Abstract

This paper reports on a current project to develop a prototype system
for the automatic identification and quantification of potato defects based on
machine vision. The system developed uses off-the-shelf hardware, including a
low-cost vision sensor and a standard desktop computer with a graphics processing
unit (GPU), together with software algorithms to enable detection, identification
and quantification of common defects affecting potatoes at near-real-time frame
rates. The system uses state-of-the-art image processing and machine learning
techniques to automatically learn the appearance of different defect types. It also
incorporates an intuitive graphical user interface (GUI) to enable easy set-up of the
system by quality control (QC) staff working in the industry.