Research On Positioning Technology Of Fish Head And Tail Based On Machine Vision

The freshwater fish which accounts for half of aquatic products in China,and its pretreatment technology is one of the major factors affecting the modernization of aquatic products processing industry in China.The head and tail removal process is the main link in the preprocessing of freshwater fish.It can realize the separation of fish head and tail from fish body,reduce the waste rate of fish body,and increase the benefit of ubsequent processing of freshwater fish.Combined with the current situation that fish heads and tails are mainly processed manually in China,this study takes silver carp as the research object,uses Matlab combined with image processing algorithm and Fasterr CNN target detection technology as the basis to realize the accurate positioning of fish heads and tails,and builds an intelligent positioning system for freshwater fish heads and tails.The system provides the key technology for the development of different sizes of automatic fish head and tail processing equipment for freshwater fish,and also has far-reaching significance for promoting the modernization of fishery processing.The main contents and conclusions of research are as follows:(1)The development trend of freshwater fish in the pretreatment processing equipment in China and overseas,and the application present situation has carried on the research and analysis,combining the advantages of machine vision technology,study the overall design scheme,and put forward can be applied to go to the head of the fish head and tail device on the tail positioning technology,to improve the freshwater fish processing equipment provides the technical support of automation,intelligent degree.(2)Set up an image acquisition system.The image acquisition system is mainly divided into two parts,hardware and software.The selection of the hardware part is considered from the needs of the actual image acquisition and the range of the body length and width of the freshwater fish,mainly for the selection of industrial cameras,lenses,light sources,computers,etc.Through the comparison of various parameters,we finally decide to choose Hai Kang robot MV-CA050-20 GM industrial array camera,MVL-KF1228M-12 MP lens,MV-LBES-300-300-W open surface light source and the desktop computer whose processor is Intelcore I9-10900 KCPU.In the software part,MATLAB is selected,and the application toolbox with rich functions of MATLAB is used to complete the transfer learning of Alex Net and Faster RCNN models and the construction of the positioning system.(3)Proposed the image segmentation technology in accordance with color and shape and template matching technology.In this text,after preprocessing the image of freshwater fish,such as grayscale,denoising and morphology,several typical image processing methods were studied and compared.color and shape features are the significant features that distinguish the fish tail from other parts.In this paper,two common image processing methods,image segmentation and template(feature)matching,are used to extract fish tail images.The shortest time of image segmentation is up to 16.57 s,and the shortest time of template(feature)matching is up to 2.97 s.(4)The deep learning based fish head and tail detection technology is studied.In this artical,the Faster RCNN network model is used.In order to prevent the overfitting of small samples,data enhancement and transfer learning methods are adopted.The classical convolutional neural network Alex Net,RCNN series target detection technology and transfer learning are introduced,and the principle of Faster RCNN algorithm is analyzed in detail.In this paper,Alex Net,transfer learning and Faster RCNN are combined and applied to the head and tail location detection technology of freshwater fish.A small sample of data was created and randomly divided into training set,verification set and test set in a 6:2:2 ratio.The optimal parameters of the training model were determined by adjusting the iteration period.The detection results of the test set with 280 images were as follows: the correct rate of fish head detection model was 96.42%,and the average detection time of each image was 0.63 s;the correct rate of fish tail detection model was95.7%,and the average detection time of each image was 0.7s.(5)Using the App Design toolbox in Matlab,the intelligent positioning system of the head and tail of freshwater fish is designed.Analyzed the requirements of the whole system,designed the main functions of each module,the mode of operation.In addition,this paper adopts the method of image calibration to transform the plane pixel coordinates obtained after image processing into actual three-dimensional coordinates.The system can not only clearly see whether the fish head and tail are accurately positioned by using a rectangular frame,but also visually display the cutting point coordinates of the fish head and tail.The interface is simple and easy to operate.The coordinate of fish head and fish tail cutting point output by the system can make the fish head and fish tail removing device automatically complete the operation of fish head and fish tail removing device automatically complete the operetion of fish head and tail removing the cutting position information,which provides technical support for automatic fish head and fish tail removing,and also lays a theoretical foundation for the automatic and intelligent of freshwater fish pretreatment and processing equipment.