Supervised Learning Independent Component Analysis Algorithms and Applications
Xianchuan Yu, Dan Hu, Jindong Xu · 2014
Considering the drawback of traditional ICA, we propose a new algorithm, supervised learning independent component analysis (SL-ICA) to solve the problem of mixed pixels in synthetic aperture radar (SAR) images. Adding supervised learning restrictive conditions to the negentropy objective function, we constrain the negentropy and restrictive conditions in a unified objective function and optimize the objective function by applying a new dual-gradient descent algorithm iteratively, which accelerates the computation speed. Results from applying SL-ICA, principal component analysis (PCA), and ICA to computer simulated SAR images and ENVISAT-ASAR images of Beijing show that SL-ICA yields more precise results than PCA and ICA.