DEEP ACTIVE LEARNING BASED ON SALIENCY-GUIDED DATA AUGMENTATION FOR IMAGE CLASSIFICATION

Image classification based on Deep learning usually requires a large amount of labeled data for model training. However, the labeling cost of data is expensive. The aim of active learning is to reduce the cost of manual labeling by selecting lesser data for labeling through a query sampling strategy. In this paper, the framework of active learning is improved based on saliency guided data augmentation, while enhancing the generality of the algorithm and reducing labeling cost. Firstly, data augmentation based on a saliency map is used to expand the sample data, which can retain the most important information in the data. In addition, a low-computation neural network, i.e., SpinalNet, is incorporated into the proposed algorithm to optimize the classification network in the active learning algorithm to compensate for the increased computational cost by data augmentation. Finally, the experiments show that the proposed framework can improve the performance of the baseline.