Food Inspection using Deep Learning

Now is the time to automate, using AI to Reduce Waste, Save Money, Improve Quality

Artificial intelligence and specifically deep learning is getting more and more powerful. Deep learning solutions can serve very specific requirements and the food industry can benefit greatly. It really is a tool that offers an answer to puzzle of how to “Reduce Waste, Save Money, Improve Quality”

Deep Learning is a very efficient technology for classification purposes and is providing inspection solutions where more traditional machine vision algorithms have struggled or failed entirely.

Ease of use is also a key benefit. It identifies parts that contain defects, and can precisely pinpoints where they are in the image.

There are two modes, typically called unsupervised and supervised anomaly detection.

Unsupervised works by learning a model of what is a “good” sample (i.e. a sample without any defect). This is done by training it only with images of “good” samples. Then, the tool can be used to classify new images as good or defective and segment the defects from these images. By training only with images of good samples, the unsupervised mode is able to perform inspection even when the type of defect is not known beforehand or when defective samples are not readily available. In the example below, only a selection of ‘good’ wafers were trained and then when the production run was inspected the deep learning algorithms automatically flagged the ‘bad’ wafers (Image courtesy of Euresys):

Another example that shows the power of this technology is the inspection on coffee beans. There can be a wide variety of ‘foreign’ objects in the coffee beans but it would be impossible to know in advance what these may be. Using Deep Learning, the system is trained to know what coffee bean looks like and it will detect and segment anything that is not a coffee bean.

There is also the supervised anomaly detection mode and this is when images of both the good and the defect parts are trained. Then during production line inspection, the results will be classified as good or as a defect, where there can be a number of different classes of defect. The feedback from the production line can be very valuable and inform the production team of specific problems.

An example of defect detection and classification is shown below where salmon is inspected and the system report back if there are any defects and if the defects are classified as bones or melanin spots;


Image courtesy of PEKAT VISION


Another example is rice cake inspection, in this case two classifications either “GOOD” or “BAD”:

Image courtesy of PEKAT VISION

Now is the perfect time to explore what Deep Learning can offer you and how it can benefit your production processes.

Multipix Imaging offer a variety of technology options and are very happy to discuss your application with you on 01730 233332

Learn more by visiting our Deep Learning Technology area