Hyperspectral-Based Classification of Individual Wheat Plants into Fine-Scale Reproductive Stages for Anthesis Prediction

Field trials are critical in the development of genetically modified and genome-edited biotechnology plants to evaluate the growth and yield of breeding lines and to test commercial viability or any potential off-target effects. In Australia, conducting field trials of biotechnology derived crops requires compliance with federally mandated regulations, including strict protocols for forecasting flowering times. Conventional practices are based on time consuming, subjective and costly visual field inspections of individual wheat plants at respective growth stages (Zadoks growth stages Z37, Z39, and Z41). To enable automatic forecasting, hyperspectral and RGB images were captured in the greenhouse, and hyperspectral reflectance data were acquired in a semi-natural environment. In the greenhouse, imaging was conducted under controlled lighting with a fixed top-view setup; in semi-natural environments, spectral data were collected manually from multiple oblique angles under supplemented natural light. Support Vector Machine classification achieved F1 scores above 0.8 for anthesis prediction when reflectance data were transformed using Standard Normal Variate, Hyper-hue, or Principal Component Analysis. After feature selection, F1 scores above 0.75 could be achieved with only five wavelengths. Furthermore, the SNV transformation demonstrated robust performance under limited training conditions, maintaining high classification accuracy and strong generalizability across varying data sizes. These findings highlight the effectiveness of transformation-enriched data and optimized feature selection for accurate growth stage classification. This study provides a low-cost approach to alleviate manual inspection burdens, improve regulatory compliance, and increase biosafety during biotechnology field trial practices.