Research on Intelligent Recognition Method of Egg cracks Based on EfficientNet Network Model

Abstract

Abstract Aiming at the egg crack detection task, a 6-position egg image acquisition method is proposed. EfficientNet was used to classify egg cracks. Using transfer learning technology, the model is pre-trained with CIFAR-10 data set to obtain initial weights. The efficiency of model training is improved. Compared with the current mainstream CNN models of Alexnet, VGG16 and ResNet50, in terms of egg crack recognition, the average correct recognition rate of EfficientNet model is as high as 98.6 %, which is significantly better than Alexnet, VGG16 and ResNet50 models. Among them, the correct recognition rate of EfficientNetB2 model is 99.5 %, the training time is only 29 min, and the comprehensive performance is the best. It is the most suitable method for egg crack detection. Finally, three learning rates are set for the EfficientNetB2 model. The experimental results show that the model has the best performance when the learning rate is 0.001. The improved model proposed in this paper improves the accuracy of egg crack detection and reduces the model training time.