A consortium of the TU Delft AgTech Institute, Delphy, Improvement Center and Birds.ai is investigating how waste in the soft fruit chain can be reduced with the help of artificial intelligence. To do this, the researchers are looking at a better 'supply chain' for strawberries, such as being able to better predict the optimal harvest date. The reason for the collaboration is that food waste is high, especially in soft fruit, because the products are fragile and have a short shelf life.

Globally, food for human consumption accounts for one third of all food grown. That percentage is considerably higher for soft fruit. For strawberries, for example, the moment of harvesting is decisive. Once the strawberries are picked, it's a race against time to keep them as fresh as possible. Determining whether a strawberry is suitable for harvesting is human work: that decision also must be taken very quickly.

Growers experience high costs if the fruit does not meet customer requirements. By using objective measurements during cultivation, harvest and throughout the chain, soft fruit growers and suppliers can better serve consumers and minimize waste. It is very important for growers that risks such as a short shelf life, surpluses, shortages, and poor quality are tackled properly. These risks create waste and lead to negative perceptions among customers. The TU Delft AgTech Institute, Delphy, Improvement Center, Birds.ai and the Innovation Pact of Greenport West-Holland joined forces to tackle this problem. 

Cameras focus on strawberries 

The research focuses on developing a model that helps growers and their employees make better and more reliable predictions for the optimal harvest time. This information can also be used to predict which distribution channel a specific strawberry is suitable for. For example, overripe fruits are not suitable for fresh produce, but are suitable for making cakes, for example. The project combines the scientific knowledge of TU Delft and Delphy's practical knowledge of cultivation and application.

In Delphy's greenhouses in Bleiswijk, various cameras are set up around several cultivation trays. The cameras show all strawberries from one side of the two cultivation trays. Data from strawberries once harvested, together with the camera images, are used by TU Delft for analysis and for the development of a harvesting model. Researcher Junhan Wen (TU Delft): “The aim of the model is to predict how many days it will take before a strawberry is ready to harvest. This is done by means of computer vision, deep learning, and other specially developed algorithms. If these practical obstacles have been overcome, we expect to be able to make a good prediction of the best harvest moment and possibly also the quality of the product at the beginning of 2022. In the second half of the year, we will work on coordination with the rest of the chain.”

Augmented reality

With such a model, it is possible in the future that employees of strawberry farms will be equipped with augmented reality glasses: these can then show which strawberries can be harvested for which distribution channel. Before we get to that point, some practical issues still need to be resolved. Researcher Stijn Jochems (Delphy): “Each strawberry is given its own name or code. Because the strawberries and trusses move (for example during harvesting), the camera can temporarily lose the strawberry. Then the right code must be put back with the right strawberry to keep following him again. In addition, we are working on a way to obtain information about all strawberries in a greenhouse, for example by moving the camera through the greenhouse. So, we use technology to reduce food waste in the chain.”

Partners

The supporting parties in this research are: Hagelunie, Achmea Agro, LTO Glastuinbouw, Hunsballe Grønt, Fruitmasters, Octinion, Greenport West-Holland en de Topsector Tuinbouw en Uitgangsmaterialen.
 

/* */