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

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