Abstract: A copper alloy gold phase diagram classification method based on an improved residual network and transfer learning was proposed for the specific metallographic datasets in the copper alloy field, which are usually very small and cannot meet the problem of large training samples required for traditional convolutional neural network modeling. Based on ImageNet, the ResNet50 residual model was pre trained using the ImageNet dataset and metallographic diagram training set. Transfer learning method was used during training, and a fully connected layer was re-established to classify and recognize copper alloy metallographic categories. The experimental results show that the accuracy of this method is 97.2%, the precision is 95.6%, the recall is 97.3% and the F1-score is 96.4%, which is superior to VGG19 and MobileV2 based on transfer learning. Moreover, the transfer learning method can overcome the problem of small metallographic image datasets, and the use of ResNet50 for feature extraction can effectively obtain the texture information of copper alloy metallographic maps. This study establishes a new method for automatic classification and identification of metallographic structures of copper alloys, which can more accurately classify and identify the metallographic structures.