arXiv:2510.27155v1 Announce Type: new 
Abstract: Remote sensing image scene classification remains a challenging task, primarily due to the complex spatial structures and multi-scale characteristics of ground objects. Existing approaches see CNNs excel at modeling local textures, while Transformers excel at capturing global context. However, efficiently integrating them remains a bottleneck due to the high computational cost of Transformers. To tackle this, we propose AFM-Net, a novel Advanced Hierarchical Fusing framework that achieves effective local and global co-representation through two pathways: a CNN branch for extracting hierarchical visual priors, and a Mamba branch for efficient global sequence modeling. The core innovation of AFM-Net lies in its Hierarchical Fusion Mechanism, which progressively aggregates multi-scale features from both pathways, enabling dynamic cross-level feature interaction and contextual reconstruction to produce highly discriminative representations. These fused features are then adaptively routed through a Mixture-of-Experts classifier module, which dispatches them to the most suitable experts for fine-grained scene recognition. Experiments on AID, NWPU-RESISC45, and UC Merced show that AFM-Net obtains 93.72, 95.54, and 96.92 percent accuracy, surpassing state-of-the-art methods with balanced performance and efficiency. Code is available at https://github.com/tangyuanhao-qhu/AFM-Net.

تقديم AFM-Net يمثل تقدمًا كبيرًا في تصنيف مشاهد صور الاستشعار عن بُعد، حيث يتناول التحديات التي تطرحها الهياكل المكانية المعقدة والخصائص متعددة المقاييس للأجسام الأرضية. من خلال دمج نقاط القوة في الشبكات العصبية التلافيفية وTransformers بشكل فعال، يقدم AFM-Net حلاً أكثر كفاءة يمكن أن يعزز دقة وسرعة تصنيف الصور في هذا المجال. هذه الابتكار مهم لأنه يفتح آفاقًا جديدة للتطبيقات في مراقبة البيئة، والتخطيط الحضري، وإدارة الكوارث.

La introducción de AFM-Net marca un avance significativo en la clasificación de escenas de imágenes de teledetección, abordando los desafíos que presentan las estructuras espaciales complejas y las características multiescala de los objetos en el suelo. Al combinar de manera efectiva las fortalezas de las CNN y los Transformers, AFM-Net ofrece una solución más eficiente que podría mejorar la precisión y la velocidad de la clasificación de imágenes en este campo. Esta innovación es crucial, ya que abre nuevas posibilidades para aplicaciones en monitoreo ambiental, planificación urbana y gestión de desastres.

L'introduction d'AFM-Net représente une avancée significative dans la classification des scènes d'images de télédétection, en répondant aux défis posés par les structures spatiales complexes et les caractéristiques multi-échelles des objets au sol. En combinant efficacement les forces des CNN et des Transformers, AFM-Net offre une solution plus efficace qui pourrait améliorer la précision et la rapidité de la classification des images dans ce domaine. Cette innovation est cruciale car elle ouvre de nouvelles possibilités d'applications dans la surveillance environnementale, la planification urbaine et la gestion des catastrophes.

The introduction of AFM-Net marks a significant advancement in remote sensing image scene classification, addressing the challenges posed by complex spatial structures and multi-scale characteristics of ground objects. By effectively combining the strengths of CNNs and Transformers, AFM-Net offers a more efficient solution that could enhance the accuracy and speed of image classification in this field. This innovation is crucial as it opens up new possibilities for applications in environmental monitoring, urban planning, and disaster management.

AFM-Net: Advanced Fusing Hierarchical CNN Visual Priors with Global Sequence Modeling for Remote Sensing Image Scene Classification

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.

Noise-Adaptive Regularization for Robust Multi-Label Remote Sensing Image Classification

One More Thing in AI – Your Shortcut to AI Mastery

AFM-Net: Advanced Fusing Hierarchical CNN Visual Priors with Global Sequence Modeling for Remote Sensing Image Scene Classification

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

Attentive AI

Octofy

The Visualizer

AINF Membership

Ready to build your own newsroom?