Master of machine learning Shebin Jose Jacob has created an automatic label inspector that can spot mistakes on printed labels.
It’s a super-simple setup, utilising Raspberry Pi 4 Model B and a 5MP security camera. But how does this lean, mean hardware team of two know what do? That’s where Shebin’s machine-learning prowess comes in.
Training the label inspector to spot mistakes
An AI model built in FOMO runs on Raspberry Pi. FOMO stands for Faster Objects, More Objects. It’s a newer object detection model which works faster than others to discern how many objects are in the frame, not just the size of those objects. With FOMO, there’s fear of missing… nope, can’t do it. You already get the joke.
Small and annoying to remove as they are, labels actually carry a lot of important product information. They can track and identify items with barcodes which are used by cashiers, the postal service and real-life human inspectors. They’re also crucial in inventory management. One tiny ink spill can render a barcode useless. Being placed a few centimetres off means a machine won’t be able to read the label at all.
Shebin’s AI model is capable of detecting ink spills and smudges, as well as recognising inverted labels and die-cut labels. There is room for the model to be enhanced to recognise more and work more robustly, but it’s already running at 97% accuracy.
A web interface sounds an alert if the FOMO system finds any of the defects it is trained to spot. You can choose whether this alert shuts the printing machine off completely until a human can come along and fix the problem, or if you’d like to continue printing but have the machine put the defective prints aside using a sorter.
Build your own label monitor
Edge Impulse has published a project page which walks you through setting up your hardware and building the machine-learning model that will monitor your label printing.