New cross-validation method improves AI fruit shelf-life prediction accuracy

Researchers have developed a new method for evaluating the accuracy of AI models that predict the remaining shelf life of fruit from images. Using the Hass Avocado Ripening dataset, they found that specimen-level cross-validation, which ensures images from the same fruit are not used in both training and testing, provides a more realistic performance estimate than observation-level splits. This approach revealed a higher error rate, with specimen-disjoint cross-validation yielding a mean RMSE of 3.12 days compared to 2.37 days for observation-level splits. The study also compared lightweight models like MobileNetV3-Small against heavier ones like ResNet-18, finding comparable accuracy with significantly better throughput for the smaller model, and used Grad-CAM to analyze visual attribution differences. AI

IMPACT This research could lead to more reliable AI tools for supply chain management and food waste reduction by improving model evaluation standards.

RANK_REASON Academic paper detailing a new methodology and experimental results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• Grad-CAM++
• Hass Avocado Ripening dataset
• ImageNet
• MobileNetV3-Small
• ResNet-18

• paper
• product
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →