Real-time monitoring of maize phenology using ground camera fusion information

• Clarified the relationship between ground camera information and maize BBCH. • Fusion of spectral and textural information made BBCH estimation more accurate. • Created an accurate way to monitor real-time phenology within the SEL framework. • Real-time maize phenology can be extracted from a single-time-phase image. Real-time crop phenology is crucial for effective crop management and food security. Traditional real-time phenology monitoring methods often relied on labor-intensive field surveys, which were susceptible to subjective biases and had limited temporal resolution. Existing remote sensing methods excessively depended on time-series data and typically only estimated specific phenology events, which differed from the scales commonly used in crop phenology, e.g., Bundessortenamt and Chemical Industry Scale, BBCH, making practical application challenging. To address these issues, this study combined ground camera and machine learning (ML) algorithms to explore the relationship between spectral and texture information derived from NDVI camera and maize phenology (BBCH), developing a user-friendly real-time maize phenology monitoring stacked ensemble learning (SEL) model. The results showed a complex nonlinear response relationship between spectral and texture information and BBCH, making it difficult to fully characterize maize phenology throughout the season by relying solely on a single type of information. Combining spectral and textural information could significantly improve the accuracy of real-time maize phenology estimation. Compared to traditional ML models, the SEL performed best, with R² of 0.967, RMSE of 5.355, rRMSE of 9.22 %, and dRMSE of 6.205 days. The SEL method effectively improved the accuracy and efficiency of BBCH estimates while demonstrating good stability and generalization ability. Overall, these results confirmed the excellent performance of the SEL method in real-time phenology monitoring, providing new insights into real-time phenological monitoring using single-time-phase ground camera images and offering new ideas for remote quantitative diagnosis of crop growth.