Study On Potato Automatic Detection And Grading Based On Machine Vision

With the development of China's rural economy, to improve our fruits and vegetables competitiveness power in the international market, the potato grading has become the theory and techniques problem solved urgently in the agricultural production and deep processing of agricultural production. Experimental study of potato grading theory and technology, not only can reduce the labor intensity of manual sorting and secondary damage, and increase potato's added value as well as quality, but also improve resource utilization for further processing. In this paper, serial potato grading tests were mainly studied with the machine vision from the aspects of image acquisition and preprocessing, shape detection, quality inspection, surface defect inspection, etc, as fellows.1. Image acquisition and preprocessing are basic image processing, and directly affect the image quality, thereby have the direct function for the grading performance and detection accuracy. A white background lighting system was designed, and B-channel grayscale, median filtering, bimodal method and mathematical morphology method were used for potato original image preprocessing.2. In shape detection, the shape parameters of R and C based on the equivalent ellipse of region were proposed, and using the improved BP neural network to classify the potato into ellipsoid, round and deformed. The grading results are consistent with those of artificial classification highly; its precision is up to 94.7% and can meet the requirements of practical application.3. In weight detection, in order to raise the image processing system speed and potato quality inspection accuracy, the MCU and weighing transducer as the core elements, the potato quality inspection system were manufactured. The weight detected was sending to PC through serial communication, and then graded by grading software.4. In surface defect detection, the potato dark-skin segmentation method was presented based on the standard deviation of the R, G, B and the potato green-skin segmentation method was proposed based on the Euclidean distance of the R, G, B. In this paper, with the above two methods for potato external defects detection and segmentation, the external defects were detected and judged by calculating the ratio of the defect area and the entire region. The experiments showed that the defective potatoes and no defects detection accuracy of potato are up to 94% and 90% respectively.