Hyperspectral Unmixing using Deep Learning

Abstract

In this chapter, the consistency and stability of the proposed unmixing methods were experimentally tested using different datasets. The performances of the proposed unmixing methods were investigated against different factor such as no. endmembers, loss function, robustness to noise, and speed up rate. To evaluate the behavior of the proposed linear unmixing method based deep convolutional autoencoder, different experiments were conducted on both syntactic and hyperspectral dataset. Experimental results indicate that the proposed method outperforms the other traditional and deep based unmixing methods. The results demonstrated the significant performance of the proposed DCAE and that it outperforms benchmark unmixing methods even in a noisy environment in terms of both Root Mean Square Error (RMSE) and Mean Square Error (MSE). The achieved results of the proposed DCAE in terms of mean absolute error were 0.0097, 0.001, 0.0141, and 0.0145 for Samson, Cuprite, Urban, and Jasper Ridge datasets, respectively. To evaluate the performance of proposed nonlinear unmixing method based on deep learning, Different experiments are conducted using syntactic and hyperspectral datasets to show the effectiveness of the proposed algorithm, and the performance was evaluated using both SAD and SID metrics. The experiments evaluated the performance in terms of accuracy assessment, weight initialization techniques, learning rate, and robustness to the noise. The results verified that the proposed autoencoder outperforms traditional endmember extraction algorithms in nonlinear cases. Also, we introduced the application of hyperspectral image unmixing algorithm on the Internet of things (IoT) environment. Finally, this chapter evaluates the proposed nonlinear unmixing method for hyperspectral band selection. Different experiment sets were conducted to study its 98