Automated detection of insect-damaged sunflower seeds by X-ray imaging

Thomas Pearson, Jarrad Prasifka, Daniel Brabec, Ronald Haff, Brent Hulke

Abstract: The development of insect-resistant sunflowers is hindered by the lack of a quick and effective method for scoring samples in terms of insect damage. The current method for scoring insect damage, which involves manual inspection of seeds for holes bored into the shell, is tedious, requiring approximately 10 minutes per 100-kernel sample. In this study, a method was developed to quickly position 72 to 144 sunflower seeds in a grid of closely packed, non-touching seeds consistently oriented for X-ray imaging. A computer program was developed to analyze the images and classify each seed as damaged or undamaged. Applying the program to 20 different samples comprising 11 sunflower lines and infestation by three different species of insects resulted in an overall classification accuracy for damaged and undamaged seeds of 95% and 99%, respectively. The method takes approximately 3 minutes per sample. The detection algorithm uses a simple but novel method for detecting seeds having asymmetrical morphology due to insect feeding. The method should aid in scoring sunflower seed varieties for insect resistance and could also be applied to other applications, such as detecting broken seeds.

To view article click here.