YOLOvFIS: a YOLO network for fruit instance segmentation

You Only Look Once (YOLO) networks have demonstrated significant advancements in fruit detection and instance segmentation. Nonetheless, challenges remain in managing dense scenes, overlaps, occlusions, generalization difficulties, dynamic environments, computational and deployment limitations, specifically concerning the application of YOLOv5 to YOLOv11 for fruit instance segmentation. This article tackles these issues by constructing a fruit image dataset with segmentation annotations, having 11,546 images with six classes, and developing a You Only Look Once version Fruit Instance Segmentation (YOLOvFIS) network for fruit detection and instance segmentation. The results demonstrate that integrating the backbone of YOLOvFIS into YOLO versions of YOLOv5n, YOLOv6n, YOLOv7t, YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv11n improved the networks, surpassing the performance of the original networks. The improved YOLOv9t outperformed other improved YOLO networks, even though it had a longer overall detection time. The parameters (params) and giga floating point operations per second (GFLOPs) of YOLOvFIS’s entire network are less than other YOLO networks. The mean average precision (mAP) of YOLOvFIS is 1.6%, 1.0%, 1.0%, 1.0%, 0.3%, 0.8%, and 0.4% more accurate than YOLOv5n, YOLOv6n, YOLOv7t, YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv11n, respectively, for fruit detection. It also exhibits superior mAP by 1.5%, 1.0%, 0.9%, 1.6%, 0.6%, 0.7%, and 0.7% compared to YOLOv5n, YOLOv6n, YOLOv7t, YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv11n, respectively, in the context of fruit instance segmentation. Regarding total detection time, YOLOvFIS is 14%, −7.2%, 112.2%, 0.0%, 84.8%, 9.9%, and 7.5% faster than YOLOv5n, YOLOv6n, YOLOv7t, YOLOv8n, YOLOv9t, YOLOv10n, and YOLOv11n, respectively. As a result, the network of YOLOvFIS is robust for generalization, appropriate for real-time applications, robotic integration, and low-power computing devices. Lastly, the obtained findings can serve as technical assistance and suitable references for YOLO-based fruit detection and instance segmentation.