Analysis of Attention Modules in Unfolding Tensor Rank Minimization-Based Pansharpening

We examine the effect of various attention modules on a low-rank tensor minimization model for pansharpening. First, the pansharpening problem is formulated as a low-rank tensor minimization task, integrating a detail injection term and an attention module to guide the model to focus on salient regions of the feature map obtained by detail injection. Then, the problem is solved using a deep unfolding network, where each stage updates the variables and the regularizer via closed-form solutions and learned deep networks. Experimental results show that a simple and parameter-free attention module outperforms the baseline model.