GAPANet: group alternate perceived attention network for optical imaging infrared thermal radiation effect correction

Abstract

To address the problem of insufficient information interaction between different dimensions of image features, and the inability of real-time simultaneously in the current thermal radiation effect correction methods, in this paper, we construct a group alternate perceived attention network. At coarser scales, the group alternate perceived attention module is constructed, which requires only fewer parameters and computational complexity. The group alternate perceived attention module mainly consists of two alternating parts: a global attention block combined with the local feature of three-branch dilated convolution, and a channel attention block for local information interaction. At the finest scale, a dimensional information interaction attention module is constructed. The correction network fully realizes attention perception at the inter-dimensional level globally and channel level locally. The experimental results of a large number of simulated and actual degraded images show that our model corrects better and realizes real-time.