Hybrid AI Pipeline for Laboratory Detection of Internal Potato Defects Using 2D RGB Imaging.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Internal potato tuber quality assessment is vital for agro-laboratory processing.
• Traditional methods fail to detect internal defects (hollow heart, bruises, insect galleries) using only surface features.
• Existing advanced techniques like hyperspectral imaging or MRI are costly and less accessible.

Purpose of the Study:

• To develop a novel, modular hybrid AI architecture for detecting internal defects in potato tubers using RGB images.
• To create a system suitable for integration into laboratory settings for efficient quality control.
• To provide a robust, interpretable, and cost-effective alternative to current defect detection methods.

Main Methods:

• A hybrid AI pipeline combining YOLO for detection, ResNet for patch validation, Segment Anything Model (SAM) for segmentation, and VGG16 with Random Forest for skin-contact analysis.
• Utilized RGB images of potato slices for defect analysis.
• Trained and validated the model on a dataset of over 6000 annotated instances.

Main Results:

• Achieved a recall rate above 95% and a precision of approximately 97.2% for most internal defect classes.
• Demonstrated superior performance compared to methods relying on hyperspectral or MRI techniques.
• The system proved to be scalable, explainable, and compatible with standard 2D imaging hardware.

Conclusions:

• The proposed hybrid AI architecture offers a robust and interpretable solution for internal potato defect detection.
• This system enhances the capabilities of agro-laboratories by providing accurate and efficient quality assessment.
• The approach represents a significant advancement in non-destructive internal quality evaluation for agricultural products.