Learning System for Estimating the Area Ratio of Flowers-fruit and Leaves-stems from Tomato Plant Images Using Semi-supervised Learning

Abstract

The authors constructed a learning system for estimating the area ratio of flowers-fruit and leaves-stems from tomato-cultivation images as one of the indices for diagnosing the plant vigor of cultivation. The annotation work for object labeling, which is performed as an image analysis preprocessing, incurs a massive computation cost. Therefore, this study proposed a semi-supervised learning method that combines classification using unsupervised learning along with a backpropagation and segmentation model through object detection using supervised learning. As a result, the area ratio of leaves-stems to flowers-fruit is recognized with a high recognition rate, suggesting the effectiveness of the proposed system with a reduced computational cost.