Nondestructive Detecting Of Prawn Freshness By Fusion Of Image And Spectral Features Information

China is the main producer,consumer and exporter of prawns,and occupies an important position in the world’s marine fisheries.In recent years,people pay more and more attention to the freshness of aquatic products.At present,domestic prawns product quality testing basically uses sensory evaluation,physical and chemical testing and microbiological testing.These testing methods are cumbersome,time-consuming,subjectively affected,and can not be fast and non-destructive.In view of the above problems,this study fused feature map information to construct a rapid detection method for the freshness quality of prawns,and compared it with the use of near infrared spectrum information and image information alone.The main test results are as follows:1.The quality changes of prawns stored at 4℃ for 12d were comprehensively evaluated to determine the freshness evaluation index applicable to prawns.The main indexes reflecting the quality change of prawn during storage at 4℃were chemical index,microbial index,texture characteristic(hardness,elasticity,adhesiveness,mastication)and color difference property;Among them,TVB-N,total amount of biological amine,b*value,elastic value and total number of bacterial colonies can be used as the freshness indexes of prawn,and TVB-N and total number of bacterial colonies can be further used as the quality indexes to determine the spoilage of prawn.2.Establish a method for quantitative analysis of prawn freshness index based on near infrared spectroscopy.This study firstly weighted fusion method for regional characteristics of 350～1000 nm and 940～1650 nm double band spectrum for effective integration,in order to get more comprehensive spectral information,and then based on the spectral information of different wavelengths,compared with the standard normal variable transformation(SNV),the first derivative(FD)and the second derivative(SD)three methods to deal with spectrum,at the same time with different methods according to the influence of different order combinations using the modeling results are analyzed,in addition,competitive adaptive weighting algorithm(CARS)was used to select the characteristic spectral variables related to freshness index from many variables,and a more robust and reliable prediction model of support vector machine(SVM)was established.The experimental results showed that,compared with the results of single-band spectral modeling,the models based on the spectral information after two-band fusion were superior to the modeling results based on single-band except for the value of parameter b*;Compared with the model based on the original spectral data,the model based on the preprocessed spectral data had better prediction effect.Among them,the prediction of TVB-N content model RPD>2.5 indicates that the prediction effect was good and the model accuracy was high,and the other index model 1.5<RPD<2.5 indicated that it was feasibled to conduct quantitative analysis of prawn freshness index by this method,but the prediction ability need to be improved.3.Establish a method for quantitative analysis of prawn freshness based on computer vision technology.The experiment took prawns stored in the refrigerator at 4℃for different days as the research object.The images of prawns on different days were obtained by CCD camera,After proper pretreatment(median filtering,binarization and morphological processing),the three channels of color R,G and B and the area,length and width,the eccentricity of the ellipse shape,the diameter of the circle of equal area and the circumference of the image were extracted.After normalization and freshness index,the SVM model was established,and each index model 1.5<RPD<2.5,indicating that these modeling methods were feasibled,but the model prediction ability need to be improved.4.Establish a method to evaluate the freshness of prawn by combining the feature information of the Image and spectral.Test by using the method of feature fusion layer extracted spectral characteristics of the information and image information fusion the SVM model was established,the index model of RPD>2.5,the results showed that the accuracy of the model and stability were improved,It was better than the model built with spectral information or image information alone and could provide reference for the application of multi-source information fusion technology in the fast quantitative detection of prawn freshness index.