U2S-HS: Unsupervised-to-Supervised Haze Stratification for Adaptive Dehazing with Multi-Profile EDCP.

Saved in:

Bibliographic Details
Title:	U2S-HS: Unsupervised-to-Supervised Haze Stratification for Adaptive Dehazing with Multi-Profile EDCP.
Authors:	Hafidh, Fathul¹ hafidh@uniska-bjm.ac.id, Shidik, Guruh Fajar² guruh.fajar@research.dinus.ac.id, Syukur, Abdul³ abah.syukur01@dsn.dinus.ac.id, Andono, Pulung Nurtantio² pulung@dsn.dinus.ac.id
Source:	IAENG International Journal of Computer Science. Jul2026, Vol. 53 Issue 7, p2834-2851. 18p.
Subjects:	Random forest algorithms, K-means clustering, Machine learning, Supervised learning, Remote sensing
Abstract:	Single-model dehazing methods often struggle with varying atmospheric conditions due to rigid parameters, while deep learning alternatives demand prohibitive computational resources for edge deployment. This study addresses the challenge of haze density variability and the scarcity of labeled remote sensing datasets by proposing U2S-HS (Unsupervisedto-Supervised Haze Stratification). The framework introduces a hybrid pipeline that uses K-Means clustering aligned via the Hungarian Algorithm to generate robust pseudo-labels from unlabeled data. A lightweight Random Forest classifier, optimized with a 5-feature spectral manifold, is then trained to dynamically route input images to the most effective Enhanced Dark Channel Prior (EDCP) engine. Experimental results on the Haze1k dataset demonstrate that U2S-HS achieves a classification accuracy of 72.59%, with a notable 93.33% precision in thick haze detection. Furthermore, the framework maintains high generalization on out-of-domain datasets (RRSHID), achieving a reliable PSNR of 16.519 dB. With an average processing speed of 0.045 seconds on a standard CPU, U2SHS offers a label-efficient, high-speed, and adaptive solution specifically optimized for real-time remote sensing applications on resource-constrained platforms. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Single-model dehazing methods often struggle with varying atmospheric conditions due to rigid parameters, while deep learning alternatives demand prohibitive computational resources for edge deployment. This study addresses the challenge of haze density variability and the scarcity of labeled remote sensing datasets by proposing U2S-HS (Unsupervisedto-Supervised Haze Stratification). The framework introduces a hybrid pipeline that uses K-Means clustering aligned via the Hungarian Algorithm to generate robust pseudo-labels from unlabeled data. A lightweight Random Forest classifier, optimized with a 5-feature spectral manifold, is then trained to dynamically route input images to the most effective Enhanced Dark Channel Prior (EDCP) engine. Experimental results on the Haze1k dataset demonstrate that U2S-HS achieves a classification accuracy of 72.59%, with a notable 93.33% precision in thick haze detection. Furthermore, the framework maintains high generalization on out-of-domain datasets (RRSHID), achieving a reliable PSNR of 16.519 dB. With an average processing speed of 0.045 seconds on a standard CPU, U2SHS offers a label-efficient, high-speed, and adaptive solution specifically optimized for real-time remote sensing applications on resource-constrained platforms. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X