Research Of Egg Crack Identification Technology Based On Machine Vision

At present,China’s large-scale egg processing enterprises generally rely on foreign processing equipment to detect and remove cracked eggs,so the expensive costs to bear hinder the development of enterprises.Therefore,the study of egg crack identification technology is of considerable economic significance to China’s egg product processing industry.Based on this,the specific research steps are as follows:First of all,the egg image acquisition platform is designed in the laboratory environment to achieve the complete acquisition of egg surface information.Through the analysis of the functions of the egg sorting system,each function is divided into modular modules: visual inspection module,sorting module,conveyor module and communication module.According to the workflow of the egg sorting system,the appropriate hardware solution is selected,and the appropriate software development environment and software design scheme are selected for the hardware linkage.Secondly,the image data pre-processing of the collected egg graphic samples is carried out,and the classification dataset of cracked eggs and intact eggs is constructed.The SVM classification method was applied to identify cracked eggs and intact eggs,and the comprehensive accuracy rate reached 84.4%;In addition,the object detection algorithm based on YOLO is applied to the egg crack image detection model,and the accuracy of training and tuning is 95%,which is better than the SVM classification method,so as to construct the egg crack image recognition algorithm based on deep learning in this paper.Finally,Based on the crack detection algorithm and the fusion target tracking algorithm,the video-based egg detection is realized,to demonstrate the accuracy of the model based on the improve deep learning image recognition algorithm,this paper uses an egg transport video as a test video,with an average accuracy of 89.7% and an average accuracy of 94.5%.Although the FPS has been reduced to 12 frames per second,it still meets the requirements for real-time performance.Detection and tracking of cracked egg targets has a more robust effect.