Automating stress detection in a vertical farming environment using an autoencoder for anomaly detection

Abstract

Feeding the world is becoming more challenging as the world’s population increases and many people are suffering from hunger. Vertical farming could be a way out, because it can improve the amount of food we can produce per square meter. Growy is a very successful company attempting to realize the vertical farming dream by automating the care for plants by robots. However, it is becoming difficult to make sure all plants are growing properly. Currently, the crops are inspected manually, but due to the increase of production, this is becoming very challenging. For this reason, the monitoring process should become automated. Insufficient nutrients can cause stress in microgreens, which will present by changing of the color of the leaves. Because of this noticeable color change, computer vision could be a good approach to solve the problem. To experiment with this, it is required that you can give plants stress and see the stress in images. We show that stress can be induced, but it is difficult to visualize due to constraints in the image-capturing system. When the system is improved, stress could be detectable using machine learning, anomaly detection, specifically. These results are a first step in automating health monitoring in microgreens growing in a vertical farm.