Automatic visual quality control of chocolate

Abstract

Automatic visual quality control of manufactured objects is a large industry in today s world, where a global market often means automatic production and inspection systems is a way to keep up with the competition. There already exists numerous systems for various tasks, but it s always possible to adapt methods to new areas. The process of visual inspection includes detection, feature extraction, classification and fault inspection of objects. This paper will focus on classification as a way to approve or reject objects, and will examine the applicability of neural networks with simple features as input vectors for this use with regards to chocolate. From the theory on the subject, a neural network has been implemented as a prototype in C#. The neural network itself seems suited for the task, but the success rate is highly dependent on the input features.