Nondestructive Testing Of Pretreatment Quality For Freshwater Fish Based On Hyperspectral Technique

The pre-treatment process of freshwater fish descaling is an important part of fish processing.The surface damage during descaling and the cleaning effect of blood in the pre-treatment process directly affect the appearance quality and commercial value of freshwater fish.At present,most of the pre-processing quality detection of freshwater fish is done by human,which seriously restricts the automation and intelligence level of fish deep processing.Hyperspectral imaging technology is a new comprehensive imaging technology,which effectively integrates the spectrum and image information,and combines the traditional spectroscopy and imaging technology.It has been widely used in the process of food quality evaluation.In this paper,based on hyperspectral imaging technology,the rapid detection of freshwater fish pretreatment quality is studied.Combined with image processing and machine learning technology,a fast and objective nondestructive testing method for freshwater fish pre-processing quality is proposed.Mechanical and water jet descaling damage and blood residue in pretreatment process are the detection targets.According to the spectral response characteristics and image characteristics of fish surface in the process of processing,the methods of combining decision tree with principal component image pixel value and synthetic image are proposed respectively to detect and recognize the damaged area and the residual area of blood pollution in freshwater fish.The specific research is as follows:(1)Non destructive testing of mechanical descaling damage and water jet descaling damage is carried out respectively.A method of combining decision tree and principal component image is proposed to reduce the dimension of data by combining the principal component and pixel value of image.A recognition model of descaling damage(mechanical descaling damage and water jet descaling damage)based on the combination of decision tree and principal component image pixel value is established.Only through the key pixel points,the visualization of descaling damage area is completed,and mechanical and water jet descaling damage is realized Identification of injury area.At the same time,according to the difference of image characteristics between the damaged area and the normal area,a synthetic image method is proposed to construct the pseudo color synthetic image corresponding to the characteristic band.Through image processing algorithms such as threshold segmentation,binary change,small area removal and logical operation,a visual method for the damaged area of the pseudo color synthetic image of mechanical descaling is proposed to realize the recognition of the damaged area of mechanical descaling In this paper,super-pixel segmentation method is used to extract the texture features of water jet descaling damage image,and a water jet descaling damage recognition model based on decision tree is established to realize the recognition of water jet descaling damage.By comparing the recognition accuracy of the above methods,it is determined that the combination of decision tree and principal component image pixel value is the best method.The recognition accuracy of mechanical and water jet descaling damage can reach 94.1% and 89.4% respectively.(2)On the basis of the above research,the method of combining decision tree with principal component image pixel value and synthetic image is used to identify and detect the residual area of blood pollution in the process of freshwater fish preprocessing.Finally,the method of combining decision tree with principal component image pixel value is determined to have a better detection effect with an accuracy of 91.3%.(3)Based on the above two NDT methods,the image processing algorithm,spectral processing algorithm and model building algorithm are integrated to build a software system for freshwater fish pre-processing effect detection.