Bicici, Ufuk CanTrimmel, MatthiasRiegler-Nurscher, PeterDörr, JörgSteckel, Thilo2025-02-042025-02-042025978-3-88579-802-6https://dl.gi.de/handle/20.500.12116/45674Soybeans are crucial in agriculture and industry, serving as a key source of protein and oil for food, feed, and various applications. Measuring individual soybean seed properties, such as size, color, and texture, aids in predicting batch parameters. In a conveyor belt setup, images captured by an RGB camera are analyzed using computer vision techniques, such as object detection and segmentation, to identify whole beans, which are then used for size measurements. For such a supervised machine learning approach, annotated data is essential. To address the challenge of manual data labeling, we propose an automated annotation system for identifying beans on a conveyor belt. Using the Segment Anything Model (SAM), contours are extracted, and ellipses are fitted to approximate bean shapes. In a dataset of 17,386 images, SAM generated over 3.5 million contours, with approximately 265,000 annotated as individual beans. Preliminary results from a deep learning-based ellipse detection model and a panoptic segmentation model, both trained on the generated soybean dataset, are presented.ensoybeansoybean size detectionmachine learningcomputer visiondeep learningText/Conference Paper10.18420/giljt2025_172944-76822944-7682