TopoGeoFusion: Integrating object topology based feature computation methods into geometrical feature analysis to...

Area of Science:

• Computer Vision
• Agricultural Technology
• Image Processing

Background:

• Automated fruit quality assessment is crucial for market suitability and intelligent camera applications.
• Traditional grading methods can be labor-intensive and subjective.
• Computer vision offers a potential solution for objective and efficient fruit grading.

Purpose of the Study:

• To develop an automated system for grading gooseberries using smartphone-captured images.
• To address challenges in separating touching or overlapping fruits for accurate assessment.
• To introduce and evaluate the 'TopoGeoFusion' method for quality and maturity assessment.

Main Methods:

• Utilized smartphone-captured RGB images of 1697 Indian Star Gooseberries.
• Developed a novel 'TopoGeoFusion' method combining geometrical and topology-aware features.
• Employed multiple classifiers including Random Forest, SVM, Naive Bayes, Decision Tree, and KNN.
• Implemented a segmentation model for fruit detection and extraction, handling occluded objects.

Main Results:

• Achieved 100% accuracy in classifying gooseberries into reject, standard, and premium classes using Random Forest with a fusion feature model.
• The segmentation model demonstrated reliable fruit detection and extraction with an average mAP of 0.56.
• The TopoGeoFusion method proved effective in assessing grade and maturity, even with overlapping fruits.

Conclusions:

• TopoGeoFusion effectively automates gooseberry grading using topological features.
• The developed models are highly accurate and reliable, even in complex scenarios with occluded fruits.
• The technique enables detection and feature computation from partial objects, enhancing grading capabilities.