Few-Shot Image Classification Method Based on Self-Supervised Task Pre-training

Deep learning has received a lot of attention in recent years and has rapidly made significant progress in research and practical applications, and has become a major approach in computer vision. However, it relies on a large annotated dataset, and the process of collecting and labeling these datasets is both labor-intensive and expensive, often making it difficult to obtain large datasets. Therefore, it is particularly important to enable deep learning to adapt to small sample size situations. Research on learning with fewer samples is still in its infancy, and the performance gap is obvious when compared to large sample scenarios. In this paper, we address this problem by exploring image classification and target detection under sample less conditions. We propose a few-shot image classification method using self-supervised pre-training to enhance the generalization capability. Experimental results show that the method exhibits competitive performance.