A research team led by Professor Jiang Ni from the Institute of Genetics and Developmental Biology (IGDB) at the Chinese Academy of Sciences (CAS) has developed a cost-effective, deep-learning-based method for in-field measurement of flag leaf angle (FLANG) in wheat. The approach combines a low-cost imaging setup with a lightweight deep learning model called LeafPoseNet, enabling accurate and high-throughput phenotyping directly in agricultural fields. FLANG is a critical trait in wheat breeding, influencing plant architecture, light interception efficiency, and ultimately, grain yield. However, traditional methods for measuring FLANG rely heavily on manual, subjective assessments, making them time-consuming, labor-intensive, and unsuitable for large-scale breeding programs. To overcome these limitations, the researchers introduced LeafPoseNet, a keypoint-based pose estimation model designed to automatically detect three essential anatomical landmarks: the center of the flag leaf (Point L), the junction where the flag leaf meets the stem (Point J), and the center of the stem (Point S). By analyzing the spatial relationship among these points, the model computes FLANG with high precision. LeafPoseNet outperforms existing state-of-the-art keypoint detection models, achieving a mean absolute error (MAE) of 1.75°, a root mean square error (RMSE) of 2.17°, and a coefficient of determination (R²) of 0.998. The model demonstrates strong robustness in accurately localizing keypoints across a wide range of leaf shapes and under complex field conditions, including varying lighting and occlusions. Thanks to its lightweight architecture and high computational efficiency, LeafPoseNet can run smoothly on standard smartphones, making it highly accessible and practical for on-site use by breeders and agronomists. This enables rapid, large-scale FLANG measurement without requiring expensive equipment or specialized infrastructure. The team validated the method by measuring FLANG across a diverse panel of 221 bread wheat accessions. Using a mixed linear model (MLM) for genome-wide association study (GWAS), they identified 10 quantitative trait loci (QTLs) significantly associated with FLANG, offering valuable genetic insights into the trait’s regulation in wheat. Published in The Crop Journal, this study presents a powerful, scalable solution for high-throughput, in-field phenotyping of FLANG. The approach not only enhances the accuracy and efficiency of wheat breeding programs but also opens new avenues for integrating advanced computer vision into agricultural research and crop improvement.