Non-destructive Testing Method For Descaling Effect Of Freshwater Fish Based On Hyperspectral Technique

Freshwater fish and freshwater fish products account for a large proportion in the global consumption of aquatic products.Scaling of freshwater fish is an important part of fish processing.The apparent quality and commercial value of freshwater fish products was directly affected by descaling effect.The existing detection technology of descaling effect is relatively backward,mostly artificial detection,which is one of the most important factors restricting the automation and intelligence level of freshwater fish processing and production.Hyperspectral technique,as a new nondestructive testing method,has been widely used in food quality evaluation because of its effective fusion of spectral and image information.In this paper,based on hyperspectral technology,combined with image processing and machine learning,the rapid non-destructive testing of descaling effect of freshwater fish was carried out.Because the spectral and image responses of fish back,belly and tail are quite different,the whole fish was divided into the above three parts for further analysis.According to the collected hyperspectral images under different degrees of fish descaling,for spectral data,on the basis of hyperspectral pretreatment method,spectral dimensionality reduction method and data modeling method,the detection model of fish descaling effect was established to realize nondestructive detection of freshwater fish descaling effect.For the image data,based on principal component analysis(PCA)and image processing,a fast algorithm for scale removal is proposed.The comparative analysis shows that the image processing method is the most suitable method.The image processing algorithm,spectral processing algorithm and model building algorithm were integrated to build a software system for fish descaling effect detection.The specific research is as follows:(1)The average spectra of fish with different scales were extracted.The original spectral data were corrected,and the spectral information was preprocessed by convolution smoothing(SG)algorithm,multiple scattering correction(MSC)and first derivative method.Continuous projection algorithm(SPA)and regression coefficient method(RC)were used to realize dimensionality reduction of hyperspectral data.Three feature bands(454.9 nm,573.8 nm and748.6 nm)was selected by SPA.RC selected Five feature band(410.2 nm,437.1 nm,555.3nm,572.1 nm and 597.0 nm)was selected by RC.Fish scaling removal effect detection model based on full spectrum information,characteristic wavelength information and segmented wavelength information was established respectively.The fish scaling removal effect detection model based on segmented wavelength information was obtained as the optimal detection model.For fish back,the best scaling removal effect detection model was established in the spectral range of 696.8–761.1 nm,which yielded the R_p~2 value of 0.966,R_c~2 of 0.971,RMSEC of 0.032,RMSEP of 0.033 and RPD of 1.639.For fish belly,the best scaling removal effect detection model was established in the spectral range of 696.8–761.1nm,which yielded the R_p~2 value of 0.948,R_c~2 of 0.978,RMSEC of 0.029,RMSEP of 0.044and RPD of 1.668.For fish tail,the best scaling removal effect detection model was established in the spectral range of 696.8–761.1 nm,which yielded the R_p~2 value of 0.988,R_c~2 of 0.995,RMSEC of 0.019,RMSEP of 0.029 and RPD of 1.837.The nondestructive testing method based on the segmented spectral data can realize the descaling effect of freshwater fish.(2)Based on image data,the six PC images were extracted by principal component analysis(PCA).PC3 images with prominent scale removal features on fish back and PC4images with prominent scale removal features on fish belly and fish tail were selected.Image processing methods such as threshold segmentation,regional growth method and morphological processing were used to obtain images of descaling areas of freshwater fish,and the accuracy of area calculation was obtained.The average accuracy of scaling effect of50 fishes can be as high as 91.12%.The testing method based on image data can be used to detect the scaling effect of freshwater fish.(3)In order to provide technical support for the automation and intellectualization of fish quality detection system,based on the above two nondestructive testing methods,the image processing algorithm,spectral processing algorithm and model building algorithm were integrated,and a software system for fish descaling effect detection was established.