Abstract: The leakage caused by damaged eggs will pollute the automatic production line and intact eggs,which will not only affect the production efficiency,but also interfere with the detection of cracked eggs.In order to realize the fast,accurate and low-cost recognition of damaged eggs,this paper proposes a damaged egg detection method based on YOLOv4 network by using machine vision technology,combined with the characteristics of deep-seated feature extraction and high-precision detection and classification of deep learning network.Build the damaged egg image data set,build the YOLOv4 deep learning network,and train the classification model containing damaged egg and intact egg images;The recognition accuracy of YOLOv4,YOLOv3 and faster RCNN network models for damaged eggs was compared;At the same time,in order to verify the online detection ability of YOLOv4,simulate and build the actual egg production environment,and compare the detection accuracy under different broken egg proportion and different moving speed.The results are as follows:under the same data set,the recognition accuracy of YOLOv4 is 4.62% higher than the average value of YOLOv3 and faster RCNN network model;In online detection,the average recognition accuracy of YOLOv4 model for damaged eggs with different proportions is 86.22%;When the moving speed of egg production line is 5～6 m/min,the average recognition accuracy is 84.91%.The results show that the damaged egg detection method based on YOLOv4 proposed in this paper has good detection effect and high detection rate for eggs moving on the convective waterline.It provides a new method for intelligent production and quality detection of eggs,and has certain practical value.