Vis-NIR hyperspectral imaging for the classification of bacterial foodborne pathogens based on pixel-wise analysis and a novel CARS-PSO-SVM model

Abstract

The objective of this study was to evaluate the advantages of meta-heuristic optimization algorithms as well as optimal wavelength selection techniques for building rigorous classification models for the classification and differentiation of bacterial foodborne pathogens grown on agar plates. The results obtained in this study provide evidence that pixel-wise extraction technique improves greatly classification accuracies. The application of Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) classification models provided classification accuracies of (91.7%, 83.9%) and (90.7%, 82.2%) for training and prediction sets respectively. In addition, the c and g SVM parameters were optimized based on the genetic algorithm (GA), grid search (GS) algorithm and particle swarm optimization (PSO) to improve classification of the spectral data, which enhanced the classification accuracy of the samples to GA (95.88%, 94.53%), GS (94.27%, 93.75%) and PSO (100%, 98.44%) for training and prediction sets respectively. Furthermore, the selected bacterial foodborne pathogens were classified via reducing the number of wavelengths by applying optimal wavelength selection techniques like CARS (Competitive Adaptive Weighted sampling), ACO (ant colony optimisation), SI (synergy interval), GA and SI-GA algorithms. CARS-PSO-SVM achieved the highest classification accuracy of 99.47% and 98.44% for training and prediction respectively. The results obtained from this study indicate that foodborne pathogens grown on agar plates could be recognized and differentiated accurately and effectively by the hyperspectral imaging (HSI) technique combined with chemometric techniques.