Semi-Supervised Few-shot Image Classification Based on Subspace Learning

Abstract Few-shot image classification is an technology that uses minimal sample data to train the classifier or network and achieves a sure accuracy. Compared with based on big data, it has advantages of unlimited number samples fast processing speed. The purpose subspace learning learn good mapping. This mapping maps high-dimensional from visual low-dimensional semantic information space, which convenient for recognition, processing, classification. Therefore, this paper proposes use method extract features. On basis, studies semi-supervised few-shot classification, self-training algorithm improved overcome performance degradation caused by adding wrong pseudo label algorithm. It trained tested four mainstream datasets. specific significance research development