三文鱼是一种营养丰富且味道鲜美的海水鱼种, 近年来, 我国三文鱼消费市场需求旺盛, 进口量不断增加, 而进口方式主要包括冰鲜和冷冻两种。 相比于冷冻三文鱼, 冰鲜三文鱼能更好的保留其优良品质, 但同时成本更高, 售价更贵。 因此存在部分不法商贩将冷冻三文鱼解冻后作为冰鲜三文鱼售卖, 以此谋取更多利润。 这种欺诈行为不仅严重损害了消费者的利益, 也不利于我国三文鱼消费市场的健康发展。 为建立一种快速、 无损的三文鱼品质检测方法, 以冰鲜和冻融三文鱼为研究对象, 对冰鲜和冻融三文鱼的高光谱光谱差异和图像差异进行了分析, 并结合化学计量学方法对冰鲜和冻融三文鱼进行快速鉴别。 三文鱼在冷冻运输过程中, 受冷链条件等因素的影响, 可能存在多次冻融的情况。 因此为提高检测方法的通用性, 制备不同冻融次数的三文鱼作为冻融组。 首先通过高光谱成像系统采集样本的高光谱图像数据。 然后利用ENVI 4.5软件提取样本高光谱图像中感兴趣区域(ROI)的平均光谱, 同时利用灰度共生矩阵法(GLCM)对前三个主成分图像的纹理信息进行提取。 原始光谱信息经过多元散射校正(MSC)等方法预处理后, 利用主成分分析法(PCA)、 竞争性自适应重加权算法(CARS)、 连续投影算法 (SPA)和CARS-SPA对光谱进行降维和变量筛选。 最后基于光谱信息、 图像信息以及融合光谱-图像信息分别结合反向传播神经网络(BPANN)、 线性判别分析(LDA)、 极限学习机(ELM)和随机森林(RF)建立冰鲜与冻融三文鱼鉴别模型。 结果显示基于MSC预处理光谱的CARS-ELM模型对冰鲜与冻融三文鱼识别效果最佳, 其校正集和预测集的识别率分别为100.00%和95.00%。 此外, 在对三文鱼的冻融次数鉴别研究中, 基于MSC预处理光谱建立的CARS-ELM模型对三文鱼冻融次数识别效果最佳, 其校正集和预测集的识别率分别为97.50%和91.67%。 研究结果表明, 基于高光谱成像技术能够对冰鲜与冻融三文鱼进行快速鉴别。

Salmon is a kind of marine fish with rich nutrition and delicious taste. In recent years, the consumption market of salmon in China wasin great demand, and the import volume of salmon was increasing. The import methods mainly included chilled salmon and frozen salmon. Compared with frozen salmon, chilled salmon can retain the excellent quality of salmon to a greater extent, but at the same time, it cost more and was more expensive. Therefore, some illegal traders sold frozen-thawed salmon as chilled salmon in order to make more profits. This kind of fraud not only seriously damaged the interests of consumers but also go against the development of salmon market inChina. In order to establish a fast and non-destructive method to detect the quality of salmon, this study took the chilled and frozen-thawed salmon as the research object, used hyperspectral imaging technology to analyze the spectral difference and image texture difference between the chilled and frozen-thawed salmon, and combined the chemometrics method -to identify the chilled and frozen-thawed salmon quickly. In the process of frozen transportation, salmon may be frozen and thawed for many times due to the cold chain conditions and other factors. Therefore, in order to improve the universality of the detection method, salmon with different frozen-thawed times were set as the frozen-thawed group in this study. Firstly, the hyperspectral image data of the samples were collected by the hyperspectral imaging system. Then, ENVI 4.5 software was used to extract the average spectrum of the region of interest in the sample’s hyperspectral image, and the texture information of the first three principal component images was extracted by using the Grey-level co-occurrence matrix(GLCM).The original spectrum was firstly pretreated by multiple scattering correction(MSC), then principal component analysis(PCA), competitive adaptive reweighting algorithm(CARS),successive projections algorithm (SPA) and CARS-SPA were used to reduce the dimension and wavelength of the spectrum. Finally, based on spectral information, image information and fusion spectroscopy-image information, the identification model of chilled and frozen-thawed salmon were established by combining Back-propagation neural network(BPANN), Linear discriminant analysis(LDA), Ultimate learning machine(ELM) and Random forest(RF).The results showed that the CARS-ELM model combined with the MSC preprocessing spectrum had the best recognition effect on the chilled and frozen-thawed salmon, the recognition rates of the calibration set and prediction set were 100.00% and 95.00%, respectively. In addition, the CARS-ELM model based on the preprocessing spectrum of MSC had the best effect on the identification of the times of frozen and thawed of salmon, the recognition rates of the calibration set and prediction set were 97.50% and 91.67%, respectively. And the fast identification of chilled and frozen-thawed salmon based on hyperspectral imaging technology was realized.