Egg food is one of the important food in people's daily life, affordable, nutritious, delicious, with the improvement of the quality of life of our people, the quality of eggs is also increasing. An important step in the egg processing process is sorting, that is, picking out damaged eggs (cracked eggs) from good eggs. At present, China's egg processing industry still uses artificial detection methods, namely according to the egg method and percussion method, respectively, through the human eyes and ears to identify. Manual testing requires skilled individuals to carry out long-term production operations, easy to produce visual or auditory fatigue and lead to the loss of damaged eggs, and the increase in the cost of employment, recruitment difficulties will cause production stagnation. In view of the above problems, modern information technology such as communication technology, computer vision technology, image processing technology, pattern recognition technology and ML technology is introduced into the field of duck egg sorting, and a new automatic duck egg surface defect detection platform is established, which has become an urgent requirement and development trend of agricultural production. With the development of deep learning technology, researchers begin to apply ML technology to image recognition, and have made great progress in recent years. The application of neural network in egg surface defect recognition can greatly reduce the workload and shorten the recognition time. The emergence of deep learning technology provides a strong guarantee for image recognition. This project proposes an egg surface defect detection platform based on improved GoogLeNet, with a view to changing the way of egg sorting, so that it can eventually replace manual detection to reduce production costs, reduce false detection rate, improve enterprise productivity, so as to truly realize the automation, intelligence and standardization of the egg production sorting stage.