Physical approaches for components identification: from spectroscopy to modelling

One of the key problems of unknown mixtures, regardless of whether they are gas, liquid, or solid agglomerates, is identifying and quantifying their relative components, even if dispersed in minute quantities. When the components are not mixable or grain-based, this becomes crucial: this problem is critical in terms of the industrial food sector, particularly in cereal production. Food security requires special care not only to supply safe and healthy food, but also to give livelihoods and income to many farmers as an essential component of rural and economic development. At the heart of the food security goal, there is the production of cereals to meet the increasing demand for food, animal feed, and biofuels. Regardless of the final use, whether food or non-food applications, the need for a reliable, precise, and accurate quality assessment of cereal composition is of fundamental importance, especially when mixtures are in use. The composition inspections of grain tons have become therefore a high priority in the EU safety guidelines and have been addressed as a priority. According to international standards, qualified analysts perform on-spot testing, physical testing, chemical testing, contaminants and microbiological analyses, and GMO testing. For grains, the place of production freedom is evaluated by visual inspection of the growing crop as well as testing of samples of grain at harvest and preshipment and finding them free from pathogens. Instruments such as digital imaging devices with pattern recognition algorithms have not still been able to replace manual visual analysis. The reason for this lies in the diversity of conditions to be examined, which include broken and shriveled kernels, immature kernels, heat-damaged and frost-damaged kernels, staining, insect-damaged kernels, sprouted kernels. Optical sensing and spectroscopic techniques have high potential for automated real-time quality and safety inspection of agricultural and food products, aside from the fact that their vast size and high cost make them challenging to operate in the industrial sector. Some solutions have been recently proposed based on imaging and machine visions systems that have achieved a good level of assessment of the quality of a given species but fail in identifying mixtures. For this precise reason, the focus of this thesis has been addressed on the exploration and development of a new technique to detect components in unknown grain-based mixtures in a non-destructive, rapid, and precise manner by using physical approaches: multispectral imaging (MSI) technology. MSI is a novel technique that combines the advantages of machine vision systems and near-infrared spectroscopy to measure a material’s spectral footprint by taking pictures of the sample at different excitation wavelengths. Throughout this dissertation, we will develop a model that, when paired with MSI, is able to discover the components of an unknown grain-based mixture by removing background and shadow effects using principal components analysis (PCA) and the N-FINDR algorithm for spectral unmixing process.