Research on image recognition of cracked eggs based on convolutional neural network model

Abstract

Objective: In order to improve the accuracy and operating efficiency of the egg crack detection method based on computer vision. Methods: Used poultry egg simulation impact equipment to obtain cracked eggs, and collected images of cracked eggs and intact eggs from different angles through egg dynamic image acquisition equipment. Then, the YOLO-v5, ResNet and SuffleNet models for cracked egg image recognition were established for the original and the preprocessed egg images, respectively. After that, the recognition accuracy and the adaptability for original egg images recognition of different models were compared. Results: The YOLO-v5, ResNet and SuffleNet models could effectively identify the preprocessed cracked egg images, and the accuracy rates of verification set were 98.8%, 97.8% and 99.4% respectively. For original eggs images, the ResNet model had a low recognition accuracy, while the SuffleNet model had the highest recognition accuracy, which was up to 99%. Conclusion: Among the convolutional neural network models, SuffleNet model is most suitable for cracked egg image recognition, and the egg image preprocess is not necessary. This study provides a reference for the further improvement of crack egg detection methods based on computer vision.

Keywords

• egg
• crack
• computer vision
• convolutional neural network
• discrimination accuracy
• detection