Cherry Tomato Online Grading Based On Machine Vision Detection Research

Cherry tomato,also known as the little tomato,can be used as both a vegetable and a fruit,and is favored by consumers.The traditional appearance quality inspection mainly relies on manual classification or machine grading.When manual inspection,human measurement and grading will be misclassified due to fatigue and personal discrimination methods;machine grading is to use a special machine to detect the size of the fruit,but not the texture,surface defects and color of the fruit.in order to solve this problem,this thesis designs a cherry tomato online grading detection system based on machine vision,which extracts several important features of cherry tomato by using the collected images to establish a fast and efficient hierarchical detection model is used to achieve automatic preferred grading of fruits.Firstly,based on the research on the basic theories and methods of image processing,the thesis analyzes the collected cherry tomato images,combines binary images and color images,and uses image local threshold segmentation,highlight extraction,morphological processing,and external Contour extraction,local property calculation,etc.,select the best image to remove the background,count the number of connected areas of the image N,use the size of the R,G,B values in the identification index and the parameter value metric as the main basis for discrimination,Determining whether the cherry tomato is a good fruit and completing the first classification.Secondly,in order to improve the real-time processing efficiency and quality of images,the image is scaled and denoised by the secondary wavelet decomposition and median filtering.The Canny algorithm is used for edge detection,and the feature extraction parameters selecte fruit width,color,defect area,fruit shape,texture and maturity,among which,the classification of fruit diameter,defect area and color is graded by discriminant tree grading method;the SVM hierarchical model based on particle swarm optimization classifies complex high-dimensional features such as fruit shape,texture and maturity.In the training phase of SVM model,KPCA is used to reduce dimensionality of high-dimensional features.When KPCA is used for training,in order to avoid blind selection of parameters,introducing particle swarm optimization algorithm to optimize selection parameters,and then the final sample level is determined by decision fusion,and the second classification is completed.According to the algorithm proposed in this thesis,in the cherry tomato online identification system based on Microsoft Visual Studio 2010 and OpenCV2.4.9 environment,through the collection of cherry tomato image and homemade image database to complete the test verification,the experimental results show that based on single classifier In the recognition result,the discriminant tree classification accuracy rate is higher than SVM,and the KPCA dimension reduction feature for SVM classification is improved compared with the single SVM classifier in the recognition accuracy rate and the classification rate,the decision fusion and single classifier Compared,the recognition accuracy of cherry tomato was significantly improved,the classification accuracy rate was over 95%,and the average classification rate was 4/s,which not only improved the recognition rate,but also met the real-time performance.