Transferred Multi-Perception Attention Networks for
7 C' C' m: m2 t' Z; x+ o% FRemote Sensing Image Super-Resolution
q0 `$ p6 Y) {6 U0 z4 J6 Z m- q$ n! q. n
8 C, v0 U7 }, Y7 d& f3 ]- P _$ s
+ D z" P$ Y# ^4 Q$ b5 UImage super-resolution (SR) reconstruction plays a key role in coping with the increasing
' {) R4 i: X2 u, Xdemand on remote sensing imaging applications with high spatial resolution requirements. Though( L a( g- a' |8 |! H
many SR methods have been proposed over the last few years, further research is needed to improve! {% I' P0 ^; p4 A& y1 \
SR processes with regard to the complex spatial distribution of the remote sensing images and the$ S. C7 r0 ~/ b7 e- L' V
diverse spatial scales of ground objects. In this paper, a novel multi-perception attention network
0 f! u% G# k5 I2 |4 C(MPSR) is developed with performance exceeding those of many existing state-of-the-art models.2 W1 h, F; Y& o% H0 ]6 }
By incorporating the proposed enhanced residual block (ERB) and residual channel attention group! b. ]! b; z, B! j, e' \# s6 n
(RCAG), MPSR can super-resolve low-resolution remote sensing images via multi-perception learning
, ]5 z3 P9 |+ K) I: mand multi-level information adaptive weighted fusion. Moreover, a pre-train and transfer learning$ m( _8 s, z0 W# k9 M9 u. o
strategy is introduced, which improved the SR performance and stabilized the training procedure.4 @6 h! k5 @' G) d- q
Experimental comparisons are conducted using 13 state-of-the-art methods over a remote sensing7 V# ~5 f% {) P6 f- t# j/ L, B$ ]$ n
dataset and benchmark natural image sets. The proposed model proved its excellence in both objective% P+ R$ ~1 ?5 M3 b. p2 H
criterion and subjective perspective.
& s0 Q1 R' ^: H2 u$ g: ]; E6 D7 P. @' y3 ]. F' W' Q
# E4 G2 E& A$ d6 H& S! Z- \/ N# r |