Research on Label-free Super-resolution Land Cover Mapping Based on Deep Learning

In this study, we propose a deep learning-based unlabeled land cover mapping method to achieve 2 m resolution land cover mapping in Glossy County, Fujian Province, without local high-resolution labels by integrating label super-resolution (SR) and Instance Batch Normalized Network (IBN−Net). The results show that label super-resolution can be achieved using the improved fully convolutional neural network (FCN) model, which upgrades the low-resolution labels to high-resolution and effectively improves the classification accuracy. The IBN−Net network enhances the generalization ability of the model and significantly improves the effectiveness of cross-domain applications. Compared with endogenous low-resolution labels, using high-precision exogenous labels improved the overall accuracy of the model in Glossy County by 2.55% to 85.48%. The method in this study significantly improves the accuracy of land cover mapping without matching labels, which can provide effective technical support for regional ecological monitoring and management.