In recent years,deep learning technology has made remarkable achievements in big data training models.However,due to the particularity of the field,it is difficult to obtain a large number of labeled/unlabeled samples.And manual annotation of data will consume a lot of manpower,limiting its application and promotion in this field.Few-shot object recognition and classification only requires a small amount of labeled sample data to achieve the recognition and classification of typical objects under limited sample conditions.Taking recognition and classification as the main application field,the research status of three commonly used algorithms for improving the performance of few-shot object recognition and classification based on data enhancement,transfer learning and metric learning is introduced and the advantages and disadvantages of these algorithms are elaborated.Finally,some challenges still faced by few-shot learning and prospects for future research directions are sorted out and summarized.
关键词
小样本目标识别分类/数据增强/迁移学习/度量学习
Key words
few-shot object recognition and classification/data augmentation/transfer learning/metric learning
维普
万方数据