Research On Potato Shape And External Defects Based On Machine Vision

With the start of potato staple food strategy in China,related industries have been developed rapidly.The quality of potatoes affects the economic benefit of deep-processing industries directly,the grading of which is the basis of industrial production.At present,mechanical grading is easy to cause secondary damage to potatoes and can only detect the single feature.While the manual grading is inefficient and high-cost.Therefore,it is significant to study the algorithm for grading potato quality fast and without prejudice.This paper not only studies the shape classification and external defect detection of potatoes,but also realizes the primary verification on the hardware implementation.The main research content and results are shown as follows:1.Design an acquisition system for potato images based on machine vision,including determining the internal layout of the light chamber,calibrating the CCD camera and eliminating the distortion effect on the collected images.2.Compare the implementation effects of several algorithms in the image preprocessing and select the best method.3.Study the application of moment invariants on detection of potato shape.Optimize relative parameters of the kernel function of the support vector machine by using particle swarm optimization,which improves the accuracy of classification.Classify the potato shape using the modified boundary moment invariants based on the region moment invariants and introduce several new moment invariants to form the feature vectors.The experimental results demonstrate that the boundary moment invariants have invariance.The time of calculating is less than the region moment invariants.Finally the way of using seven boundary moments,two introduced boundary moments and computing partial absolute value of the boundary moments improves the accuracy of classification.This paper classifies potatoes into two categories,namely malformation and non-malformation and the average accuracy is 95%.4.Combining the color features of the external defections of potatoes,a method for detecting external defections,such as dry rot,holes,budding and mechanical injury is proposed employing the HSI color model.It is lightness intercept threshold segmentation algorithm.Green spots of potatoes are detected using the RGB color model.Finally,a method is proposed to detect multiple defections simultaneously by combining two algorithms.These regions are marked using rectangular frames.5.Study the realization of the method for shapes and external defection detection based on machine vision on Zedboard platform.Build the hardware and software environment in SoC chip,which is constituted by PS and PL.Realize the potato quality detection by combining the Linux system,Qt frame and OpenCV visual database.The experiments indicate that the average time of detecting a 640×480 image is 4s.