Seed Segregation using Deep Learning

Abstract

A superior crop yield is a vital part of the agricultural industry. The principal component for a good yield is good quality seeds. Generally, seeds are sown without prior quality checks and inspections as these processes are tedious and labor intensive. This tends to diminish the crop yield as well as crop quality. This paper proposes a novel method to automatically sort seeds as good or bad based on the visual characteristics of the seed using a Convolutional Neural Network. The data set used to train the model comprised of images of the top and bottom profiles of the seeds. The Convolutional Neural Network provided a classification accuracy of 96.875%. This study uses a hardware solution which classifies seeds using the CNN model. The device performs significantly better as it scans both profiles of a seed rather than one profile. A classification accuracy of 93.00% was obtained using our hardware setup.