Research On Detection Technology Of Spraying Code Character Defects In Milk Production Data Based On Computer Vision

With the increasing living standards of people,milk has gradually become a daily drink for people,and the quality and safety of milk has become more and more important to consumers.The date of manufacture is an important criterion for milk safety and its content should be accurate.However,in the process of printing,there are inevitably many defects in the printing date of the production date,such as missing printing,missing printing,misprinting,ink pollution and the like.Once the milk with the defects of the production date is in the market,the manufacturer will inevitably be punished by the relevant departments,and will also affect its own brand image,causing unnecessary losses.Therefore,the detection of the character defects of the milk production date is necessary.In view of the problems of traditional manual detection,such as time-consuming and high cost,this paper studies the character detection of milk-printing date based on computer vision technology.The research content includes the following aspects:(1)In terms of milk production date positioning,aiming at the problem that the milk image background is complex and the production date is difficult to extract,a method of extracting the region of interest(ROI)based on prior knowledge is proposed.The method performs the production date ROI extraction on the milk image by analyzing the printing process of the production date to reduce the interference of the background information.Aiming at the need of detection of code character defects in milk production date,a positioning algorithm based on mathematical morphology combined with line scanning is proposed.The algorithm firstly performs image preprocessing on the extracted production date ROI,then processes the image by mathematical morphology method to obtain the milk production date candidate area.Finally,the line scanning method is used to detect the defect of the coded characters.Quickly and accurately locate the milk production date area.(2)In the aspect of coding character segmentation on the production date,a milk production date correction algorithm based on least squares method is proposed for the problem that the milk production date is tilted and the character segmentation is inaccurate.The algorithm uses the least squares method to perform straight line fitting to obtain the tilt angle of the milk production date,and rotates the image for correction.Aiming at the adhesion of the characters of milk production date,an improved algorithm of vertical projection segmentation based on dichotomy is proposed.The algorithm improves the vertical projection segmentation method by combining the dichotomy of finding the optimal segmentation threshold multiple times to realize the accurate segmentation of the code characters of milk production date.In the judgment stage of single character defect,Blob analysis method is used to obtain information such as the area and perimeter of the coded characters,and the defects of the code characters of the milk production date are detected by comparing with the standard information.(3)In the aspect of coding character identification on the production date,a recognition algorithm based on improved LeNet-5 network model is proposed for the characteristics of the code characters of milk production date.Based on the research of the classical convolutional neural network structure LeNet-5,the network structure and parameters are optimized,the activation function is improved,and an improved LeNet-5 model suitable for milk-coded date character recognition is constructed.The improved LeNet-5 model was trained and the improved LeNet-5 model was used to identify and detect the character defects of the milk production date.In this thesis,the Python 3.6 development environment is used to study the fingerprint detection technology of milk production date based on computer vision.The detection algorithm is verified and tested by the large amount of milk images collected in the actual production environment.The simulation results show that the algorithm can be accurate.The production date of the milk image character defect in the milk image is detected,and the processing speed can basically meet the requirements of realtime.