Multi-scale dense spatially-adaptive residual distillation network for lightweight underwater image super-resolution

• DOI: 10.3389/fmars.2023.1328436
• Publication Date: 2024-01-05T04:37:52Z
• Authors: Bingzan Liu; Xin Ning; Shichao Ma; Yizhen Yang
• Keywords: Feature (linguistics)

Underwater images are typically of poor quality, lacking texture and edge information, blurry full artifacts, which restricts the performance subsequent tasks such as underwater object detection, path planning for unmanned submersible vehicles (UUVs). Additionally, limitation equipment, most existing image enhancement super-resolution methods cannot be implemented directly. Hence, developing a weightless technique improving resolution submerged while balancing parameters is vital. In this paper, multi-scale dense spatially-adaptive residual distillation network (MDSRDN) proposed aiming at obtaining high-quality (HR) with odd fast running time. particular, module (MDSRD) developed to facilitate global-to-local feature extraction like multi-head transformer enriching spatial attention maps. By introducing layer (SFT layer) spatial-adaptive (RSFA), an enhancing map modulation generated. Furthermore, maintain lightweight enough, blue separable convolution (BS-Conv) applied. Extensive experimental results illustrate superiority MDSDRN in reconstruction, can achieve great balance between (only 0.32M), multi-adds 13G), (26.38 dB on PSNR USR-248) scale <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" id="im1"><mml:mrow><mml:mo>×</mml:mo><mml:mn>4</mml:mn></mml:mrow></mml:math> .