UC Riverside

The rapid advancement in sensor technology has significantly transformed pre-cision agriculture, enabling more efficient monitoring and management of agricultural practices. However, the high-frequency data generated by these sensors presents chal- lenges in calibration, especially when extensive ground-truth data is required. This thesis introduces an advanced approach called sensor-directed sampling (SDS), which is designed to optimize the sampling process while minimizing the number of nec- essary soil samples without compromising the accuracy of predictive models. The hyperspectral sensor setup requires synchronization and fusion of multiple individual sensors. We explore the short-vectorization approach for data fusion and create a library for ease of use. The research explores integrating sampling methodologies with the Central Com- posite Design (CCD) of Response Surface Methodology (RSM) within the SDS frame- work. These methods allow for a more robust design of experiments, facilitating the accurate capture of soil properties across large geographic areas. Additionally, the thesis presents the development and implementation of a Python-based software tool, SDSampling, which enhances the practical application of the proposed methodologies in ESAP RSSD Software. The effectiveness of the SDS approach is validated through a series of synthetic and real-world experiments, demonstrating its capability to reduce sampling efforts while maintaining model accuracy. This work contributes to the field of precision agri- culture by offering a scalable solution for sensor-based data collection and analysis, ultimately supporting more informed decision-making processes in agricultural man- agement. The extension of this work also involves stochastic optimization and GPU acceleration. Code can be found here: https://github.com/jayanta-banik/Sensor- Directed-Sampling