Study On Potato Grading Based On Machine Vision

The potato is one of the main agricultural products in Gansu province.While potato appearance quality directly affects its competitiveness in the market, potato quality inspection and grading are conducive not only to the storage and sales, but also to its further deep-added processing. Recent years, with the rapid development of computer technology, machine vision and spectroscopy techniques have been widely used in the detection and grading of agricultural products, and many scholars have used machine vision technology for agricultural products inspection and grading. Potato sorting on machine vision system can not only reduce the secondary mechanical sorting damage and the artificial classification errors, but also improve productivity and classification accuracy. In this paper one kind of potato grading machine vision system is mainly put out and developed in the following five aspects1Acquiring potato image:In the potato grading machine vision system obtaining precise potato images is key factor for image processing. A two-camera machine vision system was designed to acquire potato images in the experimental grading system, and a specialty of Gansu-Longshu No.4was selected as potato samples.2The image preprocessing:In order to accurately and efficiently extract feature parameters from the potato images, and to improve the detection accuracy and speed, a serious of operations, such as image smoothing, image enhancement, threshold segmentation and mathematical morphology algorithms, were used in potato image pretreatment.3Mass sorting:Geometric characteristics is the main factors to quality of the image. In this study potato image area and perimeter of two characteristic parameters were extracted to detect its quality. After extracting parameters of top view area and side view perimeter, a potato mass grading model was constructed with stepwise regression analysis to achieve the quality of the potato sorting. The experimental results showed that precision ratio of potato mass grading was95.3%.4Shape sorting:The shape characteristics can be described by squareness, roundness, eccentricity and invariant moments. In this paper image invariant moments were used to the shape classification of potatoes. Six invariant moments of vertical view image were input the trained BP neural network, which was trained by the sample sets of potato images, to carry out shape sorting. The experimental results showed that the accuracy of potato shape grading was96%.5Surface defects sorting:In this paper, an RGB color characteristic parameters combined with BP network of potato defect detection and identification methods was proposed. First rough set K-means clustering algorithm was used for potato defect segmentation, and divided potato into normal and defective based on the area threshold determination. Then6color characteristic optimizing parameters were extracted from the region of the potato surface defects. Finally using BP network and the characteristic parameters, a type of defect recognition model was established. The experimental results showed that the accuracy of potato defe grading was94.8%.