Soutenance de thèse de Monsieur Sherif HAMDY9h00 | Amphi A | POLYTECH Angers | 62, avenue Notre-Dame du Lac 49000 ANGERS

Le 26 mars 2026

 Sujet : Contributions to computer vision and machine learning for safe, high-throughput, and non-destructive seed quality analysis and phenotyping.

Directeur de thèse : Monsieur David ROUSSEAU

Résumé

Seed testing still relies largely on destructive methods, manual inspection, and conventional laboratory practices, despite major advances in imaging and phenotyping. This thesis focuses on X-ray imaging as a core non-invasive technology and synthesises current progress to identify the main challenges: validating X-ray seed analysis physiological safety, enabling reliable AI-based automation under data scarcity, and extending AI and X-ray-based methods to complex cases beyond naked seeds. A key contribution is the first systematic validation of X-ray safety on germination across diverse species, which supports the development of safe and standardised imaging protocols and reveals that a potentially positive or negative X-ray effect is influenced by genetics, seed quality, and density. Building on these foundations, deep learning pipelines and the X-Robustifier tool were developed to enhance rapid and robust defect detection in 2D radiographs. The thesis also introduces PelletRayTion, a high-throughput, unsupervised 3D Tomography and β-VAE-based system for non-destructive contaminant detection in pelleted seeds. Overall, our work provides essential foundations for modernising seed testing.