Modeling and Compensation of Backreflection Signal in Diffuse Reflection Optical MEMS Spectrometers

Abstract

Near-infrared (NIR) spectrometry, particularly in the diffuse reflection mode, offers significant advantages for measuring material spectral response without specialized sample preparation or destruction. Moreover, it provides both qualitative and quantitative analysis of materials in various applications, including industrial, medical, agriculture, and food. However, the presence of specular reflection from spectrometer and sample holder introduces unwanted signals, leading to non-linear photometric spectral errors. This issue is especially pronounced when measuring highly absorbing materials, limiting the accuracy of determining their maximum absorbance levels. Moreover, it is more significant in miniaturized MEMS spectrometers since the portability necessitates a compact optical head, where the stray light cannot be easily avoided. This work investigates and models the backreflection, considering the reflective index of windows and the refractive index of the material under test. Furthermore, the research proposes an effective method for compensating the backreflection error. To validate our compensation method, experiments were conducted using milk as a test material. Milk is a suitable candidate for this experiment due to its high water content, which gives it strong light absorption properties in the realm of diffuse reflection. Backreflection compensation leads to a substantial reduction in absorbance saturation, increasing maximum measured absorbance of milk from 1.77 AU to 2.5 AU and decreasing variations of different sampling configurations more than 10 times. Furthermore, the backreflection compensation was evaluated using a chemometric model for high-moisture feed material, demonstrating a reduction in prediction error by 29% and an improvement in bias variations by a factor of 11.