基于图像光谱信息融合的鱼不同冻藏时间及冻融次数鉴别

应用高光谱成像技术(380~1023 nm),基于信息融合实现鱼不同冻藏时间后冻融次数鉴别。首先,提取鱼样品感兴趣区域(region of interest,ROI)光谱并结合竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)得到57个变量作为光谱信息,同时对鱼样品做主成分分析(principal component analysis,PCA),提取第一主成分图像信息如中值、协方差、同质性、能量、对比度、相关、熵、逆差距、反差、差异性、二阶距和自相关12个灰度共生矩阵(gray level co-occurrence matrix,GLCM)纹理特征参量,结合灰度共生矩阵纹理特征与光谱特征,作为模型偏小最二乘支持向量机(least squares support vector machines,LS-SVM)的输入建立区分模型,预测集识别率达到98%。结果表明,高光谱成像技术可以用于鱼不同冷冻时间以及冻融次数的鉴别。

Classification of fresh and frozen-thawed fish fillets based on information fusion of image and spectrum

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.