The freshness of meat products during storage has received unprecedented attention. This study was conducted to investigate the feasibility of a hyperspectral imaging (HSI) technique to determine the freshness state of cooked beef during storage and identify the contaminated areas on the surface of spoiled samples. Hyperspectral images of cooked beef were acquired in the wavelength range of 400–1000 nm and the freshness state of all samples was divided into three classes (freshness, medium freshness, and spoilage) using the measured total viable count (TVC) of bacteria. Fifteen feature spectra variables were extracted by random frog (RF); based on this, six optimal wavelength variables were further selected by correlation analysis (CA). Partial least squares (PLS) and least squares–support vector machine (LS-SVM) classification models were established using different spectral variables. The results indicated that the performance of the RF-CA-LS-SVM classification model with a high overall classification accuracy of 97.14%, the results of sensitivity and specificity in the range of 0.92–1, and the κ coefficient of 0.9575 in the prediction set were obviously superior to other models. Spoiled samples were further obtained using a RF-CA-LS-SVM model, and then six feature images were extracted and further fused by principal component analysis (PCA). A PC3 image was used to segment successfully the contaminated areas from normal areas of cooked beef images using the Otsu threshold algorithm. The results demonstrated that HSI has great potential in classifying the freshness of cooked beef and identifying the contaminated areas. This current study provides a foundational basis for the classification and grading of meat production in further online detection.
© 2017 The Author(s)PDF Article