Quality classification of Jatropha curcas seeds using radiographic images and machine learning

Abstract

Efficient seed quality assessment methodologies are important for the seed industry. Advanced seed technology research requires the use of high productivity methods that provide detailed information on seed structural integrity and predict its physiological potential quickly and accurately. The aim of this study was to propose a method for predicting germination capacity and discriminate Jatropha curcas L. seeds regarding germination speed and seedling vigor by combining automatic X-ray analysis and machine learning model. The study was performed using automated analysis of radiographic images of seeds, obtaining a series of morphological and tissue integrity descriptors. After the X-ray test, the seeds were submitted to physiological assessments. Based on all individual seed descriptors, quality classes were created and LDA models were applied. Prediction of seed viability, germination speed and seedling vigor resulted in an average of 94.36, 83.72 and 89.72% of correctly classified seeds, respectively. High throughput X-ray image analysis can provide information needed to discriminate individual Jatropha curcas seeds into different classes of quality, i.e., germination capacity, germination speed and seedling vigor. The methodology proposed can be used to discriminate between seed classes quickly and robustly.