Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications

Haze removal is essential for improving image visibility in applications like autonomous vehicles and surveillance. While models like GridDehazeNet enhance dehazing performance, our research emphasizes that selecting appropriate image data and application methods is even more crucial. We propose an...

Full description

Saved in:
Bibliographic Details
Main Author: Sungkwan Youm
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10955216/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849731045940789248
author Sungkwan Youm
author_facet Sungkwan Youm
author_sort Sungkwan Youm
collection DOAJ
description Haze removal is essential for improving image visibility in applications like autonomous vehicles and surveillance. While models like GridDehazeNet enhance dehazing performance, our research emphasizes that selecting appropriate image data and application methods is even more crucial. We propose an adaptive haze removal system that integrates Bayesian Optimization with GridDehazeNet to automatically find the optimal network width and height for specific environments. Our experiments show that the optimal architectural parameters of the neural network—specifically, the height (h) and width (w) of the network architecture—vary for images captured by fixed-position cameras in real-world settings. By employing our algorithm, we effectively identify the best architectural parameters, achieving higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) compared to models without this optimization. Notably, fine-tuning the model for specific environments yields a 36% average boost in PSNR over the baseline while reducing the model parameter count by 63%. This is a pivotal outcome, as it addresses two critical demands in dehazing: improved performance and computational efficiency. By demonstrating that a smaller, optimized network can surpass a larger, generic baseline in both accuracy and speed, our work suggests a practical pathway toward real-time dehazing solutions, particularly in resource-constrained or edge-based scenarios.
format Article
id doaj-art-a88a13a4d34e4261b8242e543abf4ab8
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-a88a13a4d34e4261b8242e543abf4ab82025-08-20T03:08:40ZengIEEEIEEE Access2169-35362025-01-0113628076281910.1109/ACCESS.2025.355863210955216Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World ApplicationsSungkwan Youm0https://orcid.org/0000-0002-9751-8732Department of Information and Communication Engineering, Wonkwang University, Iksan, Republic of KoreaHaze removal is essential for improving image visibility in applications like autonomous vehicles and surveillance. While models like GridDehazeNet enhance dehazing performance, our research emphasizes that selecting appropriate image data and application methods is even more crucial. We propose an adaptive haze removal system that integrates Bayesian Optimization with GridDehazeNet to automatically find the optimal network width and height for specific environments. Our experiments show that the optimal architectural parameters of the neural network—specifically, the height (h) and width (w) of the network architecture—vary for images captured by fixed-position cameras in real-world settings. By employing our algorithm, we effectively identify the best architectural parameters, achieving higher Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) compared to models without this optimization. Notably, fine-tuning the model for specific environments yields a 36% average boost in PSNR over the baseline while reducing the model parameter count by 63%. This is a pivotal outcome, as it addresses two critical demands in dehazing: improved performance and computational efficiency. By demonstrating that a smaller, optimized network can surpass a larger, generic baseline in both accuracy and speed, our work suggests a practical pathway toward real-time dehazing solutions, particularly in resource-constrained or edge-based scenarios.https://ieeexplore.ieee.org/document/10955216/Bayesian optimizationcomputer visiondeep learningenvironmental adaptationfixed-position camerasGridDehazeNet
spellingShingle Sungkwan Youm
Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
IEEE Access
Bayesian optimization
computer vision
deep learning
environmental adaptation
fixed-position cameras
GridDehazeNet
title Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
title_full Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
title_fullStr Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
title_full_unstemmed Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
title_short Bayesian Optimized GridDehazeNet for Adaptive Haze Removal in Real-World Applications
title_sort bayesian optimized griddehazenet for adaptive haze removal in real world applications
topic Bayesian optimization
computer vision
deep learning
environmental adaptation
fixed-position cameras
GridDehazeNet
url https://ieeexplore.ieee.org/document/10955216/
work_keys_str_mv AT sungkwanyoum bayesianoptimizedgriddehazenetforadaptivehazeremovalinrealworldapplications