Optical Techniques for Automated Evaluation of Seed Damage

Abstract

Cereal grains and oilseeds, fundamental components of global diets, face significant vulnerability to mechanical damage during various stages, including harvesting, transportation, and storage. Beyond immediate physical degradation, the repercussions of such damage extend to seed viability and consequent economic implications. Traditional assessment techniques, predominantly reliant on external visual inspections, face challenges of subjectivity and inefficiency, restricting evaluations to superficial seed alterations. To circumvent these shortcomings, this study presents a fusion of optical techniques, namely two-dimensional (2D) X-ray imaging and hyperspectral imaging (HSI) – all underpinned by machine learning and deep learning frameworks – targeting an automated, holistic assessment of flaxseed damages. Leveraging an expansive dataset of 3,600 flaxseed samples spanning varied moisture contents and impact energies, the findings underscore the amplified susceptibility of seeds to damage under heightened impact stress at minimal moisture levels. Remarkably, through the integrated approach, the study achieved classification accuracies surpassing 87% for all techniques. While X-ray imaging presented throughput limitations, Vis-NIR HSI can be considered an effective alternative. In summation, the study accentuates the profound potential harboured by optical techniques in seed damage assessments, advocating their capacity to replace conventional methods. By seamlessly integrating advanced imaging with computational intelligence, the study not only streamlines damage detection but also amplifies the possibility of curbing damage, promising heightened yields and minimized economic setbacks. Future endeavors should channel this foundational research towards broader crop varieties to ensure universal applicability and validation.