Research On Grape Variety Classification And External Quality Classification Based On Computer Vision

Grapes are rich in nutrients and are favored by consumers.In modern grape orchards,most production processes have reached the level of mechanization and automation,but grape variety identification and quality grading are still carried out manually.Conventional manual inspection methods are time-consuming,laborious,costly and subjective,In the field of fruit and vegetable product classification,single round or oval fruits are mainly used as the main inspection targets(such as oranges,apples,cherries),and there are few relevant studies on fruits with ear shape,fruit grain accumulation and serious adhesion(such as grapes,longan,litchi).Focusing on the industrial requirements of grape variety classification and external quality classification,this study focuses on the algorithm models of image recognition,image semantic segmentation and quality classification in machine vision,and builds an app application for grape variety classification and quality classification on the mobile terminal to realize automatic optimization,classification and classification of grapes.The main work of this study is as follows:(1)The design and construction includes the original grape image acquisition module,the original image data preprocessing module,the image semantic segmentation design module,the image feature extraction and analysis module,the grape variety classification and external quality grading design module,and the grape variety classification and external quality grading application system design framework based on complex scenes.(2)A sufficient number of original grape pictures are obtained by combining on-the-spot manual photography and web focused crawler technology.The labelme tool is used to label the sample set of grape pictures,which provides a data basis for the batch conversion of the JSON file of grape labels into the grape labels required for subsequent training models.(3)Through the quadratic wavelet decomposition and adaptive median filter,the original grape image is scaled and denoised to improve the image quality,improve the data availability to a certain extent,facilitate semantic background segmentation,and improve the image segmentation effect.(4)Three methods of traditional semantic segmentation Otsu,deep learning UNET network direct training model and vgg16 pre training model are used for image semantic segmentation experiments.The comparison results show that vgg16 is significantly better than the other two methods.Vgg16 is used for image segmentation,which effectively improves the accuracy of grape image feature extraction and analysis.(5)The traditional opencv method was used to extract the grape features such as texture,color,area,defect area and fruit number;The mobilenet pre training model under tensorflow framework is constructed to realize the multi classification of grape varieties.At the same time,the decision fusion of judgment tree classification mode and SVM classification mode is used to complete the classification of grape quality.(6)The app for grape variety identification and external quality grading is developed with Android.By encapsulating the entire functional modules mentioned above,the user can operate the entire online grading process through the mobile app and obtain the specific index information value of identification and grading in real time.