2021/08/16 by Rahul Ghosh, Xiaowei Jia, Ghosh, Rahul +7
Computer Science · Earth and Planetary Sciences · Engineering · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Remote Sensing and Land Use #Remote-Sensing Image Classification
paper · pdf · doi:10.48550/arxiv.2108.07323
openalex publication_date 2021/08/16 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Collecting large annotated datasets in Remote Sensing is often expensive and\nthus can become a major obstacle for training advanced machine learning models.\nCommon techniques of addressing this issue, based on the underlying idea of\npre-training the Deep Neural Networks (DNN) on freely available large datasets,\ncannot be used for Remote Sensing due to the unavailability of such large-scale\nlabeled datasets and the heterogeneity of data sources caused by the varying\nspatial and spectral resolution of different sensors. Self-supervised learning\nis an alternative approach that learns feature representation from unlabeled\nimages without using any human annotations. In this paper, we introduce a new\nmethod for land cover mapping by using a clustering based pretext task for\nself-supervised learning. We demonstrate the effectiveness of the method on two\nsocietally relevant applications from the aspect of segmentation performance,\ndiscriminative feature representation learning and the underlying cluster\nstructure. We also show the effectiveness of the active sampling using the\nclusters obtained from our method in improving the mapping accuracy given a\nlimited budget of annotating.\n