Research on CAD Image Classification Method Based on Deep Neural Networks :Transfer Learning with ResNet Model

This paper studies the application of transfer learning techniques based on ResNet in CAD image classification. By fine-tuning the pre-trained ResNet model, experimental results indicate that transfer learning significantly improves classification accuracy, especially when training data is limited. Specifically, experiments compared three network models: ResNet-50, ResNet-101, and ResNet-152. The results show that ResNet-50 performed the best on both the training and validation sets, while the deeper ResNet-101 and ResNet-152 models exhibited relatively lower classification accuracy due to overfitting and the constraints of the dataset size. Ultimately, the ResNet-50 model demonstrated excellent performance under conditions of limited datasets and computing resources, making it suitable for CAD image classification tasks. Nevertheless, there is still room for improvement in the classification performance of certain specific categories. Future work can focus on optimizing the model structure and employing more sophisticated data augmentation techniques to enhance performance.