Semester of Graduation
Master of Science (MS)
School of Plant, Environmental and Soil Science
The first chapter of this thesis explores the predictive capabilities of random forests algorithm on datasets obtained from field plot experiments on crop management systems in soybean. Furthermore, the chapter presents a complementary analysis of model performance according to dataset sizes and two techniques on how to impute and deal with missing data. Random forests are being compared with standard statistical techniques such as linear regression on a well-structured, information-rich agronomic experiment. The key findings of this chapter includes the best hyperparameters adjustment and the identification of the dataset threshold for optimal algorithms performance. The second chapter has a research study for optimizing UAV flights parameters to measure field elevation. The research investigates the current challenges that farmers face when planning drainage and proposes a solution using a LiDAR-based sensor payload mounted on a UAV. The study compares the results obtained from flights at three different altitudes (40 m, 50 m, and 60 m) and determines the impact of the flight altitude on survey area coverage and flight times. The findings of the research show that there is no significant difference between the map results created using the three flight altitudes. However, the flights conducted at 60 m altitude reduce survey area coverage by 17.5% to 27.5% and flight times by 59.1s to 75.7s per 1000 m-2 when compared to data captured at 40 m. Overall, this research provides a valuable contribution to water management planning in agriculture by proposing a practical and efficient solution for measuring field elevation using LiDAR-UAV technology.
Santos, Leticia, "MACHINE LEARNING-BASED SOYBEAN YIELD PREDICTION AND OPTIMIZING LiDAR-MOUNTED UAV EFFICIENCY" (2024). LSU Master's Theses. 5828.