Non-Destructive Assessment of Microbial Spoilage of Broiler Breast Meat Using Structured Illumination Reflectance Imaging with Machine Learning

Meat quality has gained ample attention owing to increased consumer awareness and competition among poultry processors to deliver premium quality products. Nevertheless, chicken breast meat is susceptible to microbial spoilage resulting in economic and product losses. Conventional approaches such as organoleptic, aerobic plate count (APC), and molecular methods have been employed for assessing the microbiological quality of meat products but suffer various shortcomings. This study was a proof-of-concept evaluation of emerging structured illumination reflectance imaging (SIRI) as a non-destructive, objective means to evaluate microbial spoilage in chicken breast meat. The experimental chicken breast samples were kept on a retail tray for 1–13 days at 3-day intervals and subjected to image acquisition by broadband SIRI at varied spatial frequencies of sinusoidally-modulated structured illumination (0.05–0.40 cycles mm−1). The chicken samples were categorized into fresh and spoiled classes using the APC threshold of 5 log10 CFU g−1. Acquired pattern images were demodulated into amplitude component (AC) and direct component (DC) images (corresponding to uniform illumination). Three pre-trained deep learning models, including VGG16, EfficientNetB6, and ResNeXt101, were employed to extract the features from the demodulated images, followed by principal component analysis (PCA) to reduce feature redundancy. The selected PCs were used to build classification models using linear discriminant analysis (LDA) and support vector machine (SVM) separately to distinguish between fresh and spoiled samples. AC images consistently outperformed DC images in the resultant classification performance. When the LDA classifier was used, AC images yielded maximum accuracy improvements of 3.6