Research on the deep learning of the small sample data based on transfer learning

The Convolutional Neural Network of Deep Learning has been a huge success in the field of image recognition, however, it requires a lot of data samples to train a network of deep learning. In actual work, it’s too difficult to get a large number of training samples, and it will be easy to overfitting under the condition of the limited dataset. According to this problem, design a kind of Deep Convolutional Neural Network which based on the Transfer Learning to solve the problem of the small sample dataset. First of all, it uses the method of Data Augmentation to enlarge the number of sample dataset. Secondly, it uses the Transfer Learning to transfer the trained network (CNN) from the big sample dataset to our small sample dataset for secondary training, It use a Global Average Pooling instead of the fully connected layers to train the network, and wo use Soft max for classification. The method solves the problem of the small sample dataset in the deep learning, and improve the operation efficiency. The experimental results show that it has high recognition rate of the classification in small sample dataset.