Non-Invasive Methods for Predicting the Quality of Processed Horticultural Food Products, with Emphasis on Dried Powders, Juices and Oils: A Review

Machine Learning Application in Horticulture and Prospects for Predicting Fresh Produce Losses and Waste: A Review

The current review examines the state of knowledge and research on machine learning (ML) applications in horticultural production and the potential for predicting fresh produce losses and waste. Recently, ML has been increasingly applied in horticulture for efficient and accurate operations. Given the health benefits of fresh produce and the need for food and nutrition security, efficient horticultural production and postharvest management are important. This review aims to assess the application of ML in preharvest and postharvest horticulture and the potential of ML in reducing postharvest losses and waste by predicting their magnitude, which is crucial for management practices and policymaking in loss and waste reduction. The review starts by assessing the application of ML in preharvest horticulture. It then presents the application of ML in postharvest handling and processing, and lastly, the prospects for its application in postharvest loss and waste quantification. The findings revealed that several ML algorithms perform satisfactorily in classification and prediction tasks. Based on that, there is a need to further investigate the suitability of more models or a combination of models with a higher potential for classification and prediction. Overall, the review suggested possible future directions for research related to the application of ML in postharvest losses and waste quantification.

Comparative Analysis of Concurrent (CC), Mixed Flow (MX), and Combined Spray Drying Configurations on the Physicochemical Characteristics of Satsuma Mandarin (Citrus unshiu) Juice Powders

Satsuma mandarins are good sources of vitamin C and can be used as raw materials to produce novel plant-based food ingredients including satsuma mandarin juice powders (SJP). Food powders produced via spray drying often show thermal degradation due to the drying conditions and high drying air temperatures. The aim of this study was to evaluate the effect of using different spray drying configurations, including concurrent (CC), mixed flow (MX), and combined (CC + MX), at two inlet air temperatures (160 and 180 °C) on the physicochemical properties of SJP. Remarkably, SJP produced using the CC spray drying configuration exhibited a higher vitamin C content (3.56-4.01 mg/g) and lower moisture levels (15.18-16.35 g/100 g) than powders produced via MX or CC + MX. The vitamin C content of MX and CC + MX powders ranged from 2.88 to 3.33 mg/g. Meanwhile, all SJP had water activity values below 0.19. Furthermore, MX powders displayed the largest mean particle sizes (D50) (8.69-8.83 µm), higher agglomeration, and a rapid dissolution. Despite these differences, all SJP variants exhibited consistent color, surface area, and pore volumes. Notably, powders dried at higher inlet air temperatures (180 °C) showed less vitamin C content and increased thermal damage when compared with powders dried at 160 °C inlet air temperature. This study demonstrated the feasibility of producing high-quality SJP with an extended shelf life. SJP can be used as a novel plant-based ingredient in different food applications.

Vis-NIR and SWIR hyperspectral imaging method to detect bruises in pomegranate fruit

Introduction

Fresh pomegranate fruit is susceptible to bruising, a common type of mechanical damage during harvest and at all stages of postharvest handling. Accurate and early detection of such damages in pomegranate fruit plays an important role in fruit grading. This study investigated the detection of bruises in fresh pomegranate fruit using hyperspectral imaging technique.

Methods

A total of 90 sample of pomegranate fruit were divided into three groups of 30 samples, each representing purposefully induced pre-scanning bruise by dropping samples from 100 cm and 60 cm height on a metal surface. The control has no pre-scanning bruise (no drop). Two hyperspectral imaging setups were examined: visible and near infrared (400 to 1000 nm) and short wavelength infrared (1000 to 2500 nm). Region of interest (ROI) averaged reflectance spectra was implemented to reduce the image data. For all hypercubes a principal components analysis (PCA) based background removal were done prior to segmenting the region of interest (ROI) using the Evince® multi-variate analysis software 2.4.0. Then the average spectrum of the ROI of each sample was computed and transferred to the MATLAB 2022a (The MathWorks, Inc., Mass., USA) for classification. A two-layer feed-forward artificial neural network (ANN) is used for classification.

Results and discussion

The accuracy of bruise severity classification ranged from 80 to 96.7%. When samples from both bruise severity (Bruise damage induced from a 100cm and 60 cm drop heights respectively) cases were merged, class recognition accuracy were 88.9% and 74.4% for the SWIR and Vis-NIR, respectively. This study implemented the method of selecting out informative bands and disregarding the redundant ones to decreases the data size and dimension. The study developed a more compact classification model by the data dimensionality reduction method. This study demonstrated the potential of using hyperspectral imaging technology in sensing and classification of bruise severity in pomegranate fruit. This work provides the foundation to build a compact and fast multispectral imaging-based device for practical farm and packhouse applications.

Domain Adaptation for In-Line Allergen Classification of Agri-Food Powders Using Near-Infrared Spectroscopy

The addition of incorrect agri-food powders to a production line due to human error is a large safety concern in food and drink manufacturing, owing to incorporation of allergens in the final product. This work combines near-infrared spectroscopy with machine-learning models for early detection of this problem. Specifically, domain adaptation is used to transfer models from spectra acquired under stationary conditions to moving samples, thereby minimizing the volume of labelled data required to collect on a production line. Two deep-learning domain-adaptation methodologies are used: domain-adversarial neural networks and semisupervised generative adversarial neural networks. Overall, accuracy of up to 96.0% was achieved using no labelled data from the target domain moving spectra, and up to 99.68% was achieved when incorporating a single labelled data instance for each material into model training. Using both domain-adaptation methodologies together achieved the highest prediction accuracies on average, as did combining measurements from two near-infrared spectroscopy sensors with different wavelength ranges. Ensemble methods were used to further increase model accuracy and provide quantification of model uncertainty, and a feature-permutation method was used for global interpretability of the models.