Research Project: Advancement of Sensing Technologies for Food Safety and Security Applications

Location: Environmental Microbial & Food Safety Laboratory

2025 Annual Report

Objectives
Objective 1: Develop and validate an autonomous unmanned aerial vehicle with multimode imaging technologies for preharvest inspection of produce fields for animal intrusion and fecal contamination, and for irrigation water quality monitoring. Objective 2: Advance the development of customized compact spectral sensing technologies for food inspection and sanitation assessment in food processing, and for controlled-environment produce production, with embedded automated detection results for non-expert end users. Sub-objective 2.A: Develop a handheld line-scan hyperspectral imaging device with enhanced capabilities for contamination and sanitation inspection in food processing environments. Sub-objective 2.B: Develop a compact automated hyperspectral imaging platform for food safety and plant health monitoring for controlled-environment produce production in NASA space missions. Objective 3: Develop innovative spectroscopic and optical methods to characterize food composition and nondestructively detect adulterants and contaminants, for screening and inspecting agricultural commodities and commercially prepared food materials. Sub-objective 3.A: Develop a transportable multimodal optical sensing system for rapid, automated, and intelligent biological and chemical food safety inspection. Sub-objective 3.B: Develop a novel apparatus enabling dual-modality concomitant detection, along with associated methods and procedures, for assuring food integrity.

Approach
The overall goal of this project is to develop and validate automated sensing tools and techniques to reduce food safety risks in food production and processing environments. Engineering-driven research will develop the next generation of rapid, intelligent, user-friendly sensing technologies for use in food production, processing, and other supply chain operations. Feedback from industrial and regulatory end users, and from stakeholders throughout the food supply chain, indicates that effective automated sensing and instrumentation systems require real-time data processing to provide non-expert users with a clear understanding and ability to make decisions based on the system output. Towards this end, we will develop unmanned aerial vehicles with multimodal remote sensing platforms and on-board data-processing capability to provide real-time detection and classification of animal intrusion and fecal contamination in farm fields and of irrigation water microbial quality. We will upgrade our existing handheld imaging device for contamination and sanitation inspection with multispectral imaging and embedded computing and artificial intelligence. We are also partnering with the NASA Kennedy Space Center to develop a novel, compact, automated hyperspectral platform for monitoring food safety and plant health of space crop production systems. Food safety and integrity requires identifying adulterants, foreign materials, and microbial contamination as well as authenticating ingredients. We will develop innovative multimodal optical sensing systems utilizing dual-band laser Raman, and Raman plus infrared, for simultaneous detection on a single sampling site. Spectroscopic and spectral imaging-based methodologies will be developed to enhance detection efficacy for liquid or powder samples. These systems will be supported with intuitive, intelligent sample-evaluation software and procedures for both biological and chemical contaminants.

Progress Report
Significant progress has been made for all objectives of the project, which falls under National Program 108. For Objective 1, ARS scientists in Beltsville, Maryland, in collaboration with an industry partner, developed a real-time edge-computing video analysis pipeline on the NVIDIA Jetson platform, enabling autonomous detection of crop field anomalies from unmanned aerial vehicle(UAV)-acquired video. This AI pipeline is being integrated into the UAV-based multimodal imaging system developed in prior years, enhancing its capability for real-time analysis and autonomous decision support in field conditions. The system leverages the low-latency, high-efficiency processing capabilities of Jetson modules to support precision agriculture applications. Model training and testing began using region-specific datasets. Current development focuses on optimizing two deep learning models, YOLOv8 and SegFormer, for object segmentation from UAV video streams. Both models are being fine-tuned and accelerated using NVIDIA TensorRT to achieve real-time inference on embedded systems. To improve detection accuracy under varying field conditions, data augmentation and video stabilization techniques are being applied. These efforts advance the integration of intelligent, field-ready monitoring technology into ARS’s drone imaging platform. In support of Objective 2A, ARS scientists in Beltsville, Maryland, in cooperation with an industry licensee, advanced the commercial CSI-D (Contamination and Sanitation Inspection-Disinfection) device designed and developed based on an ARS patent (USPTO #, 8,310,544). A second U.S. patent was issued on the advanced version of the CSI device on April 15, 2025 (USPTO # 12,276,612 B2). The device uses ultraviolet (UV) A light to detect contaminants on food contact surfaces, applies ultravioloet UVC for disinfection, and records cleanliness levels. Experiments were conducted to detect and disinfect pathogenic biofilms on food contact surfaces such as stainless steel, and fecal contamination on shell egg surfaces. The experimental data have been analyzed to develop AI-models to be embedded to CSI devices capable of AI-edge computing. The ARS team developed a bench-top UV system capable of delivering controlled UVB and UVC light. The system effectively inactivated major cacao pathogens with UVC (275 nm) achieving over 90% reduction in fungal growth within minutes. Pulsed UVC and combined sonication-UVC treatments further enhanced efficacy, reducing pathogen survival to below 10%. ARS scientists in Beltsville, Maryland, collaborated with NASA Kennedy Space Center to advance hyperspectral imaging technologies for early detection of plant stress in controlled-environment agriculture for space crops. A multimodal imaging platform developed by ARS, equipped with a 3-axis gantry and sensors for reflectance and fluorescence imaging, was deployed at NASA to monitor lettuce plants under well-watered and drought-stressed conditions. Initial machine learning models using combined spectral data achieved over 90 percent accuracy in identifying stress before visible symptoms appeared. Follow-up studies further improved performance using a Vision Transformer-based deep learning model (SAM-ViT-3PE), which reached 95.4 percent accuracy and enabled detection by day 4 after stress induction. Additional modeling with PLS, SVM, and CNN confirmed robust performance across multiple experiments, with peak accuracy of 97 percent observed on day 6. To support these results, metabolomic and transcriptomic analyses identified key biochemical markers and gene expression changes associated with early drought response. These findings confirmed that spectral signatures captured by the imaging system reflect underlying physiological changes. Building on these outcomes, the research team is expanding the system’s application to other stress conditions. Ongoing experiments include assessing the impact of different LED lights on plant growth and evaluating plant responses to salt stress. The same imaging and AI analysis pipeline is being used to monitor morphological and physiological changes under these conditions. This integrated approach demonstrates strong potential for non-destructive, data-driven monitoring of plant health in both space and terrestrial environments, supporting early intervention and improved crop management. In collaboration with the University of Florida (UF) in Gainesville, Florida, we continued work to develop citrus disease detection and classification methods. Based on hyperspectral reflectance image data collected from healthy and diseased citrus leaves at Citrus Research and Education Center in Lake Alfred, Florida using a portable hyperspectral imaging system developed by ARS scientists in Beltsville, Maryland, UF collaborators developed a disease classification method based on leaf texture features and YOLOv8 network architectures. In addition, the portable hyperspectral imaging system has been used by collaborators from South Dakota State University in Brookings, South Dakota, for their research on assessing sudden death syndrome severity in soybean leaves and evaluating nitrogen content of corn leaves. For Objective 3A, in collaboration with the National Agricultural Products Quality Management Service (NAQS) in Gimcheon, South Korea, ARS scientists in Beltsville, Maryland, finalized an in-house developed multimodal optical sensing system for identification of foodborne bacteria. We added a new function in LabVIEW software for automated control of the laser for Raman spectral measurement to avoid its interference with image acquisitions using the three LED lights. Fluorescence, color, and transmission images and Raman spectra were collected from bacterial colonies of five species patched in 600 nonselective agar Petri dishes (120 each), including Bacillus cereus, Escherichia coli, Listeria monocytogenes, Staphylococcus aureus, and Salmonella enterica subsp. Enterica. We developed multimodal image and spectral processing procedures to create different datasets for classifications. We run YOLOv11 deep learning models on a new high-performance computing server using the fluorescence, color, and transmission images (600 each) to classify the five bacterial species. Also, machine learning models using nine optimizable classifiers and Raman spectra collected from a total of 14,400 (24 colonies × 600 dishes) bacterial colonies were evaluated and compared for the species classifications. In collaboration with a former CRADA partner, ARS scientists in Beltsville, Maryland, continued development of a portable multimode spectroscopy device for rapid, non-destructive detection of mycotoxins in corn. The device measures fluorescence (365 nm and 405 nm excitation) and reflectance across visible and near-infrared ranges. Custom Python-based software was developed for instrument control, calibration, and spectral data acquisition. Recent studies using ground maize samples demonstrated high classification performance. Fluorescence-based models achieved validation accuracies up to 85.5 percent, while short-wave infrared reflectance data yielded a maximum accuracy of 98.2 percent in distinguishing contaminated from uncontaminated samples. These results were comparable to laboratory-grade hyperspectral imaging systems. To enhance field applicability, the system hardware and software are being improved to enable use by non-specialists. The aim is to simplify operation and streamline data interpretation, allowing rapid screening in grain handling and processing environments. In collaboration with the University of North Dakota (UND) in Grand Forks, North Dakota, ARS scientists in Beltsville, Maryland, continued work to develop fish authentication methods to address issues of species mislabeling and fraud as well as freshness of fish fillets. UND collaborators developed a handheld multimode point spectroscopy system as an accurate and nondestructive solution to identify fish fillet species. The system utilizes a combination of fluorescence spectroscopy in visible near-infrared range and reflectance spectroscopy in visible near infrared and short-wave infrared range. A classification accuracy of 90% was achieved using machine learning to differentiate 11 species for thawed, frozen, and combined fish fillets. The CRADA partner will integrate the AI classification models into the handheld sensing device for industrial applications for on-site fish species and freshness inspection. For Objective 3B, we conducted the dual-modality infrared (IR) and Raman measurements of fish oil samples. Fish oils are rich in polyunsaturated fatty acids (PUFAs) and are nutritionally important and essential to a healthy diet, primarily due to their rich source of omega-3 fatty acids. Because of the chemical structural similarities among fish oil samples, one of the IR and Raman spectral data sets cannot precisely differentiate any single fish oil from the others. We successfully applied the dual-modality IR and Raman technique to identify a unique spectral pattern for each fish oil sample. Previous dual modality studies have found exactly the wavenumbers in IR signals were the highest (and IR signals were the lowest) plus those in which IR was the highest (and Raman intensity were the lowest) could be a powerful mechanism to distinguish other structural analogs. We found this dual modality technique is also applicable to a structurally very different set of compounds as well and can indeed enable in situ measurement in real-time of PUFA in agricultural products like salmon, beef, pork, poultry, nuts and seeds products. Omega-3 fatty acids have specific positive health benefits, and most fish oils are mixtures of multiple PUFA. The critical purpose of this study is to provide the capacity to distinguish docosahexaenoic acid (DHA) fatty acids from eicosapentaenoic acid (EPA) in situ in commercial products like fish oil supplements.

Accomplishments
1. Classification of citrus leaf diseases using hyperspectral reflectance and fluorescence imaging and machine learning techniques. Citrus diseases pose serious risks to Florida's citrus farms, leading to economic losses due to smaller fruit, surface imperfections, early fruit loss, and even tree mortality. ARS scientists in Beltsville, Maryland, developed an advanced hyperspectral imaging (HSI) method that combines reflectance and fluorescence techniques to enhance the detection and classification of citrus diseases, such as canker, Huanglongbing (HLB), greasy spot, melanose, scab, and zinc deficiency, thereby improving management and mitigation efforts in citrus groves. The combination of the full spectrum and spectral bands from the HSI with pixel-based and leaf-based spectrum were trained using nine machine learning classifiers. The highest overall classification accuracy of 90.7% was achieved by using a support vector machine (SVM) classifier and pixel-based, whereas the best accuracy of 94.5% was acquired by a discriminant analysis classifier and leaf-based analysis. The reflectance and fluorescence HSI with the machine learning techniques should assist the citrus industry and regulatory agencies (e.g., FDA and USDA APHIS) in enforcing standards for the quality and safety of citrus-related food and beverage products.

2. Detection of citrus black spot fungi using handheld fluorescence imaging and deep learning. Citrus black spot (CBS) is a quarantine fungal disease that can limit market access and thus cause economic losses for citrus growers. Early detection of CBS would enable control for the spread of the disease and prevent infected fruits from entering the packing stream. ARS scientists in Beltsville, Maryland, developed a method for detecting CBS fungus using a contamination and sanitation inspection and disinfection (CSI-D) handheld fluorescence imaging device, which was commercialized based on an ARS patented technology. UV-C fluorescence images were acquired from both CBS fungus Phyllosticta citricarpa and Phyllosticta capitalensis at six concentrations. Based on a YOLOv8 deep learning framework, average accuracies were achieved at 96.97% and 96.17% to classify the concentration levels of Phyllosticta citricarpa and Phyllosticta capitalensis, respectively. The CSI-D handheld UV-C fluorescence imaging and deep learning techniques would benefit the citrus industry and regulatory agencies (e.g., FDA and USDA APHIS) in ensuring and enforcing the quality and safety standards for citrus-related food and beverage products.

3. A handheld fluorescence imaging device for detection of fecal contaminants on poultry carcasses. Salmonella is a significant foodborne pathogen. The Centers for Disease Control and Prevention estimates that Salmonella causes more foodborne illnesses than any other bacteria. The pathogen is primarily associated with the consumption of raw and undercooked poultry. The presence of fecal matter on poultry carcass increases Salmonella food safety risks of raw poultry and hence it is important that carcasses with fecal contamination be accurately detected. ARS scientists in Beltsville, Maryland, investigated the use of the handheld multispectral fluorescence imaging technology combined with machine learning methods to detect fecal matter on poultry carcasses. Results indicated that fluorescence imaging combined with machine learning models can detect fecal contamination on poultry carcasses. Further detection of Salmonella on carcasses after rinsing-off fecal contamination in the lab and in-plant setting provides preliminary evidence in warranting next steps to reduce food safety risks. These results are expected to provide a new approach for detection of fecal contamination on chicken carcasses which will benefit the poultry processing industry and the USDA FSIS to further investigate novel technologies to improve Salmonella food safety of raw poultry and reduce food safety risks.

4. A portable spectral sensing technology for rapid and accurate detection of chemical food spoilage to reduce food safety risks. Meat spoilage is the degradation of a meat’s chemical components including water, protein, and fat. Total viable count (TVC) is a measure of the total amount of living microorganisms present in a sample and is used as a key indicator of meat freshness. TVC values for meat increase as meat spoilage occurs. ARS scientists in Beltsville, Maryland, developed a portable detection technology combined with machine learning techniques to predict the TVC in pork samples. The portable device acquires measurements of light reflected from the meat samples and automatically processes the measurements using a classification model to produce analysis results and can store and display the results for easy use by non-experts. Because image scanning and complex three-dimensional (3D) image processing are not required, the portable device can be used to analyze over 30 samples per minute. This portable technology is adaptable to a variety of scenarios in meat processing operations, such as slaughtering lines, storage facilities, transportation trucks, and sales markets.

Review Publications
Tao, F., Yao, H., Hruska, Z., Rajasekaran, K., Qin, J., Kim, M., Chao, K. 2024. Raman hyperspectral imaging as a potential tool for rapid and nondestructive identification of aflatoxin contamination in corn kernels. Journal of Food Protection. 87(9). 100335. https://doi.org/10.1016/j.jfp.2024.100335.
Chao, K., Schmidt, W.F., Qin, J., Kim, M.S., Tao, F. 2024. IR and Raman dual modality markers differentiate among three bis-phenols: BPA, BPS and BPF. Applied Sciences. 14(14): Article e6064. https://doi.org/10.3390/app14146064.
Frederick, Q., Burks, T.F., Yadav, P., Qin, J., Kim, M.S., Dewdney, M.M. 2024. Classifying adaxial and abaxial sides of diseased citrus leaves with selected hyperspectral bands and YOLOv8. Smart Agriculture. https://doi.org/10.1016/j.atech.2024.100600.
Min, H., Qin, J., Yadav, P., Frederick, Q., Burks, T., Dewdney, M., Baek, I., Kim, M.S. 2024. Classification of citrus leaf diseases using hyperspectral reflectance and fluorescence imaging and machine learning techniques. Horticulturae. 10. Article e1124. https://doi.org/10.3390/horticulturae10111124.
Yadav, P., Burks, T., Qin, J., Kim, M.S., Dewdney, M., Vasefi, F. 2024. Detection of citrus black spot fungi Phyllosticta citricarpa and Phyllosticta capitalensis on UV-C fluorescence images using YOLOv8. Smart Agricultural Technology. https://doi.org/10.1016/j.atech.2024.100615.
Frederick, Q., Burks, T., Watson, J., Yadav, P., Qin, J., Kim, M.S., Dewdney, M. 2025. Supervised hyperspectral band selection using texture features for classification of citrus leaf diseases with YOLOv8. Sensors. 25(4). Article e1034. https://doi.org/10.3390/s25041034.
Baek, I., Bhatt, J., Jang, J., Seunghyun, L., Lovelace, A.H., Minhyeok, C., Lakshman, D., Kim, M.S., Meinhardt, L.W., Park, S., Ahn, E.J. 2025. Dissecting trichoderma antagonism: role of strain identity, volatiles, biomass, and morphology in suppressing cacao pathogens. Biological Control. https://doi.org/10.1016/j.biocontrol.2025.105807.
Hernanda, R., Kim, J., Faqeerzada, M., Cho, B., Kim, M.S., Baek, I., Lee, H. 2024. Rapid and noncontact identification of soybean flour in edible insect using NIR spectral imager: A case study in Protaetia brevitarsis seulensis powder. Food Control. 169. Article e111019. https://doi.org/10.1016/j.foodcont.2024.111019.
Pahlawan, M.F., Kim, Y., Aline, U., Zahroh, A., Masithoh, R.E., Kim, M.S., Baek, I., Cho, B. 2025. Non-destructive identification of microplastics in soil using spectroscopy and hyperspectral imaging. Trends in Analytical Chemistry. 187: 118216. https://doi.org/10.1016/j.trac.2025.118216.
Kim, J., Faqeerzada, M., Kim, M.S., Baek, I., Cho, B. 2025. Characteristics of foot-and-mouth disease (FMD) vaccine-induced abscesses in pork meat and verification of detectability using hyperspectral imaging technique. Food Control. 175: 111330. https://doi.org/10.1016/j.foodcont.2025.111330.
Vega-Castellote, M., Sanchez, M., Kim, M.S., Hwang, C., Pezez-Marin, D. 2024. Investigating the detection of peanuts in chopped nut products using hyperspectral imaging systems. Journal of Food Engineering. 338: 112378. https://doi.org/10.1016/j.jfoodeng.2024.112378.
Lee, A., Hong, S., Baek, I., Kim, J., Kim, M.S. 2025. Deep learning approaches for bruised mandarin orange classification via fluorescence hyperspectral imaging. Postharvest Biology and Technology. 230: 113724. https://doi.org/10.1016/j.postharvbio.2025.113724.
Lee, J., Kim, M., Lee, B., Lee, J., Yang, E., Kim, M.S., Hwang, I., Jeong, C., Mo, C. 2025. Evaluating ripeness in post-harvest stored kiwifruit using VIS-NIR hyperspectral imaging. Postharvest Biology and Technology. 225: 113496. https://doi.org/10.1016/j.postharvbio.2025.113496.
Lim, S., Baek, I., Hong, S., Lee, Y., Kirubakaran, S.J., Kim, M.S., Meinhardt, L.W., Park, S., Ahn, E.J. 2025. Cacao floral traits are shaped by the interaction of flower position with genotype. Heliyon. https://doi.org/10.1016/j.heliyon.2025.e42407.
Semyalo, D., Kim, Y., Emmanuel, O., Arief, M., Kim, H., Sim, E., Kim, M.S., Baek, I., Cho, B. 2024. Nondestructive measurement of internal potato defects using visible and near-infrared spectral analysis. Agriculture. 14. Article e14112014. https://doi.org/10.3390/agriculture14112014.
Faqeerzada, Mohammad, Kim, Ye-Na, Kim, Haeun, Akter, Tanjima, Kim, Hangi, Park, Min-Seok, Kim, M.S., Baek, I., Cho, Byoung-Kwan 2024. Hyperspectral imaging system for pre- and post-harvest defect detection in paprika fruit. Postharvest Biology and Technology. 118. Article e213151. https://doi.org/10.1016/j.postharvbio.2024.113151.
Cho, S., Soleh, H.M., Choi, J., Hwang, W., Lee, H., Cho, B., Kim, M.S., Baek, I., Kim, G. 2024. Recent methods for evaluating crop qater stress using AI techniques: A review. Sensors. 24. Article e6316. https://doi.org/10.3390/s24196313.
Baek, I., Lim, S., Weerarathne, V., Leedongho, Botkin, J., Kirubakaran, S.J., Park, S., Kim, M.S., Meinhardt, L.W., Ahn, E.J. 2025. Spatial patterning of chloroplasts and stomata in developing cacao leaves. Communications Biology. https://doi.org/10.1038/s42003-025-08019-6.
Lim, S., Park, S., Baek, I., Botkin, J., Jang, J., Hong, S., Irish, B.M., Kim, M.S., Meinhardt, L.W., Curtin, S.J., Ahn, E.J. 2025. Integrative analysis of seed morphology, geographic origin, and genetic structure in Medicago with implications for breeding and conservation. BMC Plant Biology. https://doi.org/10.1186/s12870-025-06304-4.
Baek, I., Lim, S., Jang, J., Hong, S., Prom, L.K., Kirubakaran, S.J., Cohen, S.P., Lakshman, D.K., Kim, M.S., Meinhardt, L.W., Park, S., Ahn, E.J. 2025. Pathogen-specific stomatal responses in cacao leaves to Phytophthora megakarya and Rhizoctonia solani. Scientific Reports. https://doi.org/10.1038/s41598-025-94859-5.
Aliee, M., Gorji, H.T., Vasefi, F., Yaggi, K., Qin, J., Baek, I., Kim, M.S., Chan, D.E., Johnson, M., Downs, Z., Marateb, H.R., Tavakolian, K., Liang, B. 2025. GLOW-DL: Generalized light-optimized workflow with deep learning for contamination detection using fluorescence imaging in variable conditions . Journal of Biosystems Engineering. 50:240-254. https://doi.org/10.1007/s42853-025-00262-3.
Faqeerzada, M.A., Kim, H., Kim, M.S., Baek, I., Chan, D.E., Cho, B. 2025. Hyperspectral imaging Vis-NIR and SWIR fusion for improved drought-stress identification of strawberry plants. Computers and Electronics in Agriculture. 237 Part C. Article e10702. https://doi.org/10.1016/j.compag.2025.110702.
Lee, H., Shin, J., Kim, S., Kim, M., Kim, M.S., Lee, H., Mo, C. 2025. Enhancing bee mite detection with YOLO: the role of data augmentation and stratified sampling. Agriculture. 15(11). Article e1221. https://doi.org/10.3390/agriculture15111221.
Ahn, E.J., Park, S., Baek, I., Lee, D., Bhatt, J., Lim, S., Jang, J., Zhang, D., Kim, M.S., Meinhardt, L.W. 2025. Machine learning-driven GWAS uncovers novel candidate genes for resistance to frosty pod rot and witches' broom disease in cacao. The Plant Genome. 18(3). Article e70069. https://doi.org/10.1002/tpg2.70069.