Optical quality control using deep learning

Optical quality control is still often performed by people and always carries the risk of human error. A modern approach in order to solve this issue is the usage of artificial intelligence to boost performance and reliability. This paper focuses on implementing a prototype for optical quality control based on the YOLOv3 algorithm. This is a state-of-the-art object detection system that uses deep learning to detect different classes of objects within an image. Instead of different kinds of objects, the classes in this prototype were different quality levels of a strawberry. The dataset for this task was gathered by taking photos and using images from the internet. The strawberries on these images were labeled and fed to the YOLOv3 algorithm for training. Despite the poor detection rate, the results showed that it is generally possible to use such systems for detecting different quality levels of products.