Early corn stand count of different cropping systems using UAV-imagery and deep learning

Optimum plant stand density and uniformity is vital in order to maximize corn (Zea mays L.) yield potential. Assessment of stand density can occur shortly after seedlings begin to emerge, allowing for timely replant decisions. The conventional methods for evaluating an early plant stand rely on manual measurement and visual observation, which are time consuming, subjective because of the small sampling areas used, and unable to capture field-scale spatial variability. This study aimed to evaluate the feasibility of an unmanned aerial vehicle (UAV)-based imaging system for estimating early corn stand count in three cropping systems (CS) with different tillage and crop rotation practices. A UAV equipped with an on-board RGB camera was used to collect imagery of corn seedlings (~14 days after planting) of CS, i.e., minimum-till corn-soybean rotation (MTCS), no-till corn-soybean rotation (NTCS), and no-till corn-corn rotation with cover crop implementation (NTCC). An image processing workflow based on a deep learning (DL) model, U-Net, was developed for plant segmentation and stand count estimation. Results showed that the DL model performed best in segmenting seedlings in MTCS, followed by NTCS and NTCC. Similarly, accuracy for stand count estimation was highest in MTCS (R2 = 0.95), followed by NTCS (0.94) and NTCC (0.92). Differences by CS were related to amount and distribution of soil surface residue cover, with increasing residue generally reducing the performance of the proposed method in stand count estimation. Thus, the feasibility of using UAV imagery and DL modeling for estimating early corn stand count is qualified influenced by soil and crop management practices