Seed non-destructive intelligent detection technology: A review of current status, advances, and future directions

Escalating food security challenges demand accurate and efficient seed quality evaluation, yet conventional destructive methods suffer from subjectivity, inefficiency, and irreversible sample loss. This review synthesizes the state-of-the-art in seed non-destructive detection technologies published between 2021 and 2025, focusing on the integration of machine vision, hyperspectral imaging, and deep learning. Across diverse tasks and datasets, studies consistently report high accuracies ranging from 90% to 97%, although many publicly reported datasets contain fewer than 10,000 samples. These results underscore the potential of non-destructive approaches, but they also reveal persistent challenges. The reliance on small-scale public datasets limits generalization to real-world field conditions, while the high equipment costs and the opacity of deep learning models hinder broader adoption. By tracing the evolution from laboratory studies to industrial-scale applications, this work offers valuable guidance for engineers, researchers, and industry stakeholders, presenting a clear pathway toward the development of robust, scalable, and biologically meaningful seed non-destructive detection technologies.