Research on persimmon fruit diameter accurate detection method based on improved RCNN instance segmentation algorithm

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Inaccurate fruit recognition and diameter detection hinder persimmon inspection.
• Existing methods struggle with overlapping or occluded fruits.

Purpose of the Study:

• To develop a novel algorithm for accurate persimmon recognition and fruit diameter detection.
• To improve segmentation accuracy and reduce measurement errors in persimmon inspection.

Main Methods:

• Utilized Mask R-CNN with instance segmentation, incorporating cropping, morphological processing, and concave point segmentation.
• Integrated template matching for image recognition and addressed fruit sticking issues.

Main Results:

• Achieved a mean Average Precision (mAP) of 94.25%, an 8.05% improvement over the original algorithm.
• Increased Mean Intersection-over-Union (MIoU) by 18.5% and reduced maximum relative error in diameter measurement to 1.3%.

Conclusions:

• The improved Mask RCNN algorithm enhances persimmon recognition and diameter measurement accuracy.
• Provides valuable insights for intelligent inspection, yield estimation, and mechanized picking in agriculture.