Hybrid Approach for Segmenting and Counting Overlapped Cashew Nuts Using AI Based LED and FRS

Cashew nuts are one of leading tropical crops in India. It has a high commercial value worldwide, since it is enriched with minerals and vitamins. In the context of export quality cashew nut kernels, counting plays an important role in classifying different grades. Cashew nut kernels are packed according to the size and count per gram. Accordingly, this paper proposed a novel hybrid approach for segmenting and counting the overlapped cashew nut kernels by combining LED (Local Edge Thresholding) and FRS (Fast Radial Symmetry) methods using GNN (graph neural networks). The Local edge adaptive thresholding method performs edge detection and localization by keeping the edge points above the threshold. Fast Radial symmetry technique is a voting-based dimensionality reduction technique used to extract seed points on detected symmetry regions (center point of cashew nut). The segmentation improvisation was done by an AI based Graph neural network. Cashew nut kernels were detected and separated by combining the edge contours and symmetry centers to achieve an accuracy of 97.78 %.