Research On Candy Defect Detection And Grading Technology Based On Machine Vision

In the production process of candy,due to mechanical extrusion and mutual collision between candy and other reasons,it is inevitable that defective candy will appear and can not be put into the market,so it is necessary to eliminate defective candy.However,as a large candy producing country,China adopts the traditional method of manual detection in the grading process.Although manual detection may be superior in some cases,it is inefficient and tedious,which is easy to lead to eye fatigue,and it will also lead to classification errors due to different judgments of different people.Because of its low sampling rate,poor real-time performance and low detection reliability,the traditional manual detection method can not meet the efficiency and quality requirements of modern industrial production line.In this context,this paper designs a set of candy defect detection and classification system,including the corresponding hardware equipment and algorithms.Aiming at the three types of candy defects: damage,hole and small area,this paper realizes the detection and elimination of defective candy,which has good application value.Firstly,this paper designs the detection and grading process of defective candy,takes the green hard candy provided by a candy company in Nantong as the experimental sample,and statistically analyzes the three most common types of defective candy: hole,damage and small area.At the same time,the hardware structure of the system is designed,and the corresponding image acquisition equipment is analyzed and selected,including industrial camera,lens,light source,main control computer and industrial control board.The reasonable selection of image acquisition equipment provides an important guarantee for the quality of image.Secondly,the flow of image processing is designed,and the principle of image processing algorithm of candy defects is analyzed in detail.Filter the candy image and filter out the noise points in the image due to the on-site environment.By comparing various filtering algorithms:median filter,mean filter and Gaussian filter,the median filter algorithm with the best processing effect is finally selected.Next,the image is enhanced,and the methods of color image graying and histogram equalization are adopted to improve the contrast between background and foreground and better present the characteristics of the image.Then the image background is segmented,and the threshold segmentation algorithm based on threshold and the segmentation algorithm based on boundary edge are analyzed and compared.The results show that using Otsu threshold segmentation method,the distinction between background and foreground is the largest,and the candy target can be segmented from the background better.Then the coordinates of each candy in the image are located by the connected domain marking method.On this basis,the performance of eight classification models(alexnet,googlenet,vgg16,resnet-18,resnet-34,resnet-50,mobilenetw2 and mnasnet0＿5)are constructed and tested.Among them,the classification accuracy of the classification model based on resnet-50 is the highest(98.71%),and combined with the accuracy,speed and running time,the classification model based on Alex net is the most suitable for the system.Finally,the software and hardware facilities of candy defect classification system are designed.In view of the problems of confection adhesion encountered in the production process,the SDVC10-S vibratory feeding controller is used to make the candy be sprayed discretely on the conveyor belt,and the extended circuit module embedded in the industrial control board is designed for the signal transmission problem between the CMOS image sensor camera and the EPC-9600 industrial control board.The layout of man-machine interface is introduced,which can realize the functions of the system designed in this paper,including the setting of sorting threshold and the correction of light intensity,and the performance of the system is tested.The results show that the recall rate of the system is more than 95% and the take-out rate is less than 2%,which meets the needs of industrial production,and realizes the integration of candy defect detection and defect elimination.