Innovative Cucumber Phenotyping: A Smartphone-Based and Data-Labeling-Free Model

Abstract

Sustaining global food security amid a growing world population demands advanced breeding methods. Phenotyping, which observes and measures physical traits, is a vital component of agricultural research. However, its labor-intensive nature has long hindered progress. In response, we present an efficient phenotyping platform tailored specifically for cucumbers, harnessing smartphone cameras for both cost-effectiveness and accessibility. We employ state-of-the-art computer vision models for zero-shot cucumber phenotyping and introduce a B-spline curve as a medial axis to enhance measurement accuracy. Our proposed method excels in predicting sample lengths, achieving an impressive mean absolute percentage error (MAPE) of 2.20%, without the need for extensive data labeling or model training.