Combining Open-Set Recognition and Incremental Learning Approaches for Monitoring Agricultural Crops

• Grant number: 25/18352-8
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: November 01, 2025
• End date: October 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: André Luis Debiaso Rossi
• Grantee: Alexandre Luís Frata Selani
• Host Institution: Faculdade de Ciências (FC). Universidade Estadual Paulista (UNESP). Campus de Bauru. Bauru , SP, Brazil

Abstract

The advancement of Deep Learning has broadened artificial intelligence applications in various domains, including agricultural aerial imaging. However, traditional classifiers trained in closed-set scenarios struggle with unseen instances, such as emerging plant diseases or the presence of weeds, which undermines model reliability and slows management decisions. Open Set Recognition (OSR) addresses this limitation by detecting unknown samples, while Open World Recognition (OWR) further allows incremental learning of new classes, mitigating catastrophic forgetting. This research investigates and compares measures such as entropy and silhouette for novelty detection, and evaluates recent OSR methods, including OpenMax and OpenGAN, for recognizing species and diseases in crops from images captured by Unmanned Aerial Vehicles (UAVs). Additionally, it explores integrating these approaches with incremental learning algorithms, such as iCaRL and PackNet, to assess their effectiveness in OWR scenarios. The expected outcome is the development of more robust and adaptive classifiers capable of detecting novelties, learning continuously, and supporting efficient, scalable crop monitoring (AU)