7661_2_art_1_rs9mr3.pdf (3.19 MB)
Download file

Underwater motion deblurring based on cascaded attention mechanism

Download (3.19 MB)
journal contribution
posted on 07.07.2022, 08:49 authored by T Li, S Rong, L Chen, Huiyu Zhou, B He

The  images  captured  in  the  underwater  scene  frequently  suffer  from  blur  effects  due  to  the  insufficient light  and  the  relative  motion  between  the  captured  scenes  and  the  imaging  system,  which  severely  hinders  the visual-based  exploration  and  investigation  in  the  ocean.  In  this  paper,  we  propose  a  feature  pyramid  attention network  (FPAN)  to  remove  themotion  blur  and  restore  the  blurry  underwater  images.  FPAN  incorporates  the cascaded  attention  modules  into  the  feature  pyramid  network  (FPN)  that  enables  it  to  learn  more  discriminative information. To facilitate the training of FPAN, we construct a weighted loss function, which consists of a content loss, an adversarial loss, and a perceptual loss. The cascaded attention module and the weighted loss function enable our proposed FPAN to generate more realistic high-quality images from the blurry underwater images. In addition, to deal with the lack of publicly available datasets in underwater image deblurring, we built two specific underwater deblurring  datasets,  namely  Underwater  Convolutional  Deblurring  Dataset  (UCDD)  and  Underwater  Multi-frame AveragingDeblurring Dataset (UMADD), to train and examine different deep learning-based networks.Finally, we conduct  sea  trial  experiments  on  our  autonomous  underwater  vehicle  (AUV).  Experimental  results  on  two underwater  deblurring  datasets  demonstrate  our  proposed  method  achieves  satisfactory results, which  validates  the potential practical values of our proposed method in real-world applications.

History

Author affiliation

School of Computing and Mathematical Sciences, University of Leicester

Version

AM (Accepted Manuscript)

Published in

IEEE Journal of Oceanic Engineering

Publisher

Institute of Electrical and Electronics Engineers

issn

0364-9059

Acceptance date

05/07/2022

Copyright date

2022

Available date

07/07/2022

Publisher DOI

Language

en