A Millimeter-Wave Radar System for Plot-Scale Canopy Phenotyping in Breeding Trials

Abstract

Plant breeding programs are increasingly limited not by the rate at which new genotypes can be generated but by the rate at which their physical traits can be evaluated in the field, a disparity commonly described as the phenotyping bottleneck. Structural traits such as canopy height and density are strongly tied to yield yet remain slow and labor-intensive to acquire, and the established sensing modalities used to automate their measurement have limitations. Optical imaging and LiDAR are sensitive to ambient conditions and interact primarily with the outermost canopy surface. Radar offers a complementary alternative by measuring range directly and independently of illumination. It can also partially penetrate a canopy. Compact, plot-scale, millimeter-wave radar instruments designed for phenotyping, together with a processing pipeline to turn their raw output into per-plot traits need to be developed. This dissertation presents the design, implementation, and field evaluation of a compact 120 GHz stepped-frequency continuous-wave (SFCW) radar for high-throughput soybean phenotyping. The instrument developed is a low-power, portable sensor built around a highly integrated millimeter-wave transceiver and a custom embedded acquisition system, with a processing pipeline that converts its raw range-time acquisitions into per-plot canopy height and structural density measures. It employs a continuous coherent calibration technique that cancels the transmit-receive leakage that often limits compact continuous-wave radars, preserving the dynamic range needed to detect weak canopy and ground returns. The processing pipeline removes the system’s characteristic artifacts and applies a particle filter with a backward smoother to track the ground through canopy shadowing, a region-based detector to extract the canopy relative to that ground, and a radar integrated energy measure as a proxy for canopy structural density, all within a calibrate-once human-in-the-loop workflow. The system was deployed in a soybean breeding trial at Kansas State University across the 2025 growing season. Over five measurement sessions spanning emergence to maturity, the radar tracked canopy development and produced height estimates in agreement with manual ground truth measurements, including close agreement under a fully closed canopy. An experiment established that the height estimate is both repeatable and consistent with hand measurement, and the integrated-energy density measure agreed with an independent optical canopy-cover estimate at a Pearson correlation of r = 0.837. The semi-automated workflow transferred a single tuned parameter set across measurement batches within a growth stage and acquired continuous along-row height profiles at a sampling density that manual measurement cannot practically match, roughly eight times faster than hand measurement at equivalent density, while concentrating the demand on skilled labor into a single tuning step. The results demonstrate that proximal millimeter-wave radar is a viable and complementary instrument for high-throughput structural phenotyping