[Paper Review] Comparing YOLOv8 and Mask R-CNN for instance segmentation in complex orchard environments

Ranjan Sapkota, Dawood Ahmed+1 more|arXiv (Cornell University)|Dec 13, 2023

Smart Agriculture and AI106 references8 citations

TL;DR

The study compares YOLOv8 (one-stage) and Mask R-CNN (two-stage) for instance segmentation in two orchard datasets, finding YOLOv8 superior in accuracy and speed across both datasets.

ABSTRACT

Instance segmentation is an important image processing operation for agricultural automation, providing precise delineation of individual objects within images and enabling tasks such as selective harvesting and precision pruning. This study compares the one stage YOLOv8 model with the two stage Mask R CNN model for instance segmentation under varying orchard conditions across two datasets. Dataset 1, collected in the dormant season, contains images of apple trees without foliage and was used to train multi object segmentation models delineating branches and trunks. Dataset 2, collected in the early growing season, includes canopy images with green foliage and immature apples and was used to train single object segmentation models delineating fruitlets. Results showed YOLOv8 outperformed Mask R CNN with higher precision and near perfect recall at a confidence threshold of 0.5. For Dataset 1, YOLOv8 achieved precision 0.90 and recall 0.95 compared to 0.81 and 0.81 for Mask R CNN. For Dataset 2, YOLOv8 reached precision 0.93 and recall 0.97 compared to 0.85 and 0.88. Inference times were also lower for YOLOv8, at 10.9 ms and 7.8 ms, versus 15.6 ms and 12.8 ms for Mask R CNN. These findings demonstrate superior accuracy and efficiency of YOLOv8 for real time orchard automation tasks such as robotic harvesting and fruit thinning.

Motivation & Objective

• Motivate accurate instance segmentation for agricultural automation tasks like selective harvesting and precision pruning in orchards.
• Evaluate the performance of one-stage YOLOv8 versus two-stage Mask R-CNN under different orchard conditions.
• Assess precision, recall, and inference times to determine suitability for real-time orchard robotic applications.
• Explore how dataset characteristics (dormant-season trees without foliage vs. growing-season canopies) affect model performance.

Proposed method

• Compare one-stage YOLOv8 and two-stage Mask R-CNN for instance segmentation.
• Use Dataset 1 (dormant-season, apples without foliage) to train multi-object segmentation models for branches and trunks.
• Use Dataset 2 (early growing season, foliage and fruitlets) to train single-object segmentation models for fruitlets.
• Evaluate precision and recall at a confidence threshold of 0.5.
• Measure inference times for both models (in ms).
• Analyze performance differences across datasets to inform real-time orchard automation tasks.

Experimental results

Research questions

• RQ1 Does YOLOv8 provide higher precision and recall than Mask R-CNN for multi-object segmentation in dormant-season orchard images?
• RQ2 Does YOLOv8 outperform Mask R-CNN for single-object fruitlet segmentation in canopy images with foliage and immature fruit?
• RQ3 How do inference times compare between YOLOv8 and Mask R-CNN in these orchard scenarios?

Key findings

• For Dataset 1, YOLOv8 achieved precision 0.90 and recall 0.95 versus Mask R-CNN with precision 0.81 and recall 0.81.
• For Dataset 2, YOLOv8 achieved precision 0.93 and recall 0.97 versus Mask R-CNN with precision 0.85 and recall 0.88.
• Inference times were lower for YOLOv8 (10.9 ms and 7.8 ms) compared to Mask R-CNN (15.6 ms and 12.8 ms).
• YOLOv8 demonstrated superior accuracy and efficiency for real-time orchard automation tasks such as robotic harvesting and fruit thinning.