Tensor Canonical Correlation Analysis Networks for Multi-View Remote Sensing Scene Recognition

Convolutional neural network (CNN) has been proven an effective way to extract high-level features from remote sensing (RS) images automatically. Many variants of the CNN model have proposed, including principal component analysis (PCANet), canonical correlation (CCANet), multiple scale CCANet (MS-CCANet) and multiview (MCCANet). The PCANet is specialized for single view feature abstraction, while in many real-world practices, RS data are frequently observed more views. Although CCANet, MS-CCANet MCCANet can be applied two or data, they consider only pair-wise by calculating a series two-order covariance matrices. However, high-order consistence, which explored collectively simultaneously examining all views, remains undiscovered. In this paper, we propose tensor (TCCANet) tackle problem. Particularly, TCCANet learns filter banks maximizing arbitrary number views with high-order-correlation solves optimization problem decomposing tensor. After convolutional stage, utilize binarization block-wise histogram strategies generate final feature. Furthermore, also develop Multiple Scale version TCCANet, i.e., MS-TCCANet, enriched representation incorporating previous layers. Numerical experiment results on RSSCN7 SAT-6 datasets demonstrate advantages MS-TCCANet scene recognition.