Machine Learning For Product Quality Inspection

Abstract

Pistachio is widely consumed as food which requires the nut to be cracked before usage. Automating the cracking process requires quality control which can be done visually. A system developed to perform this task provides over 98% accuracy. First, the pistachio is segmented from the background using support vector regression (SVR) or deep convolutional networks. Then a classification method based on deep neural networks (FistikNet) is employed to decide the quality of the crack. A large data set is collected to train and test the system.