Use Case for Smart Retail: Print the price label by automated food recognition with machine learning
The implementation principle is valid for other markets, including industrial automation, medical or other markets.
Fruit classification on a retail weight scale to define the price of the fruit
Challenges:
Fruit classification in itself is not a very big challenge today, but developing the real use case within a complex environment is:
- What if the fruit is in a plastic bag?
- What if you cannot control lighting in the store?
- How do you recognize the differences between similar fruit, like tangerines and oranges?
- What if you do not have sufficient training data?
Implementation:
The fruit is recognized based on vision with an embedded camera in combination with the weight parameter from the scale. Both vision data and weight values are treated jointly for the implementation on an optimized ResNet-like network. Training of the network is realized with pictures of fruit created with synthetic data instead of real images. Synthetic data is artificially manufactured and avoids the need of a big amount of training data. The Neural network is running on an embedded module with an NXP i.MX 8M Plus processor with integrated Neural Processing Unit (NPU). The product and price details are displayed on a touch screen.
Results:
Results are remarkable and the fruit is recognized accurate and in less than 4ms on an embedded solution with integrated machine learning capabilities in a complex environment of a retail shop.
Interested in this solution? Get in touch with us.
Inference comparison on the different processing engines of the NXP i.MX 8M Plus Application Processor for the fruit classification application
More use cases
silica content library/home/solutions/technologies/artificial-intelligence-machine-learning/nxp-machine-learning-solutions/nxp-machine-learning-use-case-html
nxp-machine-learning-use-case-html
Factory Automation
Human Machine Interface
Smart City
Smart Appliances
