Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 01, 2022 - June 04, 2022
The goal of this research is to develop a method to evaluate the quality of chrysanthemum by the degree of stem bending. Criteria for the quality assessment include the state of flowering and the curvature of the stem. This research deals with an image-based quality assessment using the curvature of the stem. In order to measure the curvature of the stem, it is necessary to extract the stem regions from the image of chrysanthemum. Although a previous method used the XYZ color space for stem detection, its accuracy is lower because the colors of the stems and the leaves are similar. In this research, we perform stem detection by an image processing-based and a deep learning-based methods and compare their detection accuracy. We them define a feature that represents the curvature of the stems to judge whether the quality is good. The accuracy of stem detection by deep learning-based method is better than that by the image processing-based one. The accuracy of quality assessment is 0.869.
