基于元学习和位置信息的小样本学习方法
Few-Shot Learning Method Based on Meta-Learning and Location Information

随着小样本学习的发展，元学习已经成为一种流行的小样本学习框架，其作用是开发能够快速适应有限数据和低计算成本的小样本分类任务模型。最近有关注意力的研究已经证明了通道注意力对于特征提取效果有一定的提升，但是它忽略了位置信息的作用，位置信息对于在小样本任务中更好地从有限的数据中学习来说很重要。基于这一事实，本文提出一种新的方法，通过在所有基类上预先训练一个加入位置信息注意力的分类器，然后在基于最近质心的小样本分类算法上进行元学习，实现了将位置信息和提取特征有效的结合。通过在两个标准的数据集上实验，和当下主流的小样本图像分类方法相比，该方法在Mini-ImageNet数据集的1-shot与5-shot任务上分别提升1.23%和1.02%，在Tiered-ImageNet数据集上，也分别提升0.85%和0.78%。实验表明该方法有效的发挥了位置信息的作用，可以提升小样本图像分类的准确率。
With the development of few-shot learning, meta-learning has become a popular few-shot learning framework, and its role is to develop models for few-shot classification tasks that can quickly adapt to limited data and low computational cost. Recent studies on attention have demonstrated that channel attention can improve feature extraction to a certain extent, but it ignores the role of location information, which is important for better learning with limited data in few-shot tasks. Based on this fact, this paper proposes a new method to effectively combine location information and extracted features by pre-training a classifier with location information attention on all base classes, and then performing meta-learning on few-shot classification algorithm based on the nearest centroid. Through experiments on two standard datasets, compared with the current mainstream few-shot image classification methods, this method was improved by 1.23% and 1.02% on the 1-shot and 5-shot tasks of the Mini-ImageNet dataset, and also by 0.85% and 0.78% on the Tiered-ImageNet dataset. Experiments show that the method effectively plays the role of location information and can improve the accuracy of few-shot image classification.