[Paper Review] Comparing YOLOv11 and YOLOv8 for instance segmentation of occluded and non-occluded immature green fruits in complex orchard environment

Ranjan Sapkota, Manoj Karkee|arXiv (Cornell University)|Oct 24, 2024

Horticultural and Viticultural Research7 citations

TL;DR

The paper compares YOLOv11 and YOLOv8 for instance segmentation of occluded and non-occluded immature green fruits in a commercial apple orchard, highlighting YOLO11m-seg as the top performer for mask and box mAP@50 while YOLOv8n-seg offers fastest inference.

ABSTRACT

This study conducted a comprehensive performance evaluation on YOLO11 (or YOLOv11) and YOLOv8, the latest in the "You Only Look Once" (YOLO) series, focusing on their instance segmentation capabilities for immature green apples in orchard environments. YOLO11n-seg achieved the highest mask precision across all categories with a notable score of 0.831, highlighting its effectiveness in fruit detection. YOLO11m-seg and YOLO11l-seg excelled in non-occluded and occluded fruitlet segmentation with scores of 0.851 and 0.829, respectively. Additionally, YOLOv11x-seg led in mask recall for all categories, achieving a score of 0.815, with YOLO11m-seg performing best for non-occluded immature green fruitlets at 0.858 and YOLOv8x-seg leading the occluded category with 0.800. In terms of mean average precision at a 50\% intersection over union (mAP@50), YOLOv11m-seg consistently outperformed, registering the highest scores for both box and mask segmentation, at 0.876 and 0.860 for the "All" class and 0.908 and 0.909 for non-occluded immature fruitlets, respectively. YOLO11l-seg and YOLOv8l-seg shared the top box mAP@50 for occluded immature fruitlets at 0.847, while YOLO11m-seg achieved the highest mask mAP@50 of 0.810. Despite the advancements in YOLO11, YOLOv8n surpassed its counterparts in image processing speed, with an impressive inference speed of 3.3 milliseconds, compared to the fastest YOLO11 series model at 4.8 milliseconds, underscoring its suitability for real-time agricultural applications related to complex green fruit environments. (YOLOv11 segmentation)

Motivation & Objective

• Assess instance segmentation performance of YOLO11 and YOLOv8 on occluded and non-occluded immature green apples in a commercial orchard.
• Compare precision, recall, F1, and mAP@50 across configurations.
• Evaluate model complexity (parameters, GFLOPs, layers) and training/inference efficiency.
• Provide recommendations for real-time agricultural deployment.

Proposed method

• Dataset collection of 991 RGB images from a Scifresh apple orchard using a robotic imaging platform.
• Manual annotation of occluded vs non-occluded fruitlets and dataset formatting for YOLO11/YOLOv8.
• Train and evaluate YOLO11-seg and YOLOv8-seg configurations under identical hyperparameters (640x640 input, batch 8, IOU 0.7, 300 epochs, deterministic seed).
• Compute metrics including MIoU, AP, mAP@50, mAR, and F1, plus GFLOPs, parameters, and convolution layers.
• Assess training time and image processing speeds (preprocess, inference, postprocess).

Figure 1: Architecture diagram of YOLOv8 algorithm: YOLOv8 advances real-time object detection with its innovative backbone and anchor-free Ultralytics head, optimizing detection accuracy and speed across various tasks (primary image source: https://yolov8.org/what-is-yolov8/).

Experimental results

Research questions

• RQ1 Which YOLO-11 vs YOLOv8 configuration provides superior instance segmentation accuracy for occluded and non-occluded immature green fruits?
• RQ2 How do model complexity and inference speed trade off with segmentation performance in complex orchard scenes?
• RQ3 Which configuration is most suitable for real-time agricultural applications?
• RQ4 How do occluded and non-occluded fruitlets influence precision, recall, and mAP@50 across models?

Key findings

• YOLO11n-seg achieved the highest overall mask precision (0.831).
• YOLO11m-seg and YOLO11l-seg achieved top precision for non-occluded (0.851) and occluded (0.829) fruitlets respectively.
• YOLO11x-seg led in mask recall across categories (0.815); YOLO11m-seg had best non-occluded recall (0.858).
• YOLOv8n-seg had the fastest inference in the study (3.3 ms) overall.
• YOLO11m-seg achieved highest mAP@50 for All class box (0.876) and non-occluded box (0.908); mask mAP@50 for All (0.860) and non-occluded (0.909).
• For occluded fruitlets, YOLO11m-seg exceeded others in mask mAP@50 (0.810); box mAP@50 tied between YOLO11l-seg and YOLOv8l-seg (0.847).

Figure 2: Simplified architecture diagram of YOLO11 algorithm: YOLO11 builds on YOLOv8’s architecture with refined detection and segmentation efficiency and accuracy over benchmark dataset.