{"title":"Meta Self-Supervised Learning for Distribution Shifted Few-Shot Scene Classification","authors":"Tengfei Gong, Xiangtao Zheng, Xiaoqiang Lu","doi":"10.1109/lgrs.2022.3174277","DOIUrl":null,"url":null,"abstract":"Few-shot classification tries to recognize novel remote sensing image categories with a few shot samples. However, current methods assume that the test dataset shares the same domain with the labeled training dataset where prior knowledge is learned. It is infeasible to collect a training dataset for each domain, since remote sensing images may come from various domains. Exploiting the existing labeled dataset from another domain (source domain) to help the target dataset (target domain) classification would be efficient. In this letter, both meta-learning and self-supervised learning are jointly conducted for few-shot classification. Specifically, meta-learning is executed over a pretrained network for few-shot classification. Furthermore, self-supervised learning is exploited to fit the target domain distribution by training on unlabeled target domain images. Experiments are conducted on NWPU, EuroSAT and Merced datasets to validate the effectiveness.","PeriodicalId":13046,"journal":{"name":"IEEE Geoscience and Remote Sensing Letters","volume":"19 1","pages":"1-5"},"PeriodicalIF":4.0000,"publicationDate":"2022-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"5","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Geoscience and Remote Sensing Letters","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.1109/lgrs.2022.3174277","RegionNum":3,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 5
Abstract
Few-shot classification tries to recognize novel remote sensing image categories with a few shot samples. However, current methods assume that the test dataset shares the same domain with the labeled training dataset where prior knowledge is learned. It is infeasible to collect a training dataset for each domain, since remote sensing images may come from various domains. Exploiting the existing labeled dataset from another domain (source domain) to help the target dataset (target domain) classification would be efficient. In this letter, both meta-learning and self-supervised learning are jointly conducted for few-shot classification. Specifically, meta-learning is executed over a pretrained network for few-shot classification. Furthermore, self-supervised learning is exploited to fit the target domain distribution by training on unlabeled target domain images. Experiments are conducted on NWPU, EuroSAT and Merced datasets to validate the effectiveness.
期刊介绍:
IEEE Geoscience and Remote Sensing Letters (GRSL) is a monthly publication for short papers (maximum length 5 pages) addressing new ideas and formative concepts in remote sensing as well as important new and timely results and concepts. Papers should relate to the theory, concepts and techniques of science and engineering as applied to sensing the earth, oceans, atmosphere, and space, and the processing, interpretation, and dissemination of this information. The technical content of papers must be both new and significant. Experimental data must be complete and include sufficient description of experimental apparatus, methods, and relevant experimental conditions. GRSL encourages the incorporation of "extended objects" or "multimedia" such as animations to enhance the shorter papers.