CALIBRATION STRATEGIES FOR THE ANALYSIS OF FORAGES USING NEAR INFRARED SPECTROSCOPY (NIRS) INSTRUMENT

The quality control process for industrial inspection requires methods to quantify the concentration of chemical constituents of feed before the feed is transferred to animals and the last process line to consumers, these processes are usually achieved chemically and are destructive and time-consuming. Today, feed industries are transforming these processes by adopting methods that are more efficient, less wasteful and non-destructive to the materials being inspected. These methods rely on spectroscopic techniques, in which devices are used to acquire signals from products, and these signals can be used to predict the chemical composition of product. The combination of spectroscopy and multivariate calibration is one of the biggest breakthroughs in this field. In order to transform the composition inspection of products from this costly and destructive process to rapid non-destructive instant quantification. Chemical quantification of products based on spectral signals is a process based on the use of multivariate calibration models that are trained for further use in quantitation through predictions. Nonetheless, while chemical relationship between prediction environments and maintained over the long term has been a significant challenge since the chemical transition. Near Infrared Spectroscopy is definitely a powerful tool that should not be ignored in the feed industry, NIR technology can also assist in research projects where more analyses can be done with fewer resources, even a screening of many samples in short time can be done to select noticeable samples for future quantitative analyzes. The aim of this work is to develop efficient methods to develop, deliver between spectral sensors by calibrating the model, thereby using the long-term effectiveness of the model with a minimal effort, this thesis did three researches:The first research tested the This study evaluated the accuracy of calibrations developed on wet and dried and ground samples and also if the basis (wet vs. dry) of reference data would affect prediction performances. The second research of the present thesis aimed to evaluate the accuracy for large multi-products libraries, local calibration in four different ways, Global calibration, Shenk’s LOCAL and Hybrid Local, a new locally weighted method based on partial least squared scores kNN-LWPLSR, comparing them to the traditional Global calibrations.The third research of the present study evaluated techniques to transfer calibrations for alfalfa and grass forage samples that were developed for a scanning benchtop monochromator (Foss 6500, 400-2498 nm, Foss, DK; LAB) to a diode array instrument (AuroraNir, 950-1650 nm, GraiNit, IT; DA), a digital light processing instrument (NIR-S-G1, 950-1650 nm, Innospectra, TW; DLP) and a short wavelength instrument (SCiO, 740–1070 nm, Consumer Physics, Israel; SCIO).