Evaluation of seed radiographic images by independent component analysis and discriminant analysis

Although subjective, the use of X-ray images of seeds is an important tool for analysing seed lot quality. Here, we applied independent component analysis (ICA) for automatic processing of radiographic images of 600 sunflower seeds. The X-rayed seeds were also subjected to a germination test. The ICA technique was implemented with the FastICA algorithm, which decomposed X-ray images to independent basis images. Based on features extracted by ICA, we used discriminant analysis (DA) to classify seed quality. The classification achieved an overall accuracy of 82%. The results showed that ICA and DA were effective in X-ray analysis to associate seed morphology and seedling performance.