Automatic imaging system for tulip bulb detection and sorting using deep learning

This work aims to develop and test an imaging system for automatic detection of bulb cleaning quality and potential sorting of bulbs according to size and cleaning quality. Line scan images of bulbs were collected from conveyor belts using a custom-built camera box. Each camera box was connected to a high-performance computing (HPC) server which processes the acquired images using an instance segmentation algorithm. The developed algorithm can successfully detect and classify bulbs according to cleaning quality and size with an average F1-score of 0.87 and on a commercial bulb sorting setup and an average F1-score of 0.94 in a controlled setup. Tests were also performed to verify if the imaging system can be coupled with a bulb sorting system however results show that further work has to be done in order to achieve better positioning for effective sorting.