Skip to main content
ARS Home » Southeast Area » Athens, Georgia » U.S. National Poultry Research Center » Quality and Safety Assessment Research Unit » Research » Publications at this Location » Publication #411891
Research Project: Smart Optical Sensing of Food Hazards and Elimination of Non-Nitrofurazone Semicarbazide in Poultry

Location: Quality and Safety Assessment Research Unit

Title: Rapid and data-efficient classification of Salmonella serovars via image augmentation and deep learning on hyperspectral microscope images

Author
item WASIT, AARHAM - Michigan State University
item Park, Bosoon
item YI, JIYOON - Michigan State University

Submitted to: International Association for Food Protection
Publication Type: Abstract Only
Publication Acceptance Date: 5/17/2024
Publication Date: N/A
Citation: N/A

Interpretive Summary:

Technical Abstract: Addressing the need for rapid and accurate detection of foodborne pathogens, hyperspectral microscope imaging (HMI) has demonstrated promising results in classifying bacterial species, including Salmonella, based on their unique spectral signatures. This study aimed to enhance the utilization of spatial features in high-dimensional HMI data to enable serovar-level classification by developing image augmentation and deep learning algorithms. Darkfield HMI data of five different Salmonella serovars, including Enteritidis, Typhimurium, Kentucky, Heidelberg, and Infantis, were collected from pure bacterial isolates and a total of 243 raw HMI data were pre-processed to extract spatial data for model training. An artificial intelligence (AI) model was developed using a lightweight deep learning image classification algorithm, pre-trained on the large dataset, and subsequently trained with our HMI data for data-efficient transfer learning. Additionally, a combination of various image augmentation algorithms, relevant to the data acquisition process (e.g., illumination, target orientations, focusing, geometric transformations, sensor noise, and missing information), was designed for bacterial datasets to enhance data efficiency and model generalizability. The model performance was evaluated using a confusion matrix, accuracy, precision, and recall at different bacterial incubation times ranging from 6-24 h. Overall, the AI model prediction results for Salmonella serovar classification based solely on spatial data from HMI demonstrated high accuracy, ranging between 0.80-0.99, depending on the incubation time. The results also indicated that classification accuracy was significantly improved by employing image augmentation during model training. The results support the robustness of AI-assisted HMI as an effective method for rapid and accurate foodborne pathogen detection at the serovar level.