Deep learning supported machine vision system to precisely automate the wild blueberry harvester header

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Manual adjustment of wild blueberry harvester height is labor-intensive and affected by field variations.
• Operator fatigue and suboptimal harvester positioning can reduce fruit yield and create safety hazards.

Purpose of the Study:

• To develop and evaluate a deep learning-based machine vision system for automated wild blueberry harvester height adjustment.
• To reduce operator stress and improve harvesting efficiency through precise, real-time control of harvester head position.

Main Methods:

• Utilized the OpenCV AI Kit (OAK-D) with the YOLOv4-tiny deep learning model and Python for fruit height detection.
• Developed a system to automatically adjust the harvester's header picking teeth rake position based on detected fruit height.
• Statistically evaluated system accuracy using R² (coefficient of determination) and σ (standard deviation) compared to manual adjustment.

Main Results:

• The automated system demonstrated higher accuracy (R² = 72%, σ = 2.1 cm) compared to manual adjustment (R² = 43%, σ = 2.3 cm) in matching harvester head position to fruit height.
• The system effectively controlled harvester head height in weed-free environments.
• Further development is needed for operation in weedy field sections.

Conclusions:

• The deep learning-supported machine vision system offers a promising solution for automated wild blueberry harvesting.
• The system reduces operator stress and enhances safety by automating harvester height adjustments.
• Future work should focus on improving system performance in challenging, weedy conditions.