Hyperspectral Imaging (HSI) combines microscopy and spectroscopy to assess the spatial distribution of spectroscopically active compounds in objects, and has diverse applications in food quality control, pharmaceutical processes, and waste sorting. However, due to the large size of HSI datasets, it can be challenging to analyze and store them within a reasonable digital infrastructure, especially in waste sorting where speed and data storage resources are limited. Additionally, as with most spectroscopic data, there is significant redundancy, making pixel and variable selection crucial for retaining chemical information. Recent high-tech developments in chemometrics enable automated and evidence-based data reduction, which can substantially enhance the speed and performance of Non-Negative Matrix Factorization (NMF), a widely used algorithm for chemical resolution of HSI data. By recovering the pure contribution maps and spectral profiles of distributed compounds, NMF can provide evidence-based sorting decisions for efficient waste management. To improve the quality and efficiency of data analysis on hyperspectral imaging (HSI) data, we apply a convex-hull method to select essential pixels and wavelengths and remove uninformative and redundant information. This process minimizes computational strain and effectively eliminates highly mixed pixels. By reducing data redundancy, data investigation and analysis become more straightforward, as demonstrated in both simulated and real HSI data for plastic sorting.

Hyperspectral imaging (HSI) is a powerful, non-invasive analytical technique extensively utilized in chemistry as it simultaneously captures morphological and chemical information from samples across a broad spectrum of chemically informative wavelengths. In this context, morphological information refers to the spatial structure, shape, texture, and distribution of elements within the image. Enhancing its already widespread application requires reducing the computational load of the voluminous hyperspectral images while unmixing signals from different chemical species with unknown spectral fingerprints. Endmember extraction, which involves finding the purest spectral signatures within the data, is needed for decomposing these mixed signals. By resolving mixed pixels into their constituent endmembers, HSI enables accurate quantification and spatial mapping of chemical components, even when prior knowledge is limited. Current methods for endmember extraction, such as NFINDR, VCA, PPI, SIMPLISMA, and AMEE, are limited by issues including computational slowness, the requirement for extensive parameter optimization, and a lack of hierarchical consistency. Consequently, there is a pressing need for a method that is both faster and more accurate. Successive Projection Algorithm (SPA) is developed for forward wavelength selection to improve the predictive accuracy of regression models under strong collinearity. SPA emerges as a rapid and accurate endmember extraction technique, with applications extending beyond chemistry to areas such as food safety, environmental monitoring, and material analysis. Comparative analyses using both simulated and experimental datasets illustrate SPA's superior robustness, repeatability, absence of parameter tuning requirements, and computational efficiency when compared with the methods in current use. These findings show the value of SPA as a robust tool for computationally efficient hyperspectral image analysis in chemical applications and beyond.

The combination of near-infrared spectroscopic techniques (HSI or hyperspectral imaging) with advanced image analysis allows noninvasive, quantitative analysis of slow mixing phenomena. This is illustrated in a small study to monitor the preparation of homemade mayonnaise, specifically concerning the major challenge of “breaking” the emulsion. The context of the phenomenon in terms of the physics of mixing is explained as well. This study demonstrates that the Kullback–Leibler divergence metric, when applied to HSI data, enables quantitative tracking of emulsion mixing dynamics and can be used to quantify emulsion homogeneity and identify phase inversion during mayonnaise preparation if manual oil addition proceeds too quickly. This approach holds promise for real-time monitoring of emulsion quality in industrial settings where the state of the emulsion cannot always be determined visually.

Microplastic pollution presents major environmental and health challenges, requiring accurate identification and quantification to assess its distribution and impact. Conventional methods such as chromatography and spectrometry provide precise results but are destructive, time-consuming, and resource intensive. Hyperspectral Imaging in the Near-Infrared range (HSI-NIR) offers a non-destructive alternative by capturing both spectral and spatial information, though analysis of its large, noisy datasets remains difficult. This study introduces an analytical pipeline combining HSI-NIR with optimized preprocessing and a machine-learning-based Multi-Layer Perceptron (MLP) model for pixel-wise classification of microplastic particles. The shallow MLP architecture effectively handles high-dimensional data using predefined spectral features. The approach was applied to samples from Lanzarote Island, the Wadden Sea, and the Waal and Rhine rivers, accurately identifying polyethylene (PE), polypropylene (PP), polyethylene terephthalate (PET), and polystyrene (PS). The MLP model outperformed Support Vector Machines, Random Forests, and Partial Least Squares Discriminant Analysis in polymer identification. PE and PP were dominant across all sites, with PET and PS in lower proportions. River samples showed higher microplastic concentrations in the Rhine than in the Waal, with polymer composition stable across depths. Code available at: https://github.com/petroshatt/hyperplastics.