Lightweight CFE-YOLOv8s-Seg Model for Tomato Ripeness Instance Segmentation

Ngoc-Linh Nguyen, Quang-Anh Duc Nguyen, Thai Dinh Kim, Manh-Quan Nguyen, Truong-Giang Le Bui

Abstract: In smart agriculture, precision farming enhances productivity, optimizes crop quality, and minimizes time and economic losses. Monitoring tomato ripeness is crucial, as the flavor profile of tomatoes is influenced by their ripeness level. In this study, we introduce the C2f-Faster-EMA block, replacing the C2f-Faster-EMA block for the C2f modules in the original YOLOv8s-seg to develop the CFE-YOLOv8s-seg model for real-time segmentation of tomato ripeness stages (green, half-ripe, and ripe). The experimental results indicate that the proposed model achieves 85.8% precision, 84.7% recall, and a 90.1% mAP50 at a 50% IoU threshold, with a lower mAP50-95 of 73.7%. The proposed model has a compact architecture with 9 million parameters, 23.73% fewer than the original YOLOv8s-seg. The experimental results on real-world images demonstrate that the proposed model effectively segments tomato ripeness across diverse contexts.

External IDs:doi:10.1007/978-3-032-00267-9_17