A Method for Segmenting Grain Particles in Hyperspectral Images

Hyperspectral imaging, known for its high spectral resolution and nondestructive detection characteristics, has been widely applied in grain quality evaluation. Grain quality evaluation utilizing near-infrared hyperspectral images often uses individual particles as the evaluation unit and thus relies on the segmentation of particles. However, widely applicable methods for the segmentation of grain particles remain inadequate because of the challenge of segmenting adhesive particles. Therefore, this study introduces a new method designed to significantly improve the accuracy of segmenting grain particles. This method considers the shape characteristics of grains to correct the oversegmentation in watershed segmentation caused by adhesive particles. The feasibility of the method was tested by applying it to three grains: corn, wheat, and peanuts. Results demonstrated that the new method performed significantly better than watershed segmentation, with accuracies of no less than 90% for the three grains. The new method has the potential to support various studies related to grain quality evaluation, such as mold identification and moisture detection of individual particles.