Research on Fish Image Classification Based on Transfer Learning and Convolutional Neural Network Model

Traditional fish image classification process is complex and cumbersome, and traditional classification is lack in accuracy. Using convolutional neural networks to identify fish requires a large amount of sample which often gives rise to difficultly. Based on the convolutional neural network model, this paper puts forward the application of transfer learning to classify and identify fish images when the sample is small. First, the image data is normalized and enhanced to improve the classification results. The last layer of the ResNet50 network of the convolutional neural network model is transformed into the six-category full-connection layer, and the pre-trained weights on the ImageNet data set are transferred to the ResNet50 network. The experiment investigates the effect of ResNet50 frozen layers and training samples on fish classification accuracy and overfitting problems. The experimental results indicate that the ResNet 50 network model entertains good extraction ability and generalization ability of image features with the application of transfer learning and it is not easy to overfit. The accuracy of fish image classification can be as high as 97.19%. Meanwhile, the method has obvious time advantage.