Row-Sparse Discriminative Deep Dictionary Learning for Hyperspectral Image Classification

Abstract

In recent studies in hyperspectral imaging, biometrics and energy analytics, the framework of deep dictionary learning has shown promise. Deep dictionary learning outperforms other traditional deep learning tools when training data is limited; therefore hyperspectral imaging is one such example that benefits from this framework. Most of the prior studies were based on the unsupervised formulation; and in all cases, the training algorithm was greedy and hence sub-optimal. This is the first work that shows how to learn the deep dictionary learning problem in a joint fashion. Moreover, we propose a new discriminative penalty to the said framework. The third contribution of this work is showing how to incorporate stochastic regularization techniques into the deep dictionary learning framework. Experimental results on hyperspectral image classification shows that the proposed technique excels over all state-of-the-art deep and shallow (traditional) learning based methods published in recent times.

Global Prototypical Network for Few-Shot Hyperspectral Image Classification

“…2). For the c-th spectral band, the output (Oc) of "Fuse 2" module is realized by the Equation (15). O= [O1, O2, O3, …, ON] is the output of SSAN for all the spectral bands (OcR 5×5 and OR 5×5×N ).…”

Hierarchical Discriminative Deep Dictionary Learning

“…Even though the FDDL proposal model makes great sense to learn discriminative dictionaries and coefficients, it has been outperformed by recent dictionary and representation-based methods [7], [22] in face recognition problems. The DDL proposal on the other hand provides deep representations but which are not discriminative or induce discriminability to some degree between different classes of interest, unlike in [19] and [20]. We propose to take advantage of both approaches in the sense of using both discriminative and deep representations.…”

“…The later work takes advantage of the forward and backward pass in the neural network to learn sparse coefficients and the dictionaries at each layer respectively. Supervision was introduced in a deep dictionary learning framework in [18] for classification, improving modern techniques, and this was subsequently extended to include discriminative penalties [19]. In [20] sparse coding layers are proposed with wide and slim dictionaries to induce discriminative features and clustered representations respectively, achieving competitive results in image classification.…”

Union of Class-Dependent Collaborative Representation Based on Maximum Margin Projection for Hyperspectral Imagery Classification

“…In contrast to traditional classification methods, deep learning can automatically learn complex features of HSIs with a large number of hierarchical layers [18]. However, training a deep network is quite expensive and requires a large number of training samples [19][20]. Additionally, SR-based classifier (SRC) which represents each tested pixel sparsely by a few labeled atoms via l0 or l1-normed regularization takes the low-rank characteristic of HSI [21] into consideration, and has been shown to improve HSIC [22].…”