Research On Freshwater Fish Variety Identification And Weight Prediction Based On Machine Vision

The freshwater fish is very delicious and eutrophy, which is one of the most favorite food in people's life. But it also has perishable, seasonal, regional and concentration etc. characteristics as a special food resource. Therefore, in order to improve the economic value of freshwater products, it need to sank deep processing of the freshwater fish products. But before the deep processing, it is necessary to classify the freshwater fish by the breed and weight, in our country most of the classification work were done by people. The work environment is very bad, and the labor intensity is big, but the efficiency is low. Some classification work was done by machine which has a lot of damage to the fish, it's unfavorable for the subsequent process, and restricted the development of the freshwater fish industry. In view of this problem, this paper selects the usual freshwater fish as the objects for study, researches the fish surface color,physical characteristic and classification method by using machine vision technology,digital image processing technique and pattern recognition theory, and will provide the theoretical basis and technical base for the realization of freshwater fish and weight grading in automatic classification.The main content and conclusions of this paper were as follows:(1) The machine vision system applicable for freshwater fish identification was established. Images of four common freshwater fish samples were collected by using the machine vision system, it is a total of 240 images,180 are take as modeling sets, the rest are take as test sets..(2) The freshwater fish original images which were set as modeling sets were pretreated, such as image graying, binary conversion, image enhancement and contour extraction. The binary image and contour image will be used to get the projected area.(3) The variety identification model of the freshwater fish will be build by the RGB color components and the physical characteristic. Then using the 60 images in the test sets to test the model, the rate of recognition is 96.67%. Meanwhile found out the main reasons cause misjudgment by analyze the misjudgment sample.(4) The physical characteristic of the freshwater fish was measured by the experiments. Each part proportion distribution of the fish will be found out. Meanwhile the fish will be divided to three parts, the head, the soma and the tail, the weight of each part will be measured, each part of the proportion of the total weight will be calculated.(5) By experiment, the proportional relation with the weight of the head, the soma and the tail were found out, which were used to correct the projected area. Finally the prediction model can be extracted by the regression analysis. Then using the 60 images in the test sets to test each model, it was found that the relative error was very small, the relative error of the chub is 3.57%, the relative error of the bream is 3.56%, the relative error of the crucian is 3.41%, the relative error of the carp is 3.49%. Result shows that the computer vision can be used to grade the freshwater fish.(6) A software system based on Visual C++6.0 for freshwater fish variety identification and weight prediction was developed. The functions of the software system included image pretreatment, feature extraction, freshwater fish variety identification and weight prediction etc. Test results showed that the software system was operational simple and stable operation.