Detection and visualization of Escherichia coli in pork using hyperspectral imaging-based deep learning

Traditional detection of pathogenic microorganisms is time-consuming and costly. This study proposed a rapid, nondestructive, and environmentally friendly method for detecting Escherichia coli (E.coli) in pork using hyperspectral imaging (HSI) and deep learning. First, hyperspectral images (387.06-1003.18 nm) of fresh pork inoculated with different concentrations of E. coli were collected, and spectral reflectance was extracted from the region of interest (ROI). Six preprocessing methods and five feature wavelength selection algorithms were investigated to optimize the performance of the convolutional neural network (CNN) and support vector machine (SVM) models. The CNN-SVM based on particle swarm optimization (PSO) was proposed to develop a detection method of E. coli concentrations in the fresh pork, and PSO-CNN-SVM based on a variable iterative space shrinkage approach (VISSA) performed the best, achieving 87.50% accuracy in the testing set. Finally, the spatial distribution of E. coli in pork was visualized based on the predicted model. The results confirm HSI-based deep learning could effectively detect and visualize E. coli in fresh pork, providing a technical support for food safety. • Hyperspectral imaging effectively detected and visualized Escherichia coli in pork. • PSO could effectively improve the classification accuracy of CNN-SVM. • It is helpful for Escherichia coli detection in large-scale industrial pork production.