Research On Quality Inspection Technology Of Fruit Based On Machine Vision

For China,such as fruit production,fruit automatic grading detection of great significance.At present,the research emphasis on this kind of research has been increasing year by year.Firstly,the production line and vision system are designed.Based on the analysis of the experimental data,the background of the fruit grading line is designed as white.Then,the visual inspection technology for the quality of green plum and Gonggan was studied.There are few studies on the grading detection of green plum and Gonggan,these two kinds of fruits,the purpose of this paper is to detect and classify the above two kinds of fruits by means of computer vision.The automatic detection of fruit and vegetable quality is rating to provide technical support.In the classification of green plum,four kinds of state image,including normal state,circular point defect,scratch damage defect and strip block defect,were collected and used to detect the defects of 1000 samples.The classifier is trained and tested by using convolution neural network as the classifier.An exploratory analysis of the color model component,H component is determined in green plum image segmentation,has used K means clustering and fuzzy C means clustering for image analysis of Prunus salicina l do clustering results,and ultimately determine the use of K means clustering algorithm,combined enhancement algorithm and area constraints Retinex the algorithm is improved.Test the accuracy of the classifier can reach 97.5%,the overall effective rate was 92.7%.The results show that the visual detection algorithm and the result of the classifier are the same,and the segmentation of the different skin defect state is more accurate more stable.In the classification test of Gonggan,50 samples of normal and defect were collected for defect detection.Then,the data of different color space of Gonggan image were analyzed and compared.The color component of the image that is applied to the image o f the tribute.In the study of epidermal defects of Gonggan,two kinds of image segmentation algorithms were studied.The first is the use of traditional Otsu algorithm for Gonggan image segmentation and defect detection.The second method is to use the integrated Otsu method based on voting to image the image of Gonggan,and then use the five decision trees combined with the random forest method to explore the results of the image segmentation of citrus skin defects detection.The detection rate of the first method was 90.5%,and the detection rate of the defect was 92%.According to the characteristics of epidermis and the results of the experiment,the traditional Otsu method is more suitable for the detection of epidermal defects.The random forest method can try the other fruits with smooth skin on the problem of fruit epidermis defect detection.Through the research,we have made some progress in the quality inspection technology of green plum and Gonggan,and reached the research purpose and expected requirements,which laid the foundation for further research on the automatic detection and classification technology of fruit and vegetable quality.