DALightNet achieves real time rain and haze removal on edge devices

Background: Rain and haze degrade image visibility and contrast, while computationally intensive restoration methods remain challenging to deploy on resource-constrained edge devices. Aims: This study proposes DA-LightNet, a lightweight image restoration framework designed to balance restoration quality and computational efficiency for unified rain and haze removal. Methods: DA-LightNet integrates degradation-aware feature modulation with an efficient restoration backbone. The degradation-aware feature modulation (DAFM) module adaptively recalibrates intermediate feature representations, while the lightweight architecture reduces computational requirements. The model is evaluated on the LHP-Rain and LMHaze datasets and compared with representative CNN-based, transformer-based, all-in-one, and state-space restoration methods. Results: DA-LightNet achieves a PSNR of 33.28 dB and an SSIM of 0.9267 on LHP-Rain, and a PSNR of 17.61 dB and an SSIM of 0.7536 on LMHaze. Its average PSNR and SSIM across the two datasets are 25.45 dB and 0.8402, respectively. With 4.6 million parameters and 21.5 GFLOPs, DA-LightNet achieves an inference speed of $$34.8 \pm 3.1$$ FPS on the NVIDIA Jetson Orin Nano at an input resolution of 512 × 512 pixels and a batch size of 1. Under the evaluated configuration, the model meets the 30-FPS real-time inference criterion. Conclusion: DA-LightNet provides a favorable balance between restoration quality and computational efficiency. Its low computational complexity and measured inference speed demonstrate its potential for real-time rain and haze removal on the evaluated edge platform.

Authors

Publication Details

Journal
Scientific Reports
Published
2026-11-19
DOI
https://doi.org/10.1038/s41598-026-79833-z
Primary Topic
Image Enhancement Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

DALightNet achieves real time rain and haze removal on edge devices

Deng Fuyu, Jianhong Li, Guangqiang He, Yan Liang et al.
1 links
Scientific Reports
Image Enhancement Techniques
article

DALightNet achieves real time rain and haze removal on edge devices

Deng Fuyu, Jianhong Li, Guangqiang He, Yan Liang, Miaoqin Yuan, Ke Fang
article en

Abstract

Background: Rain and haze degrade image visibility and contrast, while computationally intensive restoration methods remain challenging to deploy on resource-constrained edge devices. Aims: This study proposes DA-LightNet, a lightweight image restoration framework designed to balance restoration quality and computational efficiency for unified rain and haze removal. Methods: DA-LightNet integrates degradation-aware feature modulation with an efficient restoration backbone. The degradation-aware feature modulation (DAFM) module adaptively recalibrates intermediate feature representations, while the lightweight architecture reduces computational requirements. The model is evaluated on the LHP-Rain and LMHaze datasets and compared with representative CNN-based, transformer-based, all-in-one, and state-space restoration methods. Results: DA-LightNet achieves a PSNR of 33.28 dB and an SSIM of 0.9267 on LHP-Rain, and a PSNR of 17.61 dB and an SSIM of 0.7536 on LMHaze. Its average PSNR and SSIM across the two datasets are 25.45 dB and 0.8402, respectively. With 4.6 million parameters and 21.5 GFLOPs, DA-LightNet achieves an inference speed of $$34.8 \pm 3.1$$ FPS on the NVIDIA Jetson Orin Nano at an input resolution of 512 × 512 pixels and a batch size of 1. Under the evaluated configuration, the model meets the 30-FPS real-time inference criterion. Conclusion: DA-LightNet provides a favorable balance between restoration quality and computational efficiency. Its low computational complexity and measured inference speed demonstrate its potential for real-time rain and haze removal on the evaluated edge platform.

Scientific Reports
Image Enhancement Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon