To obtain safe products, it is necessary to ensure the integrity of each link in the food chain, including animal feed, which is the first step in the chain. This highlights the need for rapid and non-destructive analytical tools, such as Near Infrared Spectroscopy (NIRS), to meet the levels of control currently required in the feed industry. The aim of this research was to evaluate the potential of incorporating, at online level in the plant, a new generation of NIRS sensors for the quality control of compound animal feeds. The results indicated that NIRS is suitable for on-line use in the feed industry, providing reliable, non-destructive and accurate analysis of feed quality parameters.

This study focuses on the development of real-time inline sensing technology using high-speed hyperspectral imaging (HSI) and high-performance Deep Learning (DL) for the detection of Foreign Materials (FMs) on chicken breast meat. HSI and DL are useful tools for assessing agricultural and food products' safety and quality features. However, most HSI-based DL models for food quality and safety assessment lack real-time sensing capabilities critical for industrial deployment, due to the high computational demands of both HSI and DL. Therefore, we propose near-infrared hyperspectral imaging (NIR-HSI) coupled with a DL model based on a semi-supervised generative adversarial network, suitable for high-speed food production applications such as real-time inline sensing for detecting FMs during poultry processing. A line-scan NIR-HSI camera (1000-1700 nm) is used for data acquisition. For real-time imaging and DL inferencing, the software system is implemented in C++. This technology will enable the rapid and accurate detection of FMs in hyperspectral images, suggesting applications in other food products.

Hyperspectral Imaging (HSI) emerges as a non-destructive solution for assessing the quality of Iberian ham, a luxury Spanish product. Traditional quality controls involve costly and time-consuming chemical analysis and genotyping. HIS is a suitable tool to deal with such heterogeneous products, since it allows to acquire the whole surface of the sample and to know the spatial distribution of the main composition parameters, obtaining a more representative information. This study optimized a HSI system operating between 900-1700 nm for sliced Iberian ham quality assessment. After analyzing 104 samples, the most optimal region of interest for the subsequent development of prediction models was selected. Partial Least Regression (PLS) models were developed for the prediction of the content of salt, fat and proteins. The research demonstrates HSI's potential for fast, non-destructive quality evaluation, aiding producers in maintaining premium standards.

The European Commission has stablished standards to be applied to all the olive oils subject to international trade. The standards include what is called conformity checks to know the physico-chemical and organoleptic properties, for labelling and category requirements (Delegated Regulation (EU) 2022/2104). However, most of those methods are inaccessible to most producers and retailers. Furthermore, the high cost and time required to obtain the data means that the number of samples inspected per year is very low in relation to the total volume of olive oils produced. There is therefore a growing and urgent demand for novel, fast and low-cost analytical methods to guarantee the authenticity and integrity of olive oils. The present work will provide scientific evidence of the potential of a handheld Linear Variable Filters (LVF) NIRS instrument for the on-site quality control of olive oils.

Foodborne illnesses are a significant threat to seniors, with high hospitalization and mortality rates among those aged 65+. SafetySpect's CSI Technology enhances cleanliness in senior living facilities. This tech identifies, records, and eliminates contamination residues on high-touch surfaces in real-time, integrating seamlessly into Sanitation Standard Operating Procedures (SSOPs). This approach reduces the risk of foodborne illnesses and enhances resident satisfaction while lowering costs. Embracing CSI Technology empowers facilities to proactively manage contamination risks, ensuring safety and high hygiene standards for vulnerable seniors.

The purpose of this study is to develop a bee mite detection model using hyperspectral images of bee combs and deep learning-based recognition algorithms. The hyperspectral image has a resolution of 510 × 270 and consist of 15 wavebands in the ranging from 611 nm to 850 nm. Image processing was applied to preprocess data and used for data augmentation. Convolutional Neural Network (CNN) was used to develop the bee and bee mite detection models. The developed bee mite detection model was combined with an algorithm for localization to detect honey bees in hyperspectral bee comb images. Consequently, the bee mite detection model using snapshot hyperspectral image was developed to recognize the bee mite.

Plant diseases jeopardize global food production, causing substantial yield and quality losses. Swift identification is vital for effective disease management, minimizing losses, and controlling costs. This study evaluated EfficientNetB4, a convolutional neural network, for rust disease detection in three key crops. The dataset, encompassing 857 positive and 907 negative samples from diverse environments, underwent a 70%-30% split for training and testing. Through rigorous optimization of optimizers and learning rates, results showcased EfficientNetB4's efficacy with a remarkable 94.29% average accuracy. The Adaptive Moment Estimation (Adam) optimizer excelled, paired with a 0.001 learning rate. These findings underscore the potential of deep learning, notably EfficientNetB4, in enhancing plant disease identification, contributing significantly to more efficient agricultural disease management. This research addresses immediate detection challenges and lays the groundwork for advancing agricultural technology and promoting global food security and sustainability.

Potatoes, often referred to as "earth apples," are globally cultivated crops known for their high vitamin C content, containing three times more vitamin C than apples, along with rich potassium and carbohydrates. While potatoes thrive in cold and harsh environments, they are susceptible to heat stress. Alarmingly, the International Potato Center predicts that ongoing global warming could lead to a significant decline of up to 68% in potato production by 2060. The primary goal of this research is to predict the Crop Water Stress Index (CWSI) in both the temperature gradient and conventional greenhouses and to classify stress conditions with transfer learning

Ensuring the supply of high-quality sweet potatoes to consumers requires efficient sorting of harvested produce, a task made complex by various factors. This paper introduces a prototype of an innovative automated sweet potato sorting system and its preliminary evaluation. The system integrates a machine vision-based grading module and a pneumatic actuation cylinder-based sorting mechanism. The vision system captures multiple views of rotating sweet potatoes on a conveyor, utilizing a deep learning algorithm to track and grade them based on size, shape, and surface defects. The integrated sorting mechanism, activated by a computer-controlled cylinder, automatically segregates the sweet potatoes into designated areas based on quality grades. Future experimentation aims to quantify the efficacy of the integrated system, promising a potentially valuable tool for sweet potato packers.

The food service sector continually faces contamination risks, demanding advanced detection solutions. This study introduces the Contamination Sanitization Inspection (CSI) handheld scanner, employing fluorescence imaging and the You Only Look Once (YOLO) architecture for real-time contamination detection. Utilizing a dataset from 11 institutional kitchens and restaurants, the model achieved an 83% mean Average Precision (mAP). Deployed on two edge computing platforms, the CSI scanner exemplifies real-time, on-site contamination detection, offering a substantial stride toward elevating food safety standards and mitigating foodborne illnesses in diverse food service settings.

Food deception is a worldwide concern. Incorrect labeling of fish, costing the US $13B with 40% inaccurately tagged, underscores this concern's magnitude. Additionally, 53M tons of wasted meat and poultry each year exacerbate the situation. The FDA's proactive steps, with rigorous data and traceability rules, target this problem. Still, a straightforward authentication process for the supply chain remains elusive. The repercussions, surpassing the annual $40 Billion financial burden, also pose health dangers, erode consumer confidence, and devalue brands. Responding to this need, we present an innovative portable multi-spectroscopy tool for Quality Adulteration and Traceability (QAT). Utilizing fluorescence at 365 and 405 nm, VisNIR, and SWIR, this tool aids in identifying fish types and evaluating freshness. Various machine learning techniques were employed on this data. The promising outcomes in distinguishing fish types and gauging freshness hint at the potential of this spectroscopy approach to replace traditional, expensive lab procedures.

In broiler breeder production, the level of abdominal fat is of utmost importance as it affects their reproductive performance. The industry relies on subjective palpation of the pelvic bones, which depends on the operator's experience, to assess subcutaneous fat and readiness for light stimulation. An alternative involves euthanizing the bird, but this is impractical for a large number of birds in a flock. Therefore, NIRS technology is postulated as a promising tool for in vivo determination of pelvic fat in pullet broiler breeders. This work aims to evaluate the use of NIR portable sensors for this purpose, attempting to optimize an efficient analysis methodology. The results suggest that NIRS technology offers a non-destructive and easy-to-use solution to improve in vivo assessments in the farm.

International standards for Olive Oil (OO) analysis face challenges, especially with the rampant adulteration of Extra Virgin Olive Oil (EVOO). As demands grow for innovative methods beyond conventional techniques, VIS + NIRS spectroscopy emerges prominently. This study postulates enhanced efficacy through integrating VIS + NIRS with Fluorescence spectroscopy. Addressing challenges like instrument optimization and data security is paramount. Our research evaluates a hand-held multi-mode spectroscopy system combining fluorescence and reflectance. Employing three prototypes, we analyzed EVOO, VOO, and LOO categories. Results, compared against a benchtop instrument, provide insights into tackling EVOO adulteration through advanced spectral sensing.

With the rising demand for strict cleanliness checks in kitchens to identify raw food surface residue, the traditional ATP swab tests for detecting biological contamination are facing challenges. Although ATP tests can identify contamination by detecting a compound present in most biological residues, they sometimes fail to detect invisible contaminants, leading to inaccurate results. A potential solution is the fluorescence imaging systems that can spot unseen contamination from food remnants, microbes, and molds. Our team introduced the Contamination, Sanitization Inspection, and Disinfection device (CSI-D+) to pinpoint contaminated spots, enhancing the accuracy of sampling. We tested contamination residues of various raw meats on different cutting board materials, imaged them using the CSI Technology, and subsequently assessed them using ATP swabs. This research aims to correlate imaging results with those of ATP to understand the effectiveness of CSI Technology in detecting food contamination.

To address the need for compact and automated sensing systems in NASA's controlled-environment space crop production, we introduce "MoonLight." This innovative illumination system is tailored for hyperspectral reflectance and fluorescence imaging of plants, primarily focusing on early stress detection within the Visible and Near-Infrared (VNIR) range (400–1000 nm). Moonlight integrates white, and UVA LED arrays for VNIR broadband and 365 nm illumination alongside a line-scan hyperspectral camera and a linear motorized stage (0.5 m travel). Its design includes two parallel L.E.D. metal back substrates with narrow-angle L.E.D. lenses, ensuring a lightweight, compact system. An additional elliptical line collimator enhances spatial target line collimation while advanced thermal management maintains spectral stability, meeting the stringent demands of space missions.

The seafood industry faces challenges in identifying fish species and assessing freshness. Approximately 20% of fish are mislabeled due to their similar appearance, and there's no quick and cost-effective method to determine freshness. Current fish identification involves DNA analysis and polymerase chain reaction, which are time-consuming, costly, and require specialized equipment and personnel. Traditional freshness assessments involve sensory evaluation, but this method is invasive and requires skilled labor. Our team introduced a hand-held spectroscopy system that combines various spectroscopic modes for identification and freshness grading. Using this device, we studied fifteen fish samples from three species over ten days, validating species through DNA barcoding and freshness via ATP and K-value. We used the data, which employed spectroscopic fusion at the feature and decision levels, to train machine learning models leading to a system capable of accurately determining both the species and its freshness.

This paper presents an innovative approach for early detection of wheat diseases, particularly Bacterial Leaf Streak (BLS) and Scab, using a combination of hyperspectral, infrared, and RGB imaging along with Deep Convolutional Neural Networks (DCNNs). The method leverages both spatial and spectral information from wheat seed images, achieving remarkable disease classification accuracy. Advanced image preprocessing, segmentation, and feature extraction techniques are applied, and attention mechanisms enhance model robustness. The study's results outperform existing techniques, demonstrating the potential of multimodal data integration and deep learning in precision agriculture for effective wheat disease management, ultimately leading to increased global agricultural yields and reduced losses.