Recognition of Tomato Fruit Regardless of Maturity by Machine Learning Using Infrared Image and Specular Reflection

This paper presents a tomato fruit recognition method using plant characteristics of tomato and infrared images. Labor shortage and aging are problems in Japanese agriculture field. We aim to realize automatic harvesting and production management system of tomato. For that, it is necessary to detect the position and maturity of tomato fruit. Tomato fruit shows high reflectance against infrared light. The specular reflection part of the tomato fruit in the infrared image is used as training data. The Tomato harvesting robot can focus only on tomato fruit in the harvestable range by using infrared image. We use the images acquired at the actual tomato greenhouse to evaluate this proposed method. As a result of machine learning, Precision is 0.940, Recall is 0.808, and F-measure is 0.868.