UAS-Based Multi-Temporal Rice Plant Height Change Prediction

Abstract

Analyzing rice growth is essential for examining pests, illnesses, lodging, and yield. To create a Digital Surface Model (DSM ) of three important rice breeding stages, an efficient and fast (compared to manual monitoring) Unoccupied Aerial System was used to collect data. Outliers emerge in DSM as a result of the influence of environ- ment and equipment, and the outliers related to rice not only affect the extraction of rice growth changes but are also more challenging to remove. Therefore, after using ground control points uniform geodetic level for filtering, statistical outlier removal (SOR ) and quadratic surface filtering (QSF ) are used. After that, differential operations are applied to the DSM to create a differential digital surface model that can account for the change in rice plant height. Comparing the prediction accuracy before and after filtering: R2 = 0.72, RMSE = 5.13cm, nRMSE = 10.65% for the initial point cloud; after QSF, R2 = 0.89, RMSE = 2.51cm, nRMSE = 5.21%; after SOR, R2 = 0.92, RMSE = 3.32cm, nRMSE = 6.89%. The findings demonstrate that point cloud filtering, particularly SOR, can increase the accuracy of rice monitoring. The method is effective for monitoring, and after filtering, the accuracy is sufficiently increased to satisfy the needs of growth analysis. This has some potential for application and extension.