Inception residual attention network for remote sensing image super-resolution

杨利霞

Inception residual attention network for remote sensingimage super-resolution

; C$ t$ T. U* aHow to enhance the spatial resolution for a remote sensing image is
- t  }; }: X7 [' e& ian important issue that we face. Many image super-resolution (SR)
1 ^" S! j$ q0 w6 x/ G+ stechniques have been proposed for this purpose and deep con
volutional neural network (CNN) is the most effective approach in
recent years. However, we observe that most CNN-based SR meth
# M) X' y. t( N+ T/ e* V# Yods treat low-frequency areas and high-frequency areas equally,
6 w/ T$ B/ d6 D+ m" P3 x6 W1 Ohence hindering the recovery of high-frequency information. In this
" E4 _' }1 L, c5 v; Ypaper, we propose a network named inception residual attention
, q. h! s& f3 {+ `  H9 fnetwork (IRAN) to address this problem. Specifically, we propose
a spatial attention module to make the network adaptively learn
the importance of different spatial areas, so as to pay more atten
, o4 X* l  \: J% }tion to the areas with high-frequency information. Furthermore, we
present an inception module to fuse local multilevel features, so as
, ?' Q# z$ P1 ^! X$ o5 I" Cto provide richer information for reconstructing detailed textures. In
2 r8 e8 w2 e$ n1 ~+ c* Sorder to evaluate the effectiveness of the proposed method, a large
% ^& u* K* W" ]+ qnumber of experiments are performed on UCMerced-LandUse data
; v; ^: c8 O9 \/ P% tset and the results show that the proposed method is superior to
& p) |- y. B5 E2 a/ X, G. u' l( ^, tthe current state-of-the-art methods in both visual effects and
  w* N3 S" F& a3 t- ^4 oobjective indicators.
6 a, x9 ?, O1 ]& c! b5 Q. L

% W- I- \9 O' r: {1 [8 ^
0 i: V- L, [2 [6 [0 f, f) ]
$ a* y, `. r4 f% j