Study On The Appearance Quality Of Freshwater Fish Based On Visible/near Infrared Image

Aquatic food is the third largest source of human food protein.The high-quality dietary protein contained in aquatic products has become an important strategic resource to ensure the national food safety,and has a significant impact on the people's diet structure and food health.At present,the level of aquatic products processing technology in China is still not perfect compared with developed countries.Some production lines need to rely on manpower to affect the efficiency and stability of the production line.Besides improving productivity,food safety is also an important key to people's livelihood.The characteristics of hyperspectral image technology are that it contains spectral and image information of the target,and can accurately,lossless and mass quality appraisal of the target.This paper focuses on the shape and quality detection of freshwater fish,and studies the two parts of the image,visible light and hyperspectral.In terms of appearance detection,according to the needs of large freshwater fish processing line,this paper proposes a method of freshwater fish contour detection on distorted stainless steel chain plate by distortion correction,background difference and ellipse fitting.By calculating the projection matrix of the corresponding pixels of the experimental image and the template image,the distorted experimental image is corrected to the world coordinate system;Then the region of interest of the corrected image and the template image is matched and the difference is made to obtain the target image and contour points separated from the background;Finally,the ellipse is circularly fitted to ensure that the output ellipse just covers the main part of the fish body as much as possible,and the direction of the long axis is consistent with the direction of the fish body.Compared with manual measurement and other methods,this method has better accuracy,efficiency and stability.For the quality detection part,the way to store fresh water fish is to ensure that it is in the living state.Frozen crucian carp can only be stored for three to five days at most.In order to ensure the fresh fresh freshwater fish in the processing line,the hyperspectral images of the target within 7 days were collected by specim fx10 hyperspectral camera.The regions of interest were selected,and the spectra were preprocessed by multiple scattering correction(MSc),first derivative,standard normal variable transformation(SNV)and detrended method,and then the characteristic bands were extracted according to lasso model,For training samples,support vector machine(SVM)is used to build a clustering model for the spectral values of all training samples in the characteristic band,and then the accuracy of the model is calculated by the test samples.The accuracy of derivative lasso SNV model is 94.8%,which fully meets the qualitative requirements of industrial production detection.It can eliminate the samples that do not meet the requirements of food safety before they enter the processing line.This method can take into account both the algorithm efficiency and accuracy,and improve the efficiency of the production line while ensuring the quality of raw materials,It has strong stability and good practical prospect.