Multi-Genotype Rice Yield Prediction Based on Time-Series Remote Sensing Images and Dynamic Process Clustering

Predicting rice yield in a timely, precise, and efficient manner is crucial for directing agricultural output and creating food policy. The goal of this work was to create a stable, high-precision estimate model for the yield prediction of multi-genotype rice combined with dynamic growth processes. By obtaining RGB and multispectral data of the rice canopy during the whole development stage, several bands of reflectance, vegetation index, canopy height, and canopy volume were retrieved. These remote sensing properties were used to define several curves of the rice-growing process. The k-shape technique was utilized to cluster the various characteristics based on rice growth features, and data from different groups were subsequently employed to create a yield estimation model. The results demonstrated that, in comparison to utilizing solely spectral and geometric factors, the accuracy of the multi-genotype rice estimate model based on dynamic process clustering was much higher. With a root mean square error of 315.39 kg/ha and a coefficient of determination of 0.82, the rice yield calculation based on canopy volume temporal characteristics was the most accurate. The proposed approach can support precision agriculture and improve the extraction of characteristics related to the rice growth process.