Manifold-based learning for high-throughput single-peanut phenotyping.

Abstract

Peanut (Arachis hypogaea L.), a major legume crop valued for its high oil content, displays complex genotypic-phenotypic interactions shaped by environmental influences, yet these relationships remain poorly understood. We present a high-throughput phenotyping framework that captures the geometry of peanut pods using digital microscopy or smartphone imaging integrated with manifold-learning for large-scale analysis and visualization. Using over 6500 pods collected across China, we identify a geographically distinct morphological signature and demonstrate accurate cultivar discrimination. This scalable approach establishes the foundation for a Large Geometric Model capable of predicting phenotypic traits and accelerating precision agriculture. Our pipeline offers a transformative tool for peanut breeding and sustainable crop improvement.

Code and data availability

The authors state that the peanut pod image dataset, extracted phenotypic trait data, and the Orange Data Mining workflow (.ows) used for analysis are publicly available in their GitHub repository.