Abstract
A physics-based approach for pushing the limits of our custom-designed near-infrared (NIR) interactance spectroscopic instruments is presented, demonstrating how more robust measurements in demanding real-world applications can be achieved. Achieving sufficient instrument performance within given design constraints requires a balancing act between the sensor size and cost, its tolerance for extrinsic environmental influences, and the sample complexity. Using an example case of in-field, non-contact robotic measurements of sweetness in strawberries, we demonstrate a step-by-step approach for building a holistic understanding of the entire measurement solution, from 1) instrumentation to 2) sample properties and 3) data analysis. This includes incorporating knowledge about the impact of environmental influences and non-optimal instrument design, and how the physical and chemical characteristics of complex, highly scattering food products affect our measurements. When incorporated into the instrument design process, this can improve the measurements and the robustness of the multivariate prediction models.