Aug 2026· Journal of King Saud University: Computer and Information Sciences· Vol 38· 0 citations· 50 references
Computer Science
TL;DR
An illumination-guided saturation constraint loss is designed, which adaptively constrains saturation in the HSV space to reduce color cast and suppress oversaturation and shows that UAV-LIENet outperforms representative existing methods in both quantitative metrics and visual quality.
Abstract
Low-light Unmanned Aerial Vehicle (UAV) image enhancement is crucial for downstream tasks such as object detection and navigation. However, low-light UAV images often have complex illumination patterns and high noise levels. Existing mainstream low-light enhancement methods tend to introduce color cast and local overexposure on such images. To address these issues, we propose a Low-light UAV image enhancement network named UAV-LIENet, and we train it with three progressive sub-networks. UAV-LIENet first applies a Non-uniform Luminance Estimation Network (NLEN) to reconstruct a smooth and uniform illumination component. NLEN adopts quantile-clipping normalization and a parallel coarse-and-fine architecture for illumination estimation. Then, UAV-LIENet performs adaptive denoising and color restoration under the guidance of the estimated luminance component. For accurate and stable color restoration, we design an illumination-guided saturation constraint loss, which adaptively constrains saturation in the HSV space to reduce color cast and suppress oversaturation. To evaluate our method systematically, we build a low-light enhancement dataset named UAV-LLIE based on high-fidelity game-engine rendering. UAV-LLIE contains 6 typical aerial scenarios and 6,000 pixel-aligned image pairs. Experiments show that UAV-LIENet outperforms representative existing methods in both quantitative metrics and visual quality.
A Lightweight Feature-Fusion and Small-Target Enhancement Network (LFE-YOLO), a lightweight detector that coordinates partial-channel feature extraction, efficient cross-scale fusion, high-resolution prediction, background-interference suppression, and stable tiny-box regression within a unified architecture is proposed.
Mingxi Chen, Cheng Guo, Shao-Jie Ma et al.· Drones· 0 citations
UAV vision systems must operate under highly variable illumination, from daylight to full night, while adhering to strict real-time constraints. Low-light conditions suppress texture and contrast, compounding the difficulty of detecting small objects at high altitudes. Image enhancement can restore visibility, but applying it unconditionally incurs fixed compute cost regardless of scene content, degrading already-bright frames while benefiting dark ones. This paper proposes a lightweight three-signal scheduler that decides per frame whether to bypass, apply gamma correction, or invoke Zero-DCE enhancement, based on scene brightness, detector confidence feedback, and inter-frame motion. A motion veto suppresses enhancement on blurred frames where it adds artifacts rather than recovering detail. On a mixed-illumination VisDrone validation set, the scheduler achieves 45.0 FPS and 22.8% mAP@50 on a GTX 1080 Ti, invoking Zero-DCE on only 27.7% of frames, demonstrating that an adaptive per-frame scheduler makes enhancement-based detection pipelines viable for real-time UAV deployment.
Ngoc-Au Doan, Duy-Linh Nguyen, Jehwan Choi et al.· 2026 International Workshop...· 0 citations
Fog severely degrades the visibility of small unmanned aerial vehicles (UAVs) in long-range imagery, reducing the reliability of downstream detection and tracking. This paper presents a task-driven evaluation framework that links depth-aware synthetic fog generation, image restoration, object detection, and tracking within a unified pipeline. Given the practical difficulty of collecting and annotating foggy UAV scenes, synthetic fog is generated from real clear-weather outdoor images containing UAV targets using monocular depth estimation and the atmospheric scattering model. Representative restoration methods from classical, convolutional neural network (CNN)-based, and transformer-based families are first compared, after which the selected restoration model is integrated into the downstream perception pipeline. Detection is evaluated under both clean-only and fog-inclusive training regimes using multiple YOLO11 variants, while tracking-by-detection is assessed on clean, foggy, and restored video sequences. Beyond image-level restoration metrics, the study evaluates how fog and restoration affect detection robustness and tracking performance. Within the adopted synthetic-fog protocol, the results show that increasing fog severity substantially degrades both detection and tracking, primarily through increased missed detections. Fog-inclusive training provides the most consistent improvement in robustness, whereas test-time restoration is most beneficial when the detector has been trained only on clean imagery. These findings show that restoration quality does not necessarily translate into proportional gains in downstream perception and therefore should be evaluated jointly with detection and tracking performance.
Amir Pouladi, Vesal Ahsani, Hai-Jun Li et al.· Journal of Imaging· 0 citations
Unmanned aerial vehicle (UAV) imagery is widely used in urban monitoring, public security, and disaster assessment. However, object detection in UAV scenes faces multiple challenges, including a high proportion of small objects, severe occlusion in crowded areas, complex background textures, and image degradations such as haze, which often cause generic detectors to suffer from missed detections, false alarms, and unstable localization. To address these issues, we propose a lightweight multi-scale enhanced detection framework tailored for complex UAV scenarios. Built upon a MobileNet backbone, the proposed framework introduces a multi-scale enhancement module that constructs a feature pyramid and incorporates a scale-adaptive fusion mechanism to dynamically reweight the contributions of features from different scales. In addition, a fine-grained detail enhancement branch is deployed at high-resolution levels to strengthen edge and texture cues, while a context compensation module is designed to alleviate local uncertainty under dense occlusion and low-contrast conditions, thereby improving small-object separability and localization stability. Experimental results demonstrate that our method achieves strong performance on both VisDrone-DET and HazyDet, reaching mAP@0.5 = 0.312 and mAP@0.5:0.9 = 0.167 on VisDrone-DET, and mAP@0.5 = 0.719 and mAP@0.5:0.9 = 0.483 on HazyDet. The proposed method also shows favorable efficiency on an RTX 4060 Ti desktop GPU, indicating its real-time inference potential under the reported desktop hardware setting.
Xuehua Tao, Ji-Wei Sun· Engineering Research Express· 0 citations
Accurate detection and localization of small objects in unmanned aerial vehicle (UAV) images are essential for traffic monitoring, urban management, and emergency response. However, UAV imagery usually contains dense object distributions, complex backgrounds, illumination variations, and substantial scale changes, making small objects difficult to distinguish and localize accurately. To address these challenges, we propose MDF-YOLO, a context-modulated deformable feature network based on YOLOv12 for UAV small-object detection. First, a context-modulated deformable large-kernel attention module is introduced into the backbone to enhance geometric adaptability and contextual representation by incorporating large-kernel contextual information into the deformable sampling-weight generation process. Second, an rectangular self-calibration module (RCM)-based rectangular feature calibration module is embedded in the neck to strengthen foreground-aware multi-scale feature fusion and suppress background interference. Third, Focaler-intersection over union (IoU) is adopted as the bounding-box regression loss to improve the localization of difficult samples. Experiments on the VisDrone2019-DET dataset show that, compared with YOLOv12n, MDF-YOLO improves precision from 52.7% to 56.0%, recall from 40.9% to 43.8%, and mAP@50 from 43.1% to 46.0%, corresponding to gains of 3.3, 2.9, and 2.9 percentage points, respectively. MDF-YOLO contains 3.47M parameters and requires 8.50 floating point operations, maintaining a relatively compact model structure. Category-wise evaluation further demonstrates consistent AP@50 improvements across all 10 object categories, with particularly noticeable gains for visually weak and frequently small instances. These results indicate that MDF-YOLO provides a favorable balance between detection accuracy and model complexity and exhibits potential for practical UAV-based object detection applications.
RGB–infrared (RGB-IR) vehicle detection in uncrewed aerial vehicle (UAV) imagery is essential for applications, such as traffic monitoring and object tracking. However, existing methods often suffer from heterogeneous feature responses across modalities, degraded RGB feature representations under adverse illumination, and insufficient capture of fine-grained structural and edge details, which collectively impede accurate cross-modal modeling and localization. To alleviate these issues, we propose an Illumination-aware Semantic-Guided Mamba (ISGM) network. First, we design a Feature Representation Refinement module that stabilizes RGB and IR features through scale-specific channel remapping and normalized nonlinear refinement, yielding more reliable representations for subsequent cross-modal interaction. Furthermore, to enhance robustness to illumination variations and better preserve detailed structural cues and boundary information, we develop a cross-modal feature interaction mechanism comprising the Illumination-Aware Fusion Modulation (IAFM) module and the Detail-enhanced Semantic-Guided Mamba (DSGM) module. Specifically, the IAFM module estimates illumination-aware modality reliability weight maps, thereby improving robustness under challenging illumination conditions. These weight maps guide the DSGM module to integrate high-level semantic information and low-level detail cues into multiscale guidance features. These features are subsequently used to modulate the scanning parameters for adaptive RGB-IR feature interaction. This design improves the modeling of target-region features while minimizing interference from complex backgrounds. Extensive experiments on the DroneVehicle and VEDAI datasets demonstrate that ISGM outperforms state-of-the-art methods in detection performance while achieving a favorable accuracy-efficiency balance among comparable methods.
Jin-Yu Liu, Ming Li, Yu-Li Sun et al.· IEEE Journal of Selected Top...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
