Research On Automatic Wine Picking Method Based On Image Classification Algorithm

The liquor industry has been continuously working on the mechanized production for a decade,and has carried out some mechanized transformations in the stalling and other aspects.However,it is difficult to achieve automated production due to the key links relying on manual judgment such as captain and wine picking,which makes the whole liquor mechanization industry fail to form an automated whole.Therefore,the liquor industry as a whole is still at a semi-mechanized level.Among them,the wine picking process is still based on the traditional techniques of picking according to the shape of the hop.During the steaming process of liquor,the composition of the liquor changes drastically with the passage of time.The liquor is impacted from the cooler through the liquor tube into the wine container to form hops of different sizes and quantities.According to the shape of the hop,the liquor is divided into different quality base wines for separate storage.This method relies too much on the experience of the wine pickers,their level and production status,which makes the production difficult to increase,labor intensive,product quality unstable and potential safety risks.The major wineries are also actively researching automated wine picking to solve this problem,and there is currently no effective automated classification method for hops.In view of the above problems,thesis uses the image acquisition equipment to collect the wine picking process,and combines the computer vision and other related technologies to process,classify and automatically segment the collected hop images,and design an automatic wine picking system.Thereby avoiding manual operation and improving the stability of the quality of the wine.The main work and research results of thesis are as follows:1)By analyzing the artificial classification method of hops,the characteristics of hops caused by liquor components and the image characteristics of liquor,the feasibility of seeing flowers is preliminarily proved.This paper takes the wine picking link in the flavor liquor production process of a liquor factory in Sichuan as the research object,and proposes an automatic scheme of flower picking wine based on image processing and convolutional neural network.2)For the interface of liquor and container will continue to fluctuate,the wine is transparent and the boundary is difficult to extract and other factors.A method for extracting wine foreground based on rapid elliptical segmentation surface detection for automatic wine extraction is proposed.The method can effectively extract the foreground of the wine liquor in many wine-picking production lines and previous wine picking,reduce the interference of background information and reduce the storatge amount and calculation amount of the data.3)In order to accurately classify the hop image and ensure the generalization of hop detection,a random image block cropping method based on hop characteristics was proposed,and the Mini-Inception classification network for liquor was designed and trained with reference to Inception network.The hop image was tested and achieved a classification accuracy of 97.9%.4)Based on the research on the processing and classification of wine extraction images,a set of autonmatic system for wine picking was designed.The indicators for measuring system performance were proposed.The applications of image acquisition,foreground extraction and classification network training were developed.program.The representative sample image stream is used for simulation test.The averalge time error of the wine segmentation is 1.02s.The experimental results show that the system has higher real-time and classification accuracy.