Calibrating UAV thermal sensors using machine learning methods for improved accuracy in agricultural applications

Accurate temperature measurements are essential for detecting crop stress, managing irrigation, and monitoring vegetation health. However, various factors can affect thermal sensors that can introduce measurement errors. To address this, machine learning (ML) algorithms were used to calibrate unmanned air vehicle (UAV) thermal sensor measurements. In this study, commercially available two different types of UAV thermal sensors, including Micasense Altum and Flir Duo Pro-R (FDP-R), have been tested and evaluated its performance by comparing the calibrated ground thermal measurements. For this purpose, five different ML algorithms, namely Random Forest, Support Vector Machine, K-NN and XGBoost, were used to calibrate UAV thermal sensors. Results showed that, after thermal calibration with XGBoost, the RMSE decreased by 2.84 degrees C (from 4.23 degrees C to 1.39 degrees C) for Micasense Altum and by 2.51 degrees C (from 3.84 degrees C to 1.33 degrees C) for FDP-R, while R2 increased from 0.89 to 0.96 for Micasense Altum and from 0.87 to 0.94 for FDP-R. In addition, we conducted correlation analyses between the calibrated temperature measurements and various sorghum phenotype parameters, such as leaf area index, crop height, and soil moisture. The results indicate that both sensors have performed well in terms of correlation coefficients. Micasense Altum has shown slightly better performance for crop height and soil moisture (r = -0.78 and r = -0.59, respectively), while FDP-R has performed better for leaf area index (r = -0.70). This study demonstrates the potential of using calibrated UAV thermal sensors for precision agriculture tasks and highlights the importance of validating the calibration with ground measurements.