Few-shot Image Classification Algorithm Based on Graph Network Optimization and Label Propagation

In recent years， deep learning technology has achieved competitive performance in image classification tasks， but in practical applications， there is often a lack of a large number of training samples， which may result in over-fitting phenomenon. Due to its advantages in representing inter and intra relationships of different samples， of graph neural network has been used for few-shot image classification tasks. The existing algorithms often use a few convolutional blocks to obtain image features as the node feature input graph network. In order to better represent the relationship between the graph nodes， the algorithm in this paper introduces the convolutional block attention module （CBAM） to enhance the saliency of the target regional features， use more representative features to optimize the graph network. This method implicitly enhances the relationship between samples through the spread of graph network information to achieve the purpose of optimizing node relationships. After that， the label propagation mechanism is used to classify unknown samples. Intensive experimental results proved the excellent performance of the proposed algorithm in few-shot supervised and semi-supervised learning tasks， and verified its effectiveness.