Comparative analysis of stomatal pore instance segmentation: Mask R-CNN vs. YOLOv8 on Phenomics Stomatal dataset

Area of Science:

• Plant Science
• Computer Vision
• Bioinformatics

Background:

• Accurate stomata pore analysis is crucial for understanding plant physiology and responses to environmental changes.
• Instance segmentation methods offer potential for automated analysis of plant structures.
• Existing methods face challenges with low-resolution and imperfect image data.

Purpose of the Study:

• To compare the performance of Mask R-CNN and YOLOv8 for stomata pore instance segmentation.
• To introduce and utilize the novel PhenomicsStomata dataset for stomata pore analysis.
• To evaluate the impact of advanced image preprocessing on segmentation accuracy.

Main Methods:

• Comparative analysis of Mask R-CNN and YOLOv8 algorithms.
• Development and application of the PhenomicsStomata dataset.
• Implementation of image enhancement using the Lucy-Richardson Algorithm.
• Evaluation based on accuracy, precision, and recall metrics.

Main Results:

• YOLOv8 demonstrated superior performance compared to Mask R-CNN in stomata pore segmentation.
• YOLOv8 achieved higher accuracy in calculating stomata pore dimensions.
• Image enhancement techniques improved the quality of stomata pore analysis.

Conclusions:

• YOLOv8 is a highly effective method for stomata pore instance segmentation.
• Advanced preprocessing enhances the reliability of plant structure analysis.
• Findings provide a foundation for improved plant physiology research and image analysis techniques.