Inception residual attention network for remote sensing image super-resolution

杨利霞

Inception residual attention network for remote sensingimage super-resolution

5 h1 K# K+ c6 @1 o0 ^" i
& z* R6 v5 E' m; s0 sHow to enhance the spatial resolution for a remote sensing image is
an important issue that we face. Many image super-resolution (SR)
techniques have been proposed for this purpose and deep con
volutional neural network (CNN) is the most effective approach in
) X3 `' d5 }: c8 Frecent years. However, we observe that most CNN-based SR meth
- Y4 r% i! E5 r# qods treat low-frequency areas and high-frequency areas equally,
hence hindering the recovery of high-frequency information. In this
paper, we propose a network named inception residual attention
2 \6 _9 v- t# N9 Y5 inetwork (IRAN) to address this problem. Specifically, we propose
  `( {9 H; b6 o" u1 za spatial attention module to make the network adaptively learn
the importance of different spatial areas, so as to pay more atten
/ Q, p9 W0 k6 ^% F0 e  {( Mtion to the areas with high-frequency information. Furthermore, we
9 \$ [8 A9 J% Z: y3 Cpresent an inception module to fuse local multilevel features, so as
2 r0 C) @' N" H/ oto provide richer information for reconstructing detailed textures. In
* I3 ]( c/ ]$ X- f  eorder to evaluate the effectiveness of the proposed method, a large
" `6 v  U& @  M( ~& x8 znumber of experiments are performed on UCMerced-LandUse data
0 G+ b3 m2 E: e" ?) u- ~set and the results show that the proposed method is superior to
the current state-of-the-art methods in both visual effects and
8 X1 o) {8 I# L# Eobjective indicators.


6 i! t: u' i; q8 x7 K2 C) D+ K

2 f9 Z3 v8 i3 c& [! G/ v9 H