CEHD: A Unified Framework for Detection and Height Estimation of Fresh Corn Ears in Field Conditions.

Real-time detection of fresh corn ear height can provide a basis for dynamic adjustment of harvester header parameters, reducing mechanical damage and improving harvest quality. This study proposes a corn ear height detection model (CEHD). A YOLO-HAMDF network is developed for ear recognition, in which the core modules-TBDA, GLSA, and AQE-respectively suppress background interference, enhance contextual perception, and optimize bounding-box scoring. Depth information is incorporated to filter non-target regions and improve system robustness. In addition, a DI-DeepSORT module is designed for ear tracking, where DBC-Net and IDA-Kalman, respectively, enhance the discriminability of ReID features and enable independent-dimension adaptive noise modeling with smoothed positional updates. Experimental results demonstrate that the proposed CEHD model achieves a mean absolute error (MAE) of only 3.21 ± 0.05 cm under field conditions, indicating strong stability and practical applicability. In summary, this study presents a stable and reliable corn ear height detection system, achieves real-time monitoring of ear height, and provides data support for the dynamic adjustment of header parameters in fresh corn harvesters.