Research On The Classification Technology Of Citrus Based On Machine Vision

Citrus is widely planted in the south of China. The classifications of citrus have depended on manpower or machines for a long time, which have no promising prospect due to the high labour intensity and low efficiency. The machine vision technology possessed many excellent properties, such as high detection precision, huge data gathering and nondestructive testing, it has become a focus for the automatic detection and classification of agricultural products. In this paper, based on the feature extraction of full surface image of citruses, the fuzzy theory is successfully applied to the citrus classification processing rely on the pre theoretical study and experimental analysis.In this study, by analyzing the principles of the machine vision system, i bui lt a visual detection system for full surface image of citrus, which is composed of a proper light source, a industrial camera, some lens, and two plane mirrors. In image preprocessing, the respective components of citrus image was analyzed in RGB color space and HSI color space, respectively. The H-component image was picked up for background segmentation, the error of segmentation will be eliminated through the morphology method and the area comparing method, then, the complete image of citrus can be obtained.By calculating the related parameters including the minimum enclosing rectangle, circularity, and yellow green ratio, the feature of citrus, such as transverse diameters, shape, and maturity was extracted. According to the illumination-reflection model, the illumination complement of original image was obtained using the low pass filtering method. After original image divided by illumination image, the corrected image was obtained. Based on previous results, the ulcer and scab of citrus can be success fully detected by using a threshold value. In the process of classification, firstly, the weight of transverse diameter, shape, and maturity in the fuzzy set must be determined rely on the survey, secondly, based on the prior knowledge and existed date, th e membership function can be determined, and using the weighted mean method, the fuzzy comprehensive evaluation can be computed, finally, the citrus grading can be easily obtained on the base of maximum subordination principle.The experimental results show that this method is effective and reliable, and the classification accuracy for the high-class, the first-class, the second-class and the substandard citrus is 100%, 88.9%, 84.6% and 100%, respectively.