Leaf Area Estimation by Semantic Segmentation of Point Cloud of Tomato Plants

Takeshi Masuda; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2021, pp. 1381-1389

Growth monitoring is an essential task in agriculture for obtaining good crops and sustainable management of cultivation. Though it is essential, it is also a hard task requiring much labor and working time, and many automation approaches have been proposed. We present an attempt to estimate the leaf area of the tomatoes grown in a sunlight-type plant factory. We scanned tomato plants by an RGB-D sensor that moves vertically to scan one side of the plants from the pathway. We built a point cloud by merging the scanned data, and we segmented it into four classes (Stem, Leaf, Fruit, and Other) based on annotation. With a limited amount of data, we estimated the stem from Stem points, and from the number of Leaf points around the stem, we estimate the leaf area of a specific tomato plant in a plant factory with the relative error of about 20%.