Classification of fresh and frozen-thawed fish fillets based on information fusion of image and spectrum
Zhang Hailiang,Zhu Fengle,Liu Xuemei,He Yong
DOI: https://doi.org/10.3969/j.issn.1002-6819.2014.06.033
2014-01-01
Abstract:Salmon has always been regarded as a popular gourmet fish that is consumed in large quantities due to its high nutritional value. This study proposes a new rapid and non-destructive method using visible and near infrared (Vis/NIR) hyperspectral imaging for the detection of freshness, storage time, and frozen-thawed times of fillets for turbot flesh. Hyperspectral imaging technology is a rapid, non-destructive, and non-contact technique that integrates spectroscopy and digital imaging to simultaneously obtain spectral and spatial information. Hyperspectral images are made up of hundreds of contiguous wavebands for each spatial position of a sample studied, and each pixel in an image contains the spectrum for that specific position. With hyperspectral imaging, a spectrum for each pixel can be obtained and a gray scale image for each narrow band can be acquired, thereby enabling this system to reflect componential and constructional characteristics, as well as the spatial distributions, of an object. In this study, a hyperspectral imaging system (380-1 023 nm) was developed to perform classification of freshness, storage time, and frozen-thawed times of fish fillets based on a gray level co-occurrence matrix (GLCM) and least squares support vector machines (LS-SVM). Altogether, 160 fish samples from two different storage days and two different frozen-thawed times were collected for hyperspectral image scanning, and mean spectra were extracted from the region of interest (ROI) inside each image. LS-SVM was applied as a calibration method to correlate the spectral and GLCM data for 110 samples in the calibration set. The LS-SVM model was then used to predict the freshness, storage time, and frozen-thawed times for the 50 prediction samples. Spectra of fish samples were extracted from the region of interest (ROI) and a competitive adaptive reweighted sampling (CARS) algorithm was used to select the key variables. Hyperspectral imaging data and principal component analysis (PCA) were performed with the goal of selecting the first principal component (PC) image that could potentially be used for the classification system. Then, 12 texture features (i.e., mean, standard deviation, smoothness, third moment, uniformity, and entropy) based on the statistical moment were extracted from the PC1 image. Finally, 12 gray level co-occurrence matrix (GLCM) variables, combined with 57 characteristic wavelengths for each fish sample, were extracted as the LS-SVM input. Experimental results show that the discriminating rate is 98% in the prediction set. The results indicate that hyperspectral imaging technology combined with chemometrics and image processing allows the classification of freshness, storage time and frozen-thawed times for fish fillets, which builds a foundation for the automatic processing of aquatic products. The fish industry can benefit from adopting hyperspectral imaging technology.