Dehazing Remote Sensing and UAV Imagery: A Review of Deep Learning, Prior-based, and Hybrid Approaches

TL;DR

This paper surveys image dehazing with a focus on remote sensing and UAV imagery, tracing the transition from prior-based methods to CNN- and ViT-based deep learning approaches and to contrastive/few-shot paradigms. It analyzes how Atmospheric Scattering Model–driven priors interact with end-to-end learning, and contrasts performance across benchmark datasets such as RESIDE, NH-HAZE, O-HAZE, and DENSE-HAZE. It also surveys RS-specific dehazing for hyperspectral, VHR, SAR, and UAV data, highlighting trends toward efficiency, domain adaptation, and detail recovery, while identifying gaps in real-world hazy data and higher-level vision task integration. The review emphasizes the importance of diverse real-world datasets and physics-informed, hybrid architectures to advance practical RS/UAV dehazing. Overall, the work provides a comprehensive, up-to-date resource (as of 2024) for researchers and practitioners seeking robust, scalable dehazing solutions across imaging domains.

Abstract

High-quality images are crucial in remote sensing and UAV applications, but atmospheric haze can severely degrade image quality, making image dehazing a critical research area. Since the introduction of deep convolutional neural networks, numerous approaches have been proposed, and even more have emerged with the development of vision transformers and contrastive/few-shot learning. Simultaneously, papers describing dehazing architectures applicable to various Remote Sensing (RS) domains are also being published. This review goes beyond the traditional focus on benchmarked haze datasets, as we also explore the application of dehazing techniques to remote sensing and UAV datasets, providing a comprehensive overview of both deep learning and prior-based approaches in these domains. We identify key challenges, including the lack of large-scale RS datasets and the need for more robust evaluation metrics, and outline potential solutions and future research directions to address them. This review is the first, to our knowledge, to provide comprehensive discussions on both existing and very recent dehazing approaches (as of 2024) on benchmarked and RS datasets, including UAV-based imagery.

Figures (16)

• Figure 1: Schematic illustration of the ASM concept. Note that this model is only valid on daytime homogeneous haze. The $J(x)t(x)$ is the direct attenuation term, and is the result of light reflected from a scenery of interest transmitting through a layer of haze. The airlight term $A(1-t(x))$ is the result of light source passing through a layer of haze and illuminating the surrounding. Some levels of light scattering due the haze also contributes to the airlight term. $d(x)$ denotes the distance from the object to the sensor (camera). Blue, maroon, orange, and pink arrows denote attenuated, airlight, direct sunlight, and reflected light respectively.
• Figure 2: Bar graphs of frequency of each type of dehazing approach against the year (last one or two digit) in which it was published, based on all the works described so far. Bars in red denote prior-based works, bars in green denote CNN-based works, bars in blue denote ViT-based works, and bars in magenta denote contrastive or few-shot based works.
• Figure 3: Bar graphs of number of times a proposed dehazing approach utilized the commonly dehazing benchmarked datasets (RESIDE, NH-HAZE, DENSE-HAZE, O-HAZE), along with the non-benchmark ('others') for all the works discussed so far. We can see that RESIDE remained the most widely utilized, followed by NH-HAZE, then 'others', O-HAZE, DENSE-HAZE and lastly the I-HAZE.
• Figure 4: Scatter plots of the PSNR vs SSIM for the selected dehazing methods displayed in Table 7 as applied on NH-HAZE (2020). Plots in red denote prior-based dehazing, plots in green denote CNN-based dehazing, plots in blue denote ViT-based dehazing, and plots in magenta denote contrastive or few-shot-based dehazing.
• Figure 5: Scatter plots of the PSNR vs SSIM for the selected dehazing methods displayed in Table 8 as applied on DENSE-HAZE. Plots in red denote prior-based dehazing, plots in green denote CNN-based dehazing, plots in blue denote ViT-based dehazing, and plots in magenta denote contrastive or few-shot-based dehazing.