Smart food waste management : embedded machine learning vs cloud based solutions

In Switzerland, 2.8 million tons of food are lost or wasted across all stages of food production - every year. This equates to approximately 330 kg of food waste per person. By analysing and classifying discarded food with a smart waste analysis system combined with machine learning, valuable insights can be gained and the amount of wasted food can be significantly reduced. In this paper, we present how we have developed an embedded system which helps to solve this task. The embedded system operates in a decentralized manner: It captures an image every time food is thrown into a bin. The discarded food is identified and classified with machine learning algorithms. This provides a detailed insight into the structure of food waste for customers, e.g. restaurants or canteens. We implemented the machine learning algorithm directly on the embedded systems control unit. We found that running machine learning directly on embedded devices has many advantages compared to running them in the cloud: We saved significant amounts of cloud storage and reduced power consumption by up to a factor 100. In addition, privacy was increased and required bandwidth reduced because only the machine learning results are forwarded to the cloud, not the full data