Study On Detection Of Salmon Quality Based On Hyper-spectral And Ultrasound Imaging Technology

Salmon is known as the supreme fish because of its good taste,sweet flavor and high nutritional values.Fresh transportation which are stored at 0~4?and frozen transportation which are preserved below-18?are common methods to prolong the shelf life of salmon.Comparied the fresh transportation with the frozen transportation,the former can retain the taste and flavor to the maximum of salmon.Hence,it is a higher price.Unscrupulous traders often replace fresh salmon with frozen-thawed substitute and pursue high interest.The frozen-thawed salmon are thawed salmon which stored and transported in frozen before being sold.In order to avoid this undesirable phenomenon,it is necessary to monitor quality of fresh salmon from frozen-thawed and altered-chilled salmon.Although the conventional analytical methods are accurate for quality evaluation,those methods usually are destructive time-consuming,grueling,and not meet the market demand of rapid,break-less and real-time detection of batch samples.The paper of this study aims at the application of hyper-spectral and ultrasound imaging technique,combined with chemometrics methods,for rapid distinguish of fresh and frozen-thawed salmon fillets.Meanwhile,the hyper-spectral imaging technique is used to predict and visualize TVB-N,moisture and shear force of fresh and frozen-thawed salmon fillets in the process of storage.The main contents and results are as follows:(1)Fresh and frozen-thawed salmon samples were identified based on hyperspectral imaging technique:The study assessed the fresh salmon samples which was stored at 4?and freeze-thaw salmon samples included the different freezing at different days.Firstly,hyper-spectral system was applied to acquire the image information of salmon samples in the wavelengths of 431 nm to 963 nm and ENVI 4.5 software was used to treat the hyper-spectral image and collect the spectral information.Secondly,the different pre-processing methods included SG,SNV,MSC,VN,1ST,and 2ND were compared and Principal Component Analysis(PCA)was used to obtain the feature spectrum.At the same time,PCA was also used to analyze the hyper-spectral images and textural feature variables were obtained from principal component signals(PCs)images.Finally,the linear(LDA,KNN)and non-linear(LS-SVM,BP-ANN)models were used for identification based on the feature spectrum,image texture variables,and fused feature spectrum and texture variables,respectively.The results indicate that the LS-SVM model reached a good recognition rate for fresh and frozen-thawed salmon samples and BP-ANN model reached a good recognition rate for fresh and frozen-thawed salmon at different days.(2)The quality indicator of fresh and frozen-thawed salmon samples during storage were detection and visualization based on hyper-spectral imaging technology:Taking fresh and freeze-thawed salmon fillets on different storage days as the research object.Firstly,the chemical methods were used to test the values of TVB-N,water content,and shearing force.Meanwhile,hyper-spectral system was applied to acquire the image information of salmon samples and ENVI 4.5 software was used to treat the hyper-spectral image and collect the spectral information.The spectral information were treated with different pre-processing methods and PCA.The feature spectra were associated with their corresponding TVB-N,water content and shearing force using PLS,iPLS,BiPLS,SiPLS,LS-SVM and BP-ANN.The best prediction models for TVB-N,moisture content and shear force were selected.High performances were obtained by the LS-SVM models for TVB-N,water content,and shearing force prediction,respectively,with r_p of 0.9076,0.8037,and 0.8584,and RMSEP of 2.110 mg/100g,2.240%,and 0.4400 N.Finally,the LS-SVM models were used pixel-wise to hyper-spectral images of the testing samples to build pseudo-color images for visualizing TVB-N,moisture content and shear force for fresh and frozen-thawed salmon samples during storage.Finally,the spectral information corresponding to each pixel in the salmon hyper-spectral image is extracted one by one,and the corresponding model index value at each pixel is calculated by substituting each index's best model LS-SVM to obtain the fresh and freeze-thawed.The distribution map of the content of each index of the melted salmon on the fish during storage.(3)Fresh and frozen-thawed salmon samples were identified based on ultrasound imaging technique:Fresh and freeze-thaw salmon samples with different storage days were used as the research objects in this study.Firstly,the ultrasound system was used to collect and correct the ultrasound images of the samples.Secondly,the image texture variables were extracted from the ultrasound images of the samples using GLCM,and the feature signal of texture variables were extracted using PCA.Finally,the linear models and non-linear models based on the image texture variables were developed.The results indicate that the LS-SVM model acquired a good recognition rate with prediction set was 90.74%.The models reached lower recognition rate for fresh and frozen-thawed salmon at different days.