Plant leaf veins coupling feature representation and measurement method based on DeepLabV3.

Area of Science:

• Computer Vision
• Plant Science
• Image Analysis

Background:

• Accurate segmentation and measurement of plant leaf veins are crucial for botanical research.
• Existing methods face challenges including slow processing, partial occlusion, and low measurement accuracy.

Purpose of the Study:

• To develop an efficient and accurate method for plant leaf vein segmentation and parameter measurement.
• To address limitations of current techniques in speed, occlusion handling, and precision.

Main Methods:

• Utilized DeepLabV3+ with MobileNetV2 as the feature extraction network for efficient segmentation.
• Implemented the Convex Hull-Scan method for leaf vein repair and a novel Floodfill MorphologyEx Medianblur Morphological Skeleton (F-3MS) algorithm for skeleton line refinement.
• Compressed network model parameters to reduce computational load.

Main Results:

• Achieved high segmentation accuracy with mean intersection over union (MIoU) of 81.50% and mean pixel accuracy (mPA) of 92.89%.
• Reached an average segmentation speed of 9.81 frames per second.
• Demonstrated high accuracy in measuring leaf vein length (96.3642%) and width (96.1358%) after significant model parameter compression (89.375%).

Conclusions:

• The proposed DeepLabV3+ based method significantly improves plant leaf vein segmentation and measurement.
• The optimized approach offers a balance of speed, accuracy, and efficiency for analyzing leaf vein structures.
• This technique provides a valuable tool for plant phenotyping and digital botany applications.