arXiv:2503.14012v3 Announce Type: replace 
Abstract: Remote sensing object detection (RSOD) often suffers from degradations such as low spatial resolution, sensor noise, motion blur, and adverse illumination. These factors diminish feature distinctiveness, leading to ambiguous object representations and inadequate foreground-background separation. Existing RSOD methods exhibit limitations in robust detection of low-quality objects. To address these pressing challenges, we introduce LEGNet, a lightweight backbone network featuring a novel Edge-Gaussian Aggregation (EGA) module specifically engineered to enhance feature representation derived from low-quality remote sensing images. EGA module integrates: (a) orientation-aware Scharr filters to sharpen crucial edge details often lost in low-contrast or blurred objects, and (b) Gaussian-prior-based feature refinement to suppress noise and regularize ambiguous feature responses, enhancing foreground saliency under challenging conditions. EGA module alleviates prevalent problems in reduced contrast, structural discontinuities, and ambiguous feature responses prevalent in degraded images, effectively improving model robustness while maintaining computational efficiency. Comprehensive evaluations across five benchmarks (DOTA-v1.0, v1.5, DIOR-R, FAIR1M-v1.0, and VisDrone2019) demonstrate that LEGNet achieves state-of-the-art performance, particularly in detecting low-quality objects.The code is available at https://github.com/AeroVILab-AHU/LEGNet.

طور الباحثون LEGNet، وهو شبكة خفيفة الوزن تعتمد على الحافة الغاوسية، مصممة لتحسين الكشف عن الأجسام في صور الاستشعار عن بعد ذات الجودة المنخفضة. تعالج هذه الابتكارات المشكلات الشائعة مثل انخفاض الدقة المكانية وضوضاء المستشعر التي تعيق غالبًا الكشف الدقيق. من خلال تعزيز تميز الميزات والفصل بين المقدمة والخلفية، يعد LEGNet بتقدم كبير في مجال الاستشعار عن بعد، مما يسهل التعرف على الأجسام في ظروف صعبة. قد يكون لهذا الاختراق تداعيات مهمة على تطبيقات متعددة، بما في ذلك مراقبة البيئة والاستجابة للكوارث.

Investigadores han desarrollado LEGNet, una nueva red ligera de borde gaussiano diseñada para mejorar la detección de objetos en imágenes de teledetección de baja calidad. Esta innovación aborda problemas comunes como la baja resolución espacial y el ruido del sensor que a menudo obstaculizan la detección precisa. Al mejorar la distintividad de las características y la separación entre el primer plano y el fondo, LEGNet promete avanzar significativamente en el campo de la teledetección, facilitando la identificación de objetos en condiciones desafiantes. Este avance podría tener importantes implicaciones para diversas aplicaciones, incluida la monitorización ambiental y la respuesta a desastres.

Des chercheurs ont développé LEGNet, un nouveau réseau léger à bord gaussien conçu pour améliorer la détection d'objets dans des images de télédétection de faible qualité. Cette innovation répond à des problèmes courants tels que la faible résolution spatiale et le bruit des capteurs qui entravent souvent une détection précise. En améliorant la distinctivité des caractéristiques et la séparation avant-arrière-plan, LEGNet promet d'avancer considérablement le domaine de la télédétection, facilitant ainsi l'identification d'objets dans des conditions difficiles. Cette avancée pourrait avoir des implications importantes pour diverses applications, y compris la surveillance environnementale et la réponse aux catastrophes.

Researchers have developed LEGNet, a new lightweight edge-Gaussian network designed to improve object detection in low-quality remote sensing images. This innovation addresses common issues like low spatial resolution and sensor noise that often hinder accurate detection. By enhancing feature distinctiveness and foreground-background separation, LEGNet promises to significantly advance the field of remote sensing, making it easier to identify objects in challenging conditions. This breakthrough could have important implications for various applications, including environmental monitoring and disaster response.

LEGNet: A Lightweight Edge-Gaussian Network for Low-Quality Remote Sensing Image Object Detection

arXiv:2601.08446v1 Announce Type: cross 
Abstract: The development of reliable methods for multi-label classification (MLC) has become a prominent research direction in remote sensing (RS). As the scale of RS data continues to expand, annotation procedures increasingly rely on thematic products or crowdsourced procedures to reduce the cost of manual annotation. While cost-effective, these strategies often introduce multi-label noise in the form of partially incorrect annotations. In MLC, label noise arises as additive noise, subtractive noise, or a combination of both in the form of mixed noise. Previous work has largely overlooked this distinction and commonly treats noisy annotations as supervised signals, lacking mechanisms that explicitly adapt learning behavior to different noise types. To address this limitation, we propose NAR, a noise-adaptive regularization method that explicitly distinguishes between additive and subtractive noise within a semi-supervised learning framework. NAR employs a confidence-based label handling mechanism that dynamically retains label entries with high confidence, temporarily deactivates entries with moderate confidence, and corrects low confidence entries via flipping. This selective attenuation of supervision is integrated with early-learning regularization (ELR) to stabilize training and mitigate overfitting to corrupted labels. Experiments across additive, subtractive, and mixed noise scenarios demonstrate that NAR consistently improves robustness compared with existing methods. Performance improvements are most pronounced under subtractive and mixed noise, indicating that adaptive suppression and selective correction of noisy supervision provide an effective strategy for noise robust learning in RS MLC.

تم اقتراح طريقة جديدة تُدعى التنظيم التكيفي للضوضاء (NAR) لتحسين التصنيف متعدد التسميات في استشعار عن بُعد، حيث تتناول التحديات التي تطرحها التعليقات المزعجة التي قد تنشأ من طرق جمع البيانات ذات التكلفة المنخفضة. يميز NAR بين الضوضاء المضافة والضوضاء المخصومة ضمن إطار التعلم شبه الموجه، مما يعزز من قوة تصنيف الصور.

Se ha propuesto un nuevo método llamado Regularización Adaptativa al Ruido (NAR) para mejorar la clasificación multi-etiqueta en la teledetección, abordando los desafíos que presentan las anotaciones ruidosas que pueden surgir de métodos de recolección de datos rentables. NAR distingue entre ruido aditivo y ruido sustractivo dentro de un marco de aprendizaje semi-supervisado, mejorando así la robustez de la clasificación de imágenes.

Une nouvelle méthode appelée Régularisation Adaptative au Bruit (NAR) a été proposée pour améliorer la classification multi-étiquettes dans la télédétection, en abordant les défis posés par les annotations bruyantes qui peuvent découler de méthodes de collecte de données rentables. NAR distingue entre le bruit additif et le bruit soustractif dans un cadre d'apprentissage semi-supervisé, renforçant ainsi la robustesse de la classification d'images.

A new method called Noise-Adaptive Regularization (NAR) has been proposed to improve multi-label classification in remote sensing, addressing the challenges posed by noisy annotations that can arise from cost-effective data collection methods. NAR distinguishes between additive and subtractive noise within a semi-supervised learning framework, enhancing the robustness of image classification.

LEGNet: A Lightweight Edge-Gaussian Network for Low-Quality Remote Sensing Image Object Detection

Was this article worth reading? Share it

LucidQuery AI

WasItAI

Lenso.ai

Attentive AI

LexiStock AI

Unblurimage AI

Ready to build your own newsroom?