In addressing the potential performance issues and weak generalization capabilities of traditional ResNet networks in image classification tasks across various research domains,this study proposes an improved ResNet model,TA-ResNet50,based on transfer learning.The model incorporates channel and spatial attention modules to enhance its focus on critical information.Optimization designs are applied to the fully connected layers,and a Dropout layer is introduced to mitigate the risk of overfitting.Transfer learning techniques are employed to accelerate model training and improve generalization.To validate the effectiveness of the TA-ResNet50 model,a series of experiments are conducted on the GTSRB,CIFAR-10,and CIFAR-100 datasets.Experimental results demonstrate that compared to traditional ResNet models,TA-ResNet50 achieves respective increases of 12.73%,12.51%,and 10.98%in image recognition accuracy on these datasets.Additionally,the TA-ResNet50 model exhibits enhanced robustness and generalization capabilities.This study provides valuable insights for the application of transfer learning in deep neural networks.
关键词
迁移学习/ResNet50/注意力机制/图像分类
Key words
transfer learning/ResNet50/attention mechanism/image classification
维普
万方数据