Pocket Phenomics: A Smartphone-based Framework for Quantifying Maize Architectural Traits in Field

Accurate quantification of maize architectural traits, such as ear height and leaf angle, is essential for breeding density-adapted ideotypes, yet high-throughput and traceable field phenotyping remains difficult. Existing 3D phenotyping platforms are constrained by canopy occlusion, operational complexity, or high costs, while manual measurements are labor-intensive and poorly traceable. To bridge this gap, this study developed Pocket Phenomics, a smartphone-based framework that integrates smartphone RGB-D acquisition, dToF error characterization, automated organ keypoint detection, and 3D geometric trait calculation into a unified field phenotyping workflow for in-field quantification of maize architectural traits. Its acquisition module (field phenotyping collection system, PP-FPCS) uses the iPhone direct time-of-flight LiDAR sensor and ARKit to acquire aligned RGB-D data for measuring ear height and leaf angle. To improve depth reliability under field conditions, the effects of acquisition distance and camera pose were quantified, and a robust operational protocol was established (distance ≤ 1.5 m, pitch/yaw within ±30°), which constrained the relative error of reconstructed 3D distances on a planar reference target to below 10%. The trait extraction pipeline further combines a YOLOv8 model enhanced with a Gather-and-Distribute module for robust multi-scale detection under occlusion and clutter with a depth-based background removal strategy to suppress complex field backgrounds. Field validation showed good agreement between smartphone-derived and manual measurements, with mean absolute errors of 5.07 cm for ear height (r = 0.96; mean bias - 4.58 cm) and 3.54° for leaf angle (r = 0.79). A temporally independent field validation carried out in 2025 reproduced this performance (ear height: r = 0.98, MAE = 4.23 cm; above-ear leaf angle: r = 0.89, MAE = 2.47°). Overall, this study provides a portable, low-cost, and reproducible framework for maize architectural phenotyping in the field and, more importantly, demonstrates consumer-grade smartphone LiDAR as a practical sensing tool for high-throughput phenotyping, with potential applications in breeding and precision crop management.