Image Denoising of Fishing Nets Underwater with Range Selection Based on Convolutional Sparse Coding

The underwater detection imaging technology mainly includes sonar imaging and underwater optical imaging. In order to overcome the disadvantages of low resolution of sonar imaging and short detection range of traditional optical imaging, optical gating technology is applied to underwater fishing net...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:2023 4th International Symposium on Computer Engineering and Intelligent Communications (ISCEIC) s. 77 - 80
Hlavní autoři: Wu, Sheng, Liu, Chao, Zhang, Honglei, Duan, Enyue, Jiang, Nian
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 18.08.2023
Témata:
component
Convolution
Convolutional codes
Convolutional sparse coding
Feature extraction
Image coding
Image denoising
Image resolution
Images of underwater fishing nets
Optical imaging
Residual learning
Underwater range gating
Visual effects
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The underwater detection imaging technology mainly includes sonar imaging and underwater optical imaging. In order to overcome the disadvantages of low resolution of sonar imaging and short detection range of traditional optical imaging, optical gating technology is applied to underwater fishing net detection. This technology has higher resolution compared to sonar imaging, and its detection range has also been shown to be three times larger than that of conventional optical imaging. However, underwater range gating imaging can suffer from large random noise and low signal-to-noise ratio. In order to solve this problem, this paper proposes a convolutional sparse coding denoising neural network based on deep learning to improve image quality. This network consists of a local residual channel and a global convolutional sparse coding channel. Compared with DnCNN, ECNDNet and other methods, the average Peak signal-to-noise ratio of the proposed method is improved by 1.57~2.11dB, which can effectively remove random noise and reconstruct clear underwater fishing net range-gated images.
DOI:10.1109/ISCEIC59030.2023.10271234