Accurate seed classification and phenotyping underpin agricultural productivity, biodiversity assessment, and ecological conservation, yet current practice still depends largely on manual inspection or static machine-learning models trained under closed-world assumptions. These approaches typically require carefully configured imaging processes in controlled environments and must be retrained from scratch when new seed species appear, limiting scalability and efficiency. To address this gap, we present SeedStudio, a unified, low-cost platform that integrates automated multi-view seed imaging with continual, open-world learning. The system couples a portable imaging capsule—equipped with dual cameras, adaptive illumination, and on-device inference for known species—with a server-side, class-incremental learning model that incorporates images of previously unseen seeds without full model retraining. The updated server-side model can be synchronized back to capsules, enabling recognition capabilities to evolve continuously in lab or field deployments. Experiments on 20 seed species show that multi-view imaging increases per-seed accuracy by 75% on complete datasets, and that the class-incremental learning model enhances classification accuracy by more than 4%–20% on average compared with state-of-the-art baselines, while maintaining computational efficiency suitable for embedded hardware. These results demonstrate that closing the loop between edge-side morphological sensing and adaptive server-side intelligence provides a practical route to lifelong seed phenotyping, and more broadly, offers a scalable blueprint for intelligent sensing systems operating in dynamic ecological and agricultural environments. • SeedStudio unifies low-cost multi-view seed imaging and continual learning. • 3D-printed imaging capsule with dual cameras and adaptive illumination. • Multi-view imaging substantially improves seed classification performance. • Similarity-aware sampling enables efficient incremental seed classification.
Luo et al. (Wed,) studied this question.