Research On Potato Quality Classification System Based On Machine Vision

In recent years,machine vision is widely used in the detection of agricultural products with the rapid development of machine vision technology.China is a big potato producer,and potato quality testing is mainly dependent on artificial grading,which has low efficiency and high cost.So fast,real-time and non-destructive testing of potatoes has important significance and scientific value.Combined with machine vision technology,this paper studies potato mass detection,potato shape classification and external defect recognition algorithm,and designs a potato detection system based on embedded development platform.The main research details are as follows:1.The potato image acquisition system is designed and established.The light chamber layout is determined through the selection of camera,lens and light source.The camera is calibrated by Zhang's calibration method to eliminate the effect of distortion on potato image.2.The algorithms of image greying,image smoothing,image segmentation and image edge detection are compared,and the most suitable potato image preprocessing algorithm is selected.3.In the potato mass detection,through the comparison area method,perimeter method and multi-factor model method,the three-area method is finally used to fit the potato mass model.The experiment shows that the model has a high correlation,with an average detection accuracy of 97.33%.4.In the detection of potato shape classification,three new invariant moments are introduced based on the traditional Hu moment invariant seven eigenvalues,which are used to extract the characteristic parameters of potato.In order to improve the efficiency of potato detection,an improved edge invariant moment is proposed based on shape invariant moment.Chaotic particle swarm optimization(CPSO)is used to optimize the parameters of the kernel function of support vector machine(SVM).Taking the accuracy of potato classification as the standard,seven traditional boundary moments and one introduced boundary moment are determined as the characteristic parameters of potato classification,and the accuracy of potato shape classification is 96%.5.In the detection of potato external defects,the green skin of potato surface is detected by the values of R,G and B components.By using the brightness interception method in HSV color space and limiting the threshold value,the defects of potato surface such as mechanical damage,sprouting and dry rot are detected,and finally the classification of potato surface defects is realized.6.Potato detection system is designed based on the embedded Zybo Z7 development platform.Using the accelerator design of FPGA,the potato detection algorithm is encapsulated into IP core,Linux operating system is transplanted in ARM processor,and the potato mass is tested online through the Qt interactive interface design and the machine vision library.