Conference Agenda

Food adulteration remains a major concern for consumer protection and food quality, whether introduced intentionally or unintentionally. Detecting corn flour mixed into chickpea flour is particularly challenging because the adulterant closely resembles the authentic product, and conventional analytical techniques are often destructive and labor-intensive. This study presents a rapid, chemical-free, and non-destructive approach using short-wave infrared (SWIR) hyperspectral imaging (HSI) combined with machine learning to identify and quantify adulteration in chickpea flour. Samples were prepared with 0–100% (w/w) corn flour at 15 adulteration levels, along with pure controls, resulting in 442 hyperspectral images collected across 26 replicates. Spectral data extracted from processed images were used to develop classification models with linear discriminant analysis (LDA) and linear support vector machines (LSVM) to differentiate adulteration levels. Furthermore, partial least squares regression (PLSR) and support vector machine regression (SVMR) models were constructed to estimate protein and starch content. Key wavelengths were selected using competitive adaptive reweighted sampling (CARS) and iteratively retaining informative variables (IRIV). The smooth-1st derivative-LSVM model achieved the highest classification accuracy, while the smooth-SNV-IRIV-SVMR approach delivered precise predictions of nutritional components. These results demonstrate strong potential for SWIR-HSI to support real-time flour authenticity assessment.

The growing prevalence of foodborne contamination and associated illnesses necessitates the development of rapid, sensitive, and reliable strategies for the early detection of microbial pathogens. In this work, a dual-mode portable biosensing platform is developed for the detection of Salmonella typhimurium in complex food matrices. The system integrates a reflectance-based optical biosensor, employing an indirect antibody assay and horseradish peroxidase (HRP)-mediated signal amplification and photothermal analysis. Insoluble enzymatic reaction products generated within a porous silicon matrix induce changes in the reflectance spectra, enabling quantitative optical detection. In parallel, HRP-catalyzed oxidation of 3,3′,5,5′-tetramethylbenzidine (TMB) produces a blue color which absorbs laser light (NIR) and produce a photothermal signal, with the reaction temperature increasing according to logarithmic concentration. The specificity and selectivity of the anti-Salmonella antibody were validated against individual and mixed non-target bacterial strains. Under optimized conditions, the platform demonstrated a low detection limit and a linear response curve over a concentration range of 10⁰–10⁴ CFU mL^-1 using both detection modes. Practical applicability was confirmed through spiked food samples, including chicken, tahina, and chocolate, showing good recovery, reproducibility, and specificity. The label-free interferometric system operates independently of external light sources, offering a robust, field-deployable solution for rapid food safety monitoring.

Detection of surface defects plays a crucial role in the classification and grading of fruits. With the increasing market demand for high-quality produce, accurate and reliable fruit defect detection has become essential in the agricultural and food industries. Traditional manual inspection methods are labour-intensive, time-consuming, and prone to errors caused by fatigue and subjective judgment. Consequently, there is a growing need for automated fruit quality inspection systems. The study developed a system for detecting surface defects on fruits using image processing techniques. The proposed approach a Convolutional Neural Network was utilized for defect detection and trained on a large dataset consisting of both defective and non-defective fruit images. Data augmentation techniques were applied to enhance feature diversity and improve model robustness. Feature extraction was performed using a pre-trained model to improve learning efficiency. The trained model was implemented in an automated fruit defect detection system designed for classification based on surface defect. The system’s performance was evaluated and it demonstrated high effectiveness, achieving an accuracy of 95%, precision of 0.90, recall of 0.90, and F1-score of 0.89. The proposed method was efficient and accurate for fruit surface defect detection, with a potential for deployment in automated fruit inspection processes.

Compression loads applied during storage and handling cause permanent deformation in fruit tissues, inducing softening and leaving visible pressure marks which affect postharvest performance. These pressure marks are likely a result of plastic deformation of cell walls and a reduction in the intercellular air spaces. This study investigated the detectable cellular‑scale changes in ‘Zesy002’ kiwifruit subjected to controlled compressive loading (0.7–32 N) using high‑resolution optical coherence tomography (OCT) imaging integrated with computational image analysis. A custom processing pipeline was developed to perform surface flattening, noise‑reduction, 3D segmentation, and extraction of geometric descriptors including cell flatness, sphericity, extent, and centroid displacement. Tissues compressed with 8–32 N forces showed significant decreases in sphericity and extent; at 32 N force, cell centroids shifted toward the surface and flattening increased. Minimal changes occurred at 0.7 N compression. These results confirm that compression damage induces irreversible microstructural reorganisation. The integration of OCT with computational analysis provides a non‑destructive engineering tool for evaluating mechanical responses in biological tissues, supporting data‑driven assessment of load limits, packaging design, and cultivar‑specific susceptibility to compression damage within postharvest systems.

Dragon fruit is a high‑value crop whose commercial acceptance largely depends on internal quality traits, particularly sugar content, a key factor influencing consumer preference. Developing nondestructive methods to estimate sugar levels is therefore important for improving postharvest sorting and classification. This study aimed to predict the sugar content of dragon fruits using external RGB images analyzed through supervised machine learning techniques. A total of 651 fruits from the 2023 season and 931 from 2024, belonging to three different varieties, were evaluated. The dataset was split into 70% for model development (80% training and 20% validation) and 30% reserved as an independent test set. For each sample, external images were acquired to extract color, size, and texture features, while destructive laboratory measurements (weight, firmness, and sugar content) served as ground truth for calibration. Model performance was assessed using a ±6 °Brix tolerance reflecting the physiological range of the fruit. Test accuracies reached 93.51% for 2023 and 94.62% for 2024, whereas training–validation accuracies were 98.60% and 99.23%, respectively. The consistent performance across seasons indicates strong generalization. Future work will focus on enlarging the dataset and improving image‑derived features to further enhance prediction accuracy.

Acrylamide, a carcinogenic compound formed during high-temperature processing of foods containing asparagine and reducing sugars, poses significant public health concerns. Since its detection in food in 2002, extensive research has aimed to mitigate its presence, resulting in regulatory measures such as the European Commission's Regulation (EU) 2017/2158 and the FoodDrinkEurope "Acrylamide Toolbox." These frameworks emphasize the monitoring and reduction of acrylamide levels in food products with high-frequency intake, including cereal-based items, potato products, and coffee. Despite these efforts, recent EFSA assessments indicate inconsistent reductions in acrylamide levels, underscoring the need for innovative and reliable tools to improve compliance and quality control in food production.

This study investigates the feasibility of using visible and near-infrared (VNIR) spectroscopy and near-infrared (NIR) spectral imaging as rapid, non-destructive tools for acrylamide evaluation in biscuits, with the goal of developing a robust Process Analytical Technology (PAT) solution. Samples were collected directly from the production line after the cooking phase. A Design of Experiment (DoE) approach considering seven production factors was implemented.

Overall, results demonstrated promising performances for an initial massive screening of the production trend (Accuracy=91%).

Acrylamide is an organic compound identified as potential human carcinogen. Up to date, the content of acrylamide in potato chips is quantified by methods like High-Performance Liquid Chromatography. Non-destructive technologies are being explored such as near infrared spectroscopy and hyperspectral imaging (HSI). In this study, 300 potato chips were used. From them, 128 samples exceeded the recommended limit by the European Food Safety Authority (750 μg/kg) whereas 172 were suitable. Spectra were recorded with a Xeva 1.7–320-100 Hz hyperspectral camera (900–1700nm) and a MicroNIR™ 1700 OnSite-W (908–1676nm). Since there is a trend towards the use of handheld devices, one strategy to guarantee robustness and accuracy is to transfer the calibrations from high resolution spectrometers to miniaturized ones. Piecewise direct standardization (PDS) and direct standardization were applied to transfer the calibrations from the HSI camera to the MicroNIR. Random forest (RF) classification algorithms were built after reducing the influence of the potato variety. After PDS, the RF model reached an accuracy of 96% correctly classified samples in Train and 87% in Test. Thus, this work advances in calibration transferability from benchtop to hand-held instruments, a key step towards their deployment in industrial quality control.