Algorithm for seed counting of adherent buckwheat based on area estimation

ResU-net network in the segmentation of adhesive buckwheat seed images, buckwheat adhesion degree increases will lead to a decline in the counting accuracy. In this paper, we propose a method based on area estimation and improvement, firstly, counting the area and roundness of the same buckwheat grain, clustering and setting the threshold; secondly, establishing a fuzzy inference system, input area and roundness, and output the number of grains; and finally, accumulating the output of all buckwheat grains. The experimental results show that the method can accurately count the adherent buckwheat seeds with an error rate of 2% or less, which meets the practical needs and avoids the waste of resources in expanding the dataset. This provides a solution for accurately counting adherent buckwheat kernels, which has potential for practical application.