Advances in Uncertainty-Guided Local Climate Zone Classification

Abstract

Like many other research fields, remote sensing has been greatly impacted by machine and deep learning and benefits from technological and computational advances. In recent years, considerable effort has been spent on deriving not just accurate, but also reliable models which yield a sense of predictive uncertainty. In the particular framework of image classification, the reliability is e.g. validated by cross-checking the model’s confidence in its predictions against the resulting accuracy. Predictive uncertainties, on the other hand, can be for example used to determine expressive data samples. We investigate model reliability in the framework of Local Climate Zone (LCZ) classification, using the So2Sat LCZ42 [1] data set comprised of Sentinel-1 and Sentinel-2 image pairs. [1] X. X. Zhu, J. Hu, C. Qiu, Y. Shi, J. Kang, L. Mou, H. Bagheri, M. Haberle, Y. Hua, R. Huang et al., “So2sat lcz42: a benchmark data set for the classification of global local climate zones [software and data sets],” IEEE Geoscience and Remote Sensing Magazine, vol. 8, no. 3, pp. 76–89, 2020.