Smart Image-Based Deep Learning System for Automated Quality Grading of Phalaenopsis Seedlings in Outsourced...

Area of Science:

• Horticultural science and computer vision applications.
• Integration of artificial intelligence in agricultural quality control.

Background:

• Phalaenopsis orchids are a major floral export for Taiwan, requiring consistent quality.
• Manual seedling inspection is subjective, time-consuming, and hinders international competitiveness.
• Outsourcing the early seedling stage necessitates reliable quality assessment.

Purpose of the Study:

• To develop and validate a smart image-based deep learning system for automatic quality grading of Phalaenopsis potted seedlings.
• To replace manual inspection with an objective, efficient, and reliable automated system.
• To improve the consistency and accuracy of quality assessment for export.

Main Methods:

• Utilized deep learning models (YOLOv8, YOLOv10) for defect and root detection.
• Employed machine learning classifiers (SVM, Random Forest) for defect counting and grading.
• Implemented a dual-view imaging approach: top-view RGB-D for spatial structure and side-view RGB for leaf/root conditions.
• Developed two grading strategies: a three-stage hierarchical method and a direct grading method.

Main Results:

• The system achieved high F1-scores: 84.44% for the three-stage method and 90.44% for the direct method.
• RGB-D top-view images and optimal viewing angles significantly improved grading accuracy.
• Demonstrated the system's reliability for automated quality assessment and export inspection.

Conclusions:

• The developed deep learning system offers a reliable and efficient solution for automated Phalaenopsis seedling quality grading.
• This technology has strong potential to enhance the competitiveness of Taiwan's orchid export industry.
• Automated inspection can lead to more consistent quality and streamlined export processes.