X-ray driven peanut trait estimation: computer vision aided agri-system transformation

Area of Science:

• Agricultural Science
• Food Technology
• Imaging Technology

Background:

• Manual peanut quality assessment in India is slow and prone to errors.
• Procurement delays and poor storage lead to fungal contamination and food safety issues.
• Current methods lack objectivity and efficiency in evaluating raw peanut pods.

Purpose of the Study:

• To investigate the efficacy of X-ray technology for rapid peanut quality assessment.
• To determine if X-ray imaging can accurately estimate kernel and shell weight for price determination.
• To explore a technological solution for improving peanut procurement processes.

Main Methods:

• Generated 1752 2D X-ray projections of peanut pods using computed tomography (CT).
• Employed X-ray image transformation (XRT) and a convolutional neural network (CNN) for feature prediction.
• Validated predictions against gravimetric measurements of kernel and shell weight.

Main Results:

• Both XRT and CNN accurately predicted kernel weight (R² > 0.93).
• CNN showed higher accuracy in predicting shell weight (R² = 0.91) compared to XRT (R² = 0.78).
• X-ray based methods demonstrated high correlation with actual weight measurements across diverse peanut varieties.

Conclusions:

• X-ray technology offers a viable solution for objective and rapid peanut quality assessment.
• This technology can streamline procurement, reduce post-harvest losses, and ensure fair farmer compensation.
• The system has potential applications in crop improvement programs for high-throughput cultivar selection.