TIGP (SNHCC) -- Non-destructive classification of melon sweetness levels using segmented rind properties based on semantic segmentation models
- 講者胡氏妝 教授 (淡江大學資工系)
邀請人：TIGP (SNHCC) - 時間2023-11-13 (Mon.) 14:00 ~ 16:00
- 地點資訊所新館106演講廳
摘要
Melon (Cucumis melo L.) is one of the most consumable crops globally because of its high market demand and is one of the top 10 fruits in the international market. In Taiwan, sweet melon is a leading product that generates high profits to farmers, mainly in the spring and autumn seasons. However, the emergence of the coronavirus epidemic has significantly affected the global food supply since 2019. Moreover, the greenhouse effect and global warming adversely affect agricultural processes and products. Therefore, it is crucial to develop an assistance system to support farmers in harvest management and improve the quality of melon products. The appearance and sweetness of melon are two important factors that can influence consumer decision-making when buying melon in a market. When choosing a melon, consumers prefer to buy one with a high sweetness content. Unlike small fruits such as bananas, mangoes, and apples, melon is a large fruit with a thick rind. Therefore, determining the sweetness of a melon by its smell or external appearance is difficult. On the other hand, image segmentation plays a vital role in computer vision applications because it influences critical tasks such as image analysis, feature calculation, object detection, and classification. It has been widely used in diverse fields, such as biomedicine, robotics, autonomous vehicles, agriculture.
In this talk, I would like to introduce an efficient and accurate model for classifying melon sweetness. This classification is achieved by analyzing the characteristics of the rind through a semantic segmentation approach based on U-Net architecture. We first developed a semantic segmentation approach to efficiently isolate the net pattern on melon rinds in images. Then, we extracted various rind features from the segmented objects of the melon’s rind, including net density, net thickness, and rind color, which are crucial factors in grading melon quality. Subsequently, we utilized these extracted features as inputs to develop the melon sweetness level classification model.
In this talk, I would like to introduce an efficient and accurate model for classifying melon sweetness. This classification is achieved by analyzing the characteristics of the rind through a semantic segmentation approach based on U-Net architecture. We first developed a semantic segmentation approach to efficiently isolate the net pattern on melon rinds in images. Then, we extracted various rind features from the segmented objects of the melon’s rind, including net density, net thickness, and rind color, which are crucial factors in grading melon quality. Subsequently, we utilized these extracted features as inputs to develop the melon sweetness level classification model.
BIO
Dr. Trang (also known as Trang-Thi Ho in academic publications) received her Ph.D in Computer Science at National Taiwan University of Science and Technology in 2020. From January 2021 to August 2022, she worked as a Postdoctoral Researcher at Research Center for Information Technology Innovation, Academia Sinica Academia Sinica). She is currently an Assistant Professor at Department of Computer Science and Information Engineering, TamKang University. Her research interests include machine learning, AI research and development, data science, and digital image processing.