Solutions based on near-infrared (NIR) spectroscopy, computer vision, and NIR hyperspectral imaging (NIR-HSI) for the understanding of specialty coffee quality

• Grant number: 24/13645-4
• Support Opportunities: Scholarships abroad - Research Internship - Post-doctor
• Start date: March 05, 2025
• End date: February 28, 2026
• Field of knowledge: Agronomical Sciences - Food Science and Technology - Food Science
• Principal Investigator: Cleiton Antônio Nunes
• Grantee: Yhan da Silva Mutz
• Supervisor: Alessandro Ulrici
• Host Institution: Escola de Ciências Agrárias. Universidade Federal de Lavras (UFLA). Ministério da Educação (Brasil). Lavras , SP, Brazil
• Institution abroad: Università degli Studi di Modena e Reggio Emilia, Modena (UNIMORE), Italy
• Associated to the scholarship: 23/00474-4 - Sensors associated with the internet of things to connect the environment, genetics and processing to the chemical and sensory profile of specialty coffees, BP.PD

Abstract

The specialty coffee market is experiencing significant growth driven by evolving consumer preferences. Specialty coffee, defined as green coffee beans of the highest quality with a specific geographic origin, has its quality depending on various factors such as genetics, cultivation conditions, post-harvest processing, roasting, and beverage preparation. Quality control in specialty coffee production involves evaluating physical and sensory attributes. Physically, the beans must be defect-free, appropriately sized, well-dried, and free from unusual odors or colors. Sensorial analysis includes quantitative assessments like cupping scores and taste attributes (sweetness, acidity, bitterness, aftertaste, body) and qualitative evaluations like flavor descriptors (chocolate, vanilla, fruity, etc.). These characteristics are determined through cupping assessments post-roasting and brewing. Sensory evaluation, or cupping, is essential for confirming specialty status (reached when a coffee has an 80-point or higher score) and product valuation in the coffee trade. However, cupping can be subjective, costly, time-consuming, and challenging when evaluating numerous samples. Additionally, the link between farming and post-harvest factors and coffee sensory quality is not well quantified. Thus, introducing a reliable instrumental technique to assess quality directly from dried coffee could enhance the understanding of these factors and reduce reliance on sensory evaluation. In this sense, non-destructive analytical techniques like near-infrared spectroscopy (NIR), computer vision, and NIR-hyperspectral imaging (NIR-HSI) are increasingly used in food processing. With advances in miniaturization, NIR technology shows promise for quality-related coffee purposes but lacks spatial distribution of spectral information. Conversely, computer vision provides spatial analysis, which is excellent for process monitoring approaches, but lacks spectral information, limiting its application. NIR-HSI combines NIR spectral data and spatial distribution, overcoming these limitations and showing great potential in food process control as a non-destructive, reliable technique. However, NIR-HSI data is a three-dimensional matrix, with each pixel representing a spectrum, necessitating multivariate data analysis techniques for processing, which, with further wavelength selection, can reduce spectral dimensionality without information loss, reducing the computer power needed and improving classification results by eliminating noise and redundant signals. Despite its potential, no studies on specialty coffee and HSI have been found, with only one study linking coffee beans to sensory aspects through HSI data. On the other hand, similar techniques, like infrared spectroscopy, are widely used for coffee quality-related purposes, with some studies attempting to correlate coffee's physical and sensory quality over time. The unexplored potential of HSI places it as a potential candidate for establishing new links between specialty coffee and its sensory quality, which should be explored through data dimensionality reduction techniques, addressing the complexity of data to assess and predict coffee quality. As understanding the impact of farming and post-harvest practices on final cup quality is critical, the proposed work involves obtaining spectral and physical measurements using NIR, NIR-HSI, and computer vision of harvested coffee fruits, varying processing, and final quality and evaluating their relationship with sensory responses. The data set will be analyzed with multivariate statistical methods and image processing algorithms to develop predictive models, bridging the knowledge gap on the influence of farming and processing on coffee quality. This approach aims to simplify the analytical process, benefiting producers by enabling consistent high-quality specialty coffee production, ultimately benefiting consumers with reliable product quality.