-End Convolutional Autoencoder for Nonlinear Hyperspectral Unmixing

Abstract

Hyperspectral Unmixing is the process of decomposing a mixed pixel into its pure materials (endmembers) and estimating their corresponding proportions (abundances). Although linear unmixing models are more common due to their simplicity and flexibility, they suffer from many limitations in real world scenes where interactions between pure materials exist, which paved the way for nonlinear methods to emerge. However, existing methods for nonlinear unmixing require prior knowledge or an assumption about the type of nonlinearity, which can affect the results. This paper introduces a nonlinear method with a novel deep convolutional autoencoder for blind unmixing. The proposed framework consists of a deep encoder of successive small size convolutional filters along with max pooling layers, and a decoder composed of successive 2D and 1D convolutional filters. The output of the decoder is formed of a linear part and an additive non-linear one. The network is trained using the mean squared error loss function. Several experiments were conducted to evaluate the performance of the proposed method using synthetic and real airborne data. Results show a better performance in terms of abundance and endmembers estimation compared to several existing methods.

Spatial Validation of Spectral Unmixing Results: A Systematic Review

“…It is important to emphasize that 97% of these reference maps are available online together with hyperspectral images and/or reference spectral libraries (e.g., [532-535] Figure 8). Therefore, these images were well known: Cuprite (Nevada, USA, e.g., [70,458]), Indian Pines (Indiana, USA, e.g., [78,458]), Jasper Ridge (California, USA, e.g., [68,97]), Salinas Valley (California, USA, e.g., [75,78]) datasets that were acquired with AVIRIS sensors; Pavia (Italy, e.g., [81,85]) datasets that were acquired with the ROSIS sensor; Samson (Florida, USA, e.g., [59,89]) dataset that was acquired with the Samson sensor; University of Houston (Texas, USA, e.g., [59,78],) dataset that was acquired with the CASI-1500 sensor ; Urban (Texas, USA, e.g., [59,68]) and Washington DC Mall (Washington DC, USA, e.g., [81,90]) datasets that were acquired with the HYDICE sensor. As regards the papers that analyzed the multispectral data, most of the authors chose to create the reference maps from the other images, whereas most of the authors that analyzed the hyperspectral data chose to employ the previous reference maps.…”

Modified Dynamic Routing Convolutional Neural Network for Pan-Sharpening

“…Due to its broad applications, remote sensing image processing has become an active field in computer vision. Based on different application scenarios, there are many representative research directions, such as hyperspectral image classification [1][2][3][4][5][6][7][8][9], estimation of the number of endmembers [10,11], hyperspectral unmixing [12,13] and pansharpening [14][15][16]. Pan-sharpening mainly fuses the information of LRMS images and PAN images to obtain high spatial resolution multi-spectral (HRMS) images, which contain the rich spectral information from LRMS images and the spatial details of PAN images.…”

An Improved Hyperspectral Unmixing Approach Based on a Spatial–Spectral Adaptive Nonlinear Unmixing Network

“…In addition, by combining high-order nonlinear interaction of photons with AE, additional methods have been developed [41][42][43] to simulate the non-linear mixture of photons when incident to ground features. For example, Dhaini et al [44] replaced linear layers with convolutional layers to simulate the nonlinear effects of photons and achieved good results.…”