Study Of YOLOv7-based Online Inspection System For Defective Eggs

China has now become the world’s largest egg producer,and eggs are one of the most common necessities in daily life,and demand for eggs has been increasing in recent years.Egg quality control plays an important role not only in ensuring food safety,but also in national health and safety and economic development.However,the egg industry inevitably produces eggs with shell defects during processing.Currently,most egg manufacturers still use the traditional human eye recognition to select defective eggs,which has its own physiological and empirical limitations and is prone to omissions and misdetection.In this paper,we analyze three deep learning target detection algorithm models,Faster-RCNN,YOLOv5 and YOLOv7,and compare and investigate the detection effect of Faster-RCNN,YOLOv5 and YOLOv7 network models on shell defective eggs.Meanwhile,this paper builds a computer image acquisition system for collecting raw egg image datasets,designs and develops a human-computer interaction defective egg online detection system based on PyQt5 and Qt Designer,and comparatively studies experiments with different number of defective eggs and different transmission speeds,and the detection results reach a high recognition level,providing an efficient solution for egg quality detection that meets the production inspection needs of the egg industry.The main work of this paper is as follows:(1)By analyzing common egg defect types,a computerized image acquisition system was built,and images of egg samples were collected in the Shuolai egg farm and laboratory in Hebei Province to construct a raw data set for egg detection.The egg dataset was also expanded and enriched by image enhancement,and the dataset was labeled and divided according to the target detection experiments and target detection algorithm training requirements.(2)Three target detection algorithms for defective egg detection were studied and analyzed.With the same initial value settings of the main variables of the target detection network,the defect recognition accuracy(mAP)of the three target detection algorithms Faster-RCNN,YOLOv5 and YOLOv7 for the same egg dataset were 91.08%,94.36%and 98.48%,respectively.The network model training results show that YOLOv7 has higher accuracy and lower false detection rate compared with the other two detection network models.(3)Based on the Python language software development environment Pycharm,combined with the network model image processing library OpenCV,a human-computer interaction defective egg online detection system was designed and developed using PyQt5 and Qt Designer development tools,which includes egg detection module and model management module,with six functional areas,the system operation interface is simple and logical The system has a simple interface and easy-to-understand logic.The system was also tested for different numbers of defective eggs and different transmission speeds.The experimental results proved that the system has a high level of detection and recognition and can meet the requirements of egg production.