Universidad de Cantabria | Spain
Pedro Andres Heredia | Adrian Herrera | Adrian Monge
Food waste is, as of today, one of the main sustainability issues society faces. United Nations Sustainable Development Goals [https://www.un.org/sustainabledevelopment/sustainable-development-goals/] acknowledge this in Goal 12.3, where they target halving nowadays quantities by 2030. Solely in the EU for 2012, 88 million tonnes (Mts) of food waste were measured, an equivalent to 143 billion euros and 182 Mts of CO2 emissions on the whole food chain. Studies suggest this is a social issue, with 53% of the total contributed by households. There are different initiatives for reducing food waste around Europe, among them the current REFRESH project [https://eu-refresh.org/], successor of the successful FUSIONS project [https://www.eu-fusions.org/], is the main European project developing research and solutions concerning food waste. On the other hand, Machine Learning, specifically Deep Learning, since the wide release of stable frameworks such as TensorFlow [https://www.tensorflow.org/] in 2015, coupled with the recent advancements in computing power due to Graphics Processing Units (GPUs) and just now Tensor Processing Units (TPUs), has gained a special place as one of the most versatile collection of methods for achieving predictive analytics. Cloud providers such as Amazon or Google provide Machine Learning solutions [see https://aws.amazon.com/machine-learning/ or https://cloud.google.com/ml-engine/] without the need to bother with environments, letting the developers focus on the actual models. Furthermore, Internet of Things has brought the possibility to acquire enormous amounts of data and feed it into the mentioned models. Small circuit boards such as Raspberry Pi [https://www.raspberrypi.org/] or Arduino [https://www.arduino.cc/], complemented with distinct sensors (temperature, weight, light, …) allow for building networks of small systems that are then connected to the cloud as a continuous monitoring entity. Finally, Computer Vision is recently seeing an upswing in popularity due to GPU power and Deep Learning techniques applied to it; in fact, the top most popular Computer Science conference is “Computer Vision and Pattern Recognition” [http://www.guide2research.com/topconf/]. It allows for identifying from pedestrians in an intelligent car to faces in family photos. With binSight, we try to integrate all these technologies into an innovative solution for reducing food waste and making supply chains smart. We work in a market where competency is scarce, and opportunities derived from current issues are in our favour. Our market analysis suggests a need for new methods of food waste measurement and reduction, with interest around the whole European Union. The business model defines four main customer segments: catering companies, governments, social entities and compost companies, all of them agents of the food supply chain; we provide different cloud-based services tailored to their needs. The technical feasibility shows that the tools needed for the project are already available, and the economic projection showcases a favourable net gain increase at short to mid term. Finally, we have also devised a project structure for the 5 months of following work.