|
Transferred Multi-Perception Attention Networks for 8 M: u$ Q) P G" B' u
Remote Sensing Image Super-Resolution 2 ~5 C& b3 P' ~# Z$ e- d
( o: p1 p1 B3 E- v
& Z/ v! O$ k! w) V# I# T7 i' Z
' K" C1 r4 Q# T p: Q2 }- K" `; J3 {% [/ y
Image super-resolution (SR) reconstruction plays a key role in coping with the increasing. C, x" R; Z4 b; U5 J) K" ^
demand on remote sensing imaging applications with high spatial resolution requirements. Though) a2 N! u8 @0 y" {) F
many SR methods have been proposed over the last few years, further research is needed to improve" N3 w! |+ t. c9 K, c1 M: O* M
SR processes with regard to the complex spatial distribution of the remote sensing images and the3 z& k5 s4 f2 o: K, P l- s4 U
diverse spatial scales of ground objects. In this paper, a novel multi-perception attention network7 J, L* X$ v, F }* F5 H- x
(MPSR) is developed with performance exceeding those of many existing state-of-the-art models.# W7 y* W! T8 N, x
By incorporating the proposed enhanced residual block (ERB) and residual channel attention group8 \/ H6 z* Q3 @$ E7 `; I4 E
(RCAG), MPSR can super-resolve low-resolution remote sensing images via multi-perception learning
2 W; c% n& z8 b' [1 Y! _7 E6 H' T9 yand multi-level information adaptive weighted fusion. Moreover, a pre-train and transfer learning' q3 o" N$ w1 I
strategy is introduced, which improved the SR performance and stabilized the training procedure.5 R! m# ~5 E4 H8 g1 p' w0 Y7 @' Y) S
Experimental comparisons are conducted using 13 state-of-the-art methods over a remote sensing! F6 o# D! Z' b7 b. s! c7 F
dataset and benchmark natural image sets. The proposed model proved its excellence in both objective
3 P \: G" H9 Q; U7 ^2 ]criterion and subjective perspective.
/ E! E( H: V. E/ F! A
+ a- f& i9 `* f) f% s; V" [# R6 I: V/ e: `
|