Inception residual attention network for remote sensing image super-resolution

杨利霞

Inception residual attention network for remote sensingimage super-resolution

8 c! e: }; v1 Q. L- i
  t( C# |' ^4 d
How to enhance the spatial resolution for a remote sensing image is
$ p5 i, |4 d9 m; san important issue that we face. Many image super-resolution (SR)
- J8 @! D3 Y% p5 Btechniques have been proposed for this purpose and deep con
volutional neural network (CNN) is the most effective approach in
& N, g3 \& q2 t% Srecent years. However, we observe that most CNN-based SR meth
ods treat low-frequency areas and high-frequency areas equally,
hence hindering the recovery of high-frequency information. In this
0 e7 [2 o5 [9 |2 u) ?( }0 u- gpaper, we propose a network named inception residual attention
network (IRAN) to address this problem. Specifically, we propose
! l9 u0 S; J/ C4 ia spatial attention module to make the network adaptively learn
the importance of different spatial areas, so as to pay more atten
tion to the areas with high-frequency information. Furthermore, we
# N0 |7 K; h0 J1 o& G# x7 l8 Rpresent an inception module to fuse local multilevel features, so as
to provide richer information for reconstructing detailed textures. In
order to evaluate the effectiveness of the proposed method, a large
2 R# ]" s( Q: V+ R) [! i3 Gnumber of experiments are performed on UCMerced-LandUse data
* U* v! v' K+ T; Tset and the results show that the proposed method is superior to
! d6 x6 ?; y" {the current state-of-the-art methods in both visual effects and
* W1 B0 X. S6 q- a, F0 Jobjective indicators.
- M- ]' o9 a1 _0 D5 k

  K* h. P5 Z; D6 @0 L! }/ H" S
6 k2 m8 W7 w3 m5 {% p
+ w1 f4 k: a$ Q' o! F# |