Estimation of Maize Silage Yield and Quality Using UAV-Based Multi-Source Data

Maize (Zea mays L.) silage is one of the most essential roughage resources for dairy cows. Its nutritive quality significantly impacts animal performance and, consequently, affects the final profit of the livestock industry. Mazie silage nutritive quality is primarily determined by the forage yield and compositional attributes, such as crude protein (CP), starch, neutral detergent fiber (NDF), and acid detergent fiber (ADF) concentrations. Typically, breeders combine these nutritive traits through summative criteria, such as MILK2006 (milk/acre index), to select high-performing hybrids in breeding cycles to improve maize germplasm and promote future feed quality. However, the conventional methods of measuring these traits mainly rely on laboratory chemical and near-infrared reflectance spectroscopy approaches of ground samples, which are time-consuming and labor-intensive. Recently, uncrewed aerial vehicles (UAVs) have become a new research frontier in crop high-throughput phenotyping (HTP) due to their flexibility and efficiency in data collection. In this study, we aim to investigate the potential of estimating the forage yield and nutritive quality of 507 maize silage hybrids using spectral, textural, and structural features derived from the UAV deployed hyperspectral, RGB, and light detection and ranging (LiDAR) sensors. Specifically, we extracted spectral reflectance, gray-level co-occurrence matrix (GLCM), and plant height features from the collected data and built machine learning models based on the obtained features. Results showed that the ridge regression models maintained acceptable estimation accuracies for all the yield and quality traits. The spectral reflectance supplemented by the GLCM and plant height features achieved practical yield estimation with a correlation coefficient (r) of 0.79. Moreover, there were good agreements between the estimated quality traits and ground data with r of 0.73 for milk/acre index, 0.72 for CP, 0.49 for starch, 0.41 for NDF, and 0.49 for ADF. Our results showed that the proposed approach in this study could ease the compositional attributes acquisition process and increase efficiencies for maize silage breeding.