Google AI can be used to make baby food safer
The baby food industry is realizing the importance of artificial intelligence (AI), to manage quality control. Japanese food producer Kewpie Corp. says that it is using Google’s TensorFlow artificial intelligence, to quickly inspect ingredients that go into baby food products. This includes diced potatoes it uses in baby food, an ingredient that comes under most scrutiny in terms of safety since quality can greatly vary across batches of production and packaging. Kewpie tied up with analytics firm BrainPad Inc. to train machine learning algorithms to recognize the ingredients.
The TensorFlow system was “educated” to identify the clean and healthy ingredients, by making it recognize as many as 18,000 photos of potatoes, which included acceptable and useable, as well as defective and unusable potatoes. This allowed the artificial intelligence software to understand and flag any anomalies on the raw ingredients. The system is currently deployed at Kewpie’s Plant in the city of Tosu, in the Saga Prefecture on the island of Kyushu.
TensorFlow is an open-source software library for machine learning applications. The basic premise is to replicate how a human brain would recognize and identify objects and things. Since this is an open-source platform, developers can modify this for different use cases and environments too, and it can be used for applications as well as research.
In the case of the deployment in Kewpie’s plant, the system could detect potatoes that aren’t fit for use in baby food, even as they were running at high speeds on a conveyer belt. Takeshi Ogino, production division, Kewpie, says in an official blog post that the system was “near perfect” in terms of accuracy. It is designed to be an “anomaly detector” and not exactly a “sorter”, and will supplement the human workers at the factory and not replace them. At present, the TensorFlow systems flags off the potatoes it doesn’t deem fit, but there are still people waiting to inspect the potatoes afterwards to identify and remove any pieces that the AI system could not detect.
Kewpie currently handles more than 400 food ingredients at the Tosu Plant alone, and wants to expand the AI intelligence towards similar detection with eggs, grains etc., and is open to the idea of sharing the inspection system with other manufacturers too.