arXiv:2511.16218v1 Announce Type: new 
Abstract: Real-world agricultural distributions often suffer from severe class imbalance, typically following a long-tailed distribution. Labeled datasets for crop-type classification are inherently scarce and remain costly to obtain. When working with such limited data, training sets are frequently constructed to be artificially balanced -- in particular in the case of few-shot learning -- failing to reflect real-world conditions. This mismatch induces a shift between training and test label distributions, degrading real-world generalization. To address this, we propose Dirichlet Prior Augmentation (DirPA), a novel method that simulates an unknown label distribution skew of the target domain proactively during model training. Specifically, we model the real-world distribution as Dirichlet-distributed random variables, effectively performing a prior augmentation during few-shot learning. Our experiments show that DirPA successfully shifts the decision boundary and stabilizes the training process by acting as a dynamic feature regularizer.

يقدم المقال طريقة جديدة تُسمى تعزيز الأولوية ديريشليت (DirPA) تهدف إلى تحسين تصنيف أنواع المحاصيل القليلة الأمثلة في الزراعة. يتناول مشكلة عدم توازن الفئات في مجموعات البيانات المعلّمة، والتي غالبًا ما تؤدي إلى تباين بين ظروف التدريب والاختبار. من خلال محاكاة توزيع غير معروف للتسميات أثناء تدريب النموذج، يُحسن DirPA قدرة النموذج على التعميم في السيناريوهات الواقعية، مما يزيد من دقة التصنيف.

El artículo presenta un nuevo método llamado Aumento de Prior de Dirichlet (DirPA) que busca mejorar la clasificación de tipos de cultivos con pocos ejemplos en la agricultura. Aborda el problema del desequilibrio de clases en conjuntos de datos etiquetados, que a menudo conduce a una discrepancia entre las condiciones de entrenamiento y prueba. Al simular una distribución de etiquetas desconocida durante el entrenamiento del modelo, DirPA mejora la capacidad del modelo para generalizar a escenarios del mundo real, aumentando así la precisión de la clasificación.

L'article présente une nouvelle méthode appelée Dirichlet Prior Augmentation (DirPA) visant à améliorer la classification des types de cultures en peu d'exemples dans l'agriculture. Elle aborde le problème du déséquilibre des classes dans les ensembles de données étiquetées, ce qui entraîne souvent un décalage entre les conditions d'entraînement et de test. En simulant une distribution d'étiquettes inconnue pendant l'entraînement du modèle, DirPA améliore la capacité du modèle à se généraliser à des scénarios réels, augmentant ainsi la précision de la classification.

The article presents a new method called Dirichlet Prior Augmentation (DirPA) aimed at improving few-shot crop-type classification in agriculture. It addresses the issue of class imbalance in labeled datasets, which often leads to a mismatch between training and testing conditions. By simulating an unknown label distribution during model training, DirPA enhances the model's ability to generalize to real-world scenarios, thereby improving classification accuracy.

Mind the Gap: Bridging Prior Shift in Realistic Few-Shot Crop-Type Classification

arXiv:2511.16541v1 Announce Type: new 
Abstract: The rapid advancement of generative artificial intelligence has enabled the creation of synthetic images that are increasingly indistinguishable from authentic content, posing significant challenges for digital media integrity. This problem is compounded by the accelerated release cycle of novel generative models, which renders traditional detection approaches (reliant on periodic retraining) computationally infeasible and operationally impractical.
  This work proposes a novel two-stage detection framework designed to address the generalization challenge inherent in synthetic image detection. The first stage employs a vision deep learning model trained via supervised contrastive learning to extract discriminative embeddings from input imagery. Critically, this model was trained on a strategically partitioned subset of available generators, with specific architectures withheld from training to rigorously ablate cross-generator generalization capabilities. The second stage utilizes a k-nearest neighbors (k-NN) classifier operating on the learned embedding space, trained in a few-shot learning paradigm incorporating limited samples from previously unseen test generators.
  With merely 150 images per class in the few-shot learning regime, which are easily obtainable from current generation models, the proposed framework achieves an average detection accuracy of 91.3\%, representing a 5.2 percentage point improvement over existing approaches . For the source attribution task, the proposed approach obtains improvements of of 14.70\% and 4.27\% in AUC and OSCR respectively on an open set classification context, marking a significant advancement toward robust, scalable forensic attribution systems capable of adapting to the evolving generative AI landscape without requiring exhaustive retraining protocols.

أدى التقدم السريع في الذكاء الاصطناعي التوليدي إلى إنشاء صور اصطناعية يصعب تمييزها عن المحتوى الحقيقي، مما يطرح تحديات كبيرة لسلامة الوسائط الرقمية. تقترح هذه الدراسة إطار عمل جديد للكشف من مرحلتين مصمم لمعالجة تحدي التعميم في الكشف عن الصور الاصطناعية. تستخدم المرحلة الأولى نموذج تعلم عميق مدرب من خلال التعلم التبايني المراقب لاستخراج تمثيلات تمييزية من الصور المدخلة.

El rápido avance de la inteligencia artificial generativa ha permitido la creación de imágenes sintéticas que son cada vez más difíciles de distinguir del contenido real, lo que plantea desafíos significativos para la integridad de los medios digitales. Este trabajo propone un nuevo marco de detección en dos etapas diseñado para abordar el desafío de generalización inherente a la detección de imágenes sintéticas. La primera etapa utiliza un modelo de aprendizaje profundo entrenado mediante aprendizaje contrastivo supervisado para extraer embeddings discriminativos de las imágenes de entrada.

L'avancement rapide de l'intelligence artificielle générative a conduit à la création d'images synthétiques de plus en plus difficiles à distinguer du contenu réel, posant des défis à l'intégrité des médias numériques. Cette étude propose un cadre de détection en deux étapes utilisant l'apprentissage contrastif supervisé pour améliorer la détection des images générées par l'IA. La première étape se concentre sur l'extraction d'embeddings discriminants à partir des images, abordant le défi de généralisation dans la détection d'images synthétiques.

The rapid advancement of generative artificial intelligence has led to the creation of synthetic images that are increasingly difficult to distinguish from real content, challenging digital media integrity. This study proposes a two-stage detection framework utilizing supervised contrastive learning to enhance the detection of AI-generated images. The first stage focuses on extracting discriminative embeddings from images, addressing the generalization challenge in synthetic image detection.

Mind the Gap: Bridging Prior Shift in Realistic Few-Shot Crop-Type Classification

Was this article worth reading? Share it