Near-infrared (NIR) spectroscopy, digital imaging and spectral imaging have emerged as rapid, non-destructive, and environmentally friendly tools for assessing food safety and quality. Compared to traditional wet analytical techniques, these methods are more sustainable and cost-effective, as they don’t require skilled personnel nor chemical reagents. Opposed to destructive targeted approaches, NIR spectroscopy enables untargeted analysis, providing a comprehensive assessment of the distinctive chemical profile of a sample without prior selection of specific compounds of interest. Hyperspectral imaging combines the strengths of spectroscopy and imaging, allowing the visualization of spatial distribution of the chemical features of interest across a sample’s surface. However, its data-richness requires longer computational times thus complicating data handling, storage and analysis. These limitations are typically overcome by applying deep learning methods resulting in enormous environmental implications. In this frame, Chemometrics and Multivariate Image Analysis are a sustainable alternative for addressing the curse of dimensionality and extracting useful information. The present Doctoral Thesis proposes several strategies based on chemometrics and NIR hyperspectral imaging to increase the sustainability of the agri-food systems. To this aim, practical solutions for pest monitoring, post-harvest sorting, and for food authentication involving image dimensionality reduction and original classification methods are presented. Within the HALY.ID project, several strategies were explored for Brown Marmorated Stink Bug (BMSB) management, from field pest monitoring until post-harvest fruit sorting. For pest monitoring, NIR hyperspectral images of BMSB specimens on different plant backgrounds were used to build a spectral library for pixel-level classification models aimed at enhancing automated BSMB in field detection. For the detection of internal damages to the fruit pulp invisible to the naked eye, NIR hyperspectral imaging was evaluated as a viable method to implement a post-harvest sorting system for the early detection of BMSB punctures on pears. Organic pears (cv. Williams and cv. Abate Fétel) were collected in the summers of 2022 and 2023, with half of the fruits exposed to BMSB and the remainder used as controls. Approximately 2000 hyperspectral images were acquired for each variety. An innovative approach based on hyperspectrograms and image-level classification coupled with spatial features selection was adopted to automatically annotate BMSB punctures on the collected images. Thanks to this annotation step, a library of representative spectra belonging to both punctures and sound fruits was then built for the development of classification models. Concerning authenticity issues in agri-food, herbs and spices are frequently subjected to Economically Motivated Adulterations, which are favoured by the complexity of the supply chain. Within this framework, NIR hyperspectral imaging was evaluated as a possible screening technique to differentiate authentic oregano from oregano suspected of adulteration with leaves of plants of lower commercial value. An alternative classification method is proposed to properly address authentication issues involving strongly overlapping classes. Another part of the research investigated the benefits of applying sparse-based methods in the spatial direction for the selection of representative pixels of hyperspectral data. As with spectral variables, this method may significantly improve data compression by focusing only on the most informative spatial regions, thus reducing storage and computational demands.

La spettroscopia nel vicino infrarosso (NIR), l'imaging digitale e spettrale si stanno sempre più affermando come metodi rapidi, non distruttivi e dal basso impatto ambientale per la valutazione della sicurezza e della qualità degli alimenti. Grazie alla facilità d’uso e alla mancanza di reagenti, tali metodi risultano più sostenibili ed economici rispetto alle tradizionali tecniche analitiche. Inoltre, la spettroscopia NIR permette un'analisi non mirata, fornendo una valutazione completa del profilo chimico distintivo di un campione senza la selezione a monte di composti di interesse. L'imaging iperspettrale unisce i benefici della spettroscopia con quelli delle tecniche di imaging, grazie alla visualizzazione della distribuzione spaziale dei composti chimici di interesse sulla superficie del campione. Tuttavia, l’acquisizione di immagini iperspettrali prevede l’elaborazione di una grande mole di dati con conseguenti difficoltà computazionali, di gestione, di archiviazione e di analisi. Per superare tali limitazioni vengono spesso applicati metodi di deep learning, con ripercussioni negative a livello ambientale. In questo contesto, la chemiometria e l'analisi multivariata delle immagini rappresentano un’alternativa sostenibile. Questa Tesi di Dottorato propone diverse strategie chemiometriche che prevedono l’utilizzo dell’imaging iperspettrale nel NIR (NIR-HSI) per aumentare la sostenibilità dei sistemi agroalimentari. A tal fine, vengono presentate soluzioni pratiche per il monitoraggio in campo, la cernita in post-raccolta e l'autenticazione attraverso metodi alternativi di riduzione della dimensionalità delle immagini e di classificazione. Nell'ambito del progetto HALY.ID, sono state proposte diverse strategie per la gestione della cimice asiatica: dal monitoraggio in campo per prevenirne l’attività, all’impianto di cernita per valutare la qualità delle pere nel post-raccolta. Per il monitoraggio in campo il NIR-HSI è stato utilizzato per acquisire immagini di esemplari di cimice su diversi sfondi vegetali, utili a costruire una libreria spettrale per modelli di classificazione allo scopo di migliorare l’identificazione automatizzata in campo dell’infestante. Per il controllo qualità in post-raccolta, il NIR-HSI è stato vagliato come metodo di cernita per rilevare punture di cimice asiatica su pere, che causano difetti interni alla polpa non rilevabili ad occhio nudo. Le pere biologiche (cv. Abate Fétel, cv. Williams), raccolte durante due estati consecutive, sono state suddivise in frutti esposti all’insetto e in frutti di controllo. Attraverso un approccio innovativo basato sulla conversione delle immagini in iperspettrogrammi e l’applicazione di modelli di classificazione abbinati alla selezione di features spaziali, è stato possibile annotare automaticamente le aree riconducibili alle punture di cimice. Grazie a questa fase, è stato possibile creare un dataset di spettri di riferimento per lo sviluppo di modelli di classificazione. In tema di autenticità nel settore agroalimentare, le erbe aromatiche e le spezie sono tra i prodotti più soggetti ad adulterazione per fini economici. La ricerca condotta in questo ambito ha dimostrato che NIR-HSI può essere una valida tecnica di screening per differenziare l'origano autentico dall'origano sospettato di adulterazione, proponendo un metodo di classificazione alternativo per migliorare l’efficacia dell'autenticazione di classi fortemente sovrapposte. Un'altra parte della ricerca esplora i vantaggi nell’applicare tecniche sparse nella dimensione spaziale, al fine di selezionare i pixel più rappresentativi di immagini iperspettrali. Come per le variabili spettrali, ciò può migliorare significativamente la compressione dei dati riducendo le esigenze di archiviazione e di calcolo.

Chemiometria e tecnologie green per la sostenibilità dei sistemi agroalimentari / Veronica Ferrari , 2026 Apr 16. 38. ciclo, Anno Accademico 2024/2025.

Chemiometria e tecnologie green per la sostenibilità dei sistemi agroalimentari

FERRARI, VERONICA
2026

Abstract

Near-infrared (NIR) spectroscopy, digital imaging and spectral imaging have emerged as rapid, non-destructive, and environmentally friendly tools for assessing food safety and quality. Compared to traditional wet analytical techniques, these methods are more sustainable and cost-effective, as they don’t require skilled personnel nor chemical reagents. Opposed to destructive targeted approaches, NIR spectroscopy enables untargeted analysis, providing a comprehensive assessment of the distinctive chemical profile of a sample without prior selection of specific compounds of interest. Hyperspectral imaging combines the strengths of spectroscopy and imaging, allowing the visualization of spatial distribution of the chemical features of interest across a sample’s surface. However, its data-richness requires longer computational times thus complicating data handling, storage and analysis. These limitations are typically overcome by applying deep learning methods resulting in enormous environmental implications. In this frame, Chemometrics and Multivariate Image Analysis are a sustainable alternative for addressing the curse of dimensionality and extracting useful information. The present Doctoral Thesis proposes several strategies based on chemometrics and NIR hyperspectral imaging to increase the sustainability of the agri-food systems. To this aim, practical solutions for pest monitoring, post-harvest sorting, and for food authentication involving image dimensionality reduction and original classification methods are presented. Within the HALY.ID project, several strategies were explored for Brown Marmorated Stink Bug (BMSB) management, from field pest monitoring until post-harvest fruit sorting. For pest monitoring, NIR hyperspectral images of BMSB specimens on different plant backgrounds were used to build a spectral library for pixel-level classification models aimed at enhancing automated BSMB in field detection. For the detection of internal damages to the fruit pulp invisible to the naked eye, NIR hyperspectral imaging was evaluated as a viable method to implement a post-harvest sorting system for the early detection of BMSB punctures on pears. Organic pears (cv. Williams and cv. Abate Fétel) were collected in the summers of 2022 and 2023, with half of the fruits exposed to BMSB and the remainder used as controls. Approximately 2000 hyperspectral images were acquired for each variety. An innovative approach based on hyperspectrograms and image-level classification coupled with spatial features selection was adopted to automatically annotate BMSB punctures on the collected images. Thanks to this annotation step, a library of representative spectra belonging to both punctures and sound fruits was then built for the development of classification models. Concerning authenticity issues in agri-food, herbs and spices are frequently subjected to Economically Motivated Adulterations, which are favoured by the complexity of the supply chain. Within this framework, NIR hyperspectral imaging was evaluated as a possible screening technique to differentiate authentic oregano from oregano suspected of adulteration with leaves of plants of lower commercial value. An alternative classification method is proposed to properly address authentication issues involving strongly overlapping classes. Another part of the research investigated the benefits of applying sparse-based methods in the spatial direction for the selection of representative pixels of hyperspectral data. As with spectral variables, this method may significantly improve data compression by focusing only on the most informative spatial regions, thus reducing storage and computational demands.
Chemometrics and green technologies for a low environmental impact agri-food system
16-apr-2026
ULRICI, Alessandro
CALVINI, ROSALBA
File in questo prodotto:
File Dimensione Formato  
Ferrari.pdf

Open access

Descrizione: Ferrari.Veronica.pdf
Tipologia: Tesi di dottorato
Dimensione 12.21 MB
Formato Adobe PDF
12.21 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

Licenza Creative Commons
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11380/1401851
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact