Digital Phenotyping and AI-based Diagnostics for Fruit Quality

Moonshot Award

Hybrid Award (Proposal Development & Innovation Development)

Abstract

Fruit quality defects reduce marketable yield and consumer acceptance in apple. Many of the most important defects occur on the fruit surface or in internal cortex tissues that are difficult to quantify with conventional visual grading. We initiated this project to develop new non-destructive imaging and AI-based analysis for detecting internal defects (watercore, internal browning) via CT imaging and external defects (russeting, scarf skin) via surface phenotyping. To date, we have focused on russeting, a surface disorder characterized by cracks in the fruit cuticle. To track the development of early russet risk, we are evaluating the GelSight portable elastomeric microscope as a field- and lab-deployable tool for high-resolution surface phenotyping. The method works by pressing a soft gel-based sensor against the fruit surface to capture 3D microtopography, enabling quantitative comparison of russeted and intact cuticle regions. 

Preliminary time-series data have been collected on seven apple cultivars and breeding selections that differ in their russet susceptibility, including Honeycrisp, Golden Delicious, and Snapdragon. Ongoing work will apply AI image analysis to parse the collected topographic data and develop the time-series aspects of russet development. Planned analyses include image-analysis pipelines to classify and quantify surface phenotypes, with longer-term extension to internal defect detection. This approach aims to provide objective, repeatable surface phenotyping to support russeting research and breeding.