Sponsored by:
ひろしま型スマート農業推進事業「ぶどうの⼤規模経営の実現に向けた効率的な作業体系 の構築」（2022–2024年度）「ぶどうAI等級判定ステーション」
Grape grading is an essential but labor-intensive process that traditionally relies on the experience and subjective judgment of skilled farmers. Evaluating grape quality based on size, color, density, and weight can be challenging, especially for new growers, and may result in inconsistent grading outcomes.
To address this issue, we developed an AI-based grape grading station that integrates computer vision, deep learning, and IoT technologies. The overall prediction flow of the grading model begins with the input of four images captured from different angles of a grape bunch. These images are first processed using an instance segmentation model to extract precise grape masks and isolate the grape bunch from the background. The segmented images, together with the measured weight information, are then fed into a CNN-based grading classification model, which predicts the final grape quality grade.
The station consists of a custom gripper for holding grape bunches, a custom-built weight scale using load-cell based sensor with Digital-to-Analog (HX711) converter attached to the gripper, a user interface display powered by Raspberry Pi 5, and four cameras positioned at different angles, which also controlled by Raspberry Pi 5. Users simply hang a harvested grape bunch on the station, and within a few seconds, the system captures images from four viewpoints, measures the weight, and automatically evaluates grape quality. The grading results are immediately displayed on the screen and stored in the data management system.
The initial open-box station achieved a grading accuracy of over 80%, but performance was sometimes affected by varying lighting conditions. To address this limitation, a new enclosed station design, structured as a black box, was developed to provide stable illumination and improve grading reliability. Using this new enclosed design manages to achieve grading accuracy of 90%.
Publication
- Muhammad Faris Kamarudzaman, Prawit Buayai, Yin Suan Tan, Latifah Munirah Kamarudin, Xiaoyang Mao, “Versatile and Easy-to-Operate Grading System for High-Grade Table Grapes: Leveraging Deep Learning, Computer Vision, and IoT,” 2024 International Conference on Cyberworlds (CW), pp.264-271, 2024-10, doi: 10.1109/CW64301.2024.00037.
Patent
特開2025-144414(P2025-144414A)（特願2024-044171）、2024年3月19日出願、【発明の名称】等級判定システム、保持ボックス、等級判定方法および等級判定プログラム、茅 暁陽、 ブアヤイ プラウィット