Viewpoint Analysis for Maturity Classification of Sweet Peppers

Area of Science:

• Agricultural technology
• Computer vision
• Machine learning

Background:

• Accurate classification of sweet pepper maturity is crucial for optimal harvesting and quality control in commercial greenhouses.
• Existing methods may not fully account for variations in imaging conditions, such as camera angle and fruit presentation.

Purpose of the Study:

• To evaluate the impact of camera viewpoint and fruit orientation on the performance of a sweet pepper maturity classification algorithm.
• To determine optimal imaging strategies for improving automated maturity assessment.

Main Methods:

• Collected two datasets: 789 Red Green Blue (RGB) images and 417 Red Green Blue-Depth (RGB-D) images of sweet peppers.
• Utilized a random forest algorithm for maturity level classification.
• Compared classification accuracy across different single camera viewpoints, combined viewpoints, and various fruit orientations against manual classification.

Main Results:

• The bottom camera viewpoint demonstrated the highest accuracy for single-view maturity classification.
• Combining two viewpoints improved classification accuracy by 25% for red peppers and 15% for yellow peppers compared to single viewpoints.
• Classification performance was found to be highly sensitive to the fruit's orientation on the plant.

Conclusions:

• Camera viewpoint and fruit orientation are critical factors influencing automated sweet pepper maturity classification.
• Utilizing multiple camera viewpoints, particularly the bottom view, enhances classification accuracy.
• Future algorithms should consider fruit orientation for more robust maturity assessment.