Search

Self-supervised learning (SSL) has significantly bridged the gap between supervised and unsupervised learning in computer vision tasks and shown impressive success in the field of remote sensing (RS). However, these methods have primarily focused on single-modal RS data, which may have limitations in capturing the diversity of information in complex scenes. In this paper, we propose the Asymmetric Attention Fusion (AAF) framework to explore the potential of multi-modal representation learning compared to two simpler fusion methods: early fusion and late fusion. Given that data from active sensors (e.g., digital surface models and light detection and ranging) is often noisier and less informative than optical images, the AAF is designed with an asymmetric attention mechanism within a two-stream encoder, applied at each encoder stage. Additionally, we introduce a Transfer Gate module to select more informative features from the fused representations, enhancing performance in downstream tasks. Our comparative analyses on the ISPRS Potsdam datasets, focusing on scene classification and segmentation tasks, demonstrate significant performance enhancements with AAF compared to baseline methods. The proposed approach achieves an improvement of over 7% in all metrics compared to randomly initialized methods for both tasks. Furthermore, when compared to early fusion and late fusion methods, AAF consistently outperforms in achieving superior improvements. These results underscore the effectiveness of AAF in leveraging the strengths of multi-modal RS data for SSL, opening doors for more sophisticated and nuanced RS analysis. [ABSTRACT FROM AUTHOR]

Groundwater icings, typical features of permafrost hydrology, are indicative of hydrothermal interactions between surface and ground waters, and permafrost. Their main morphological parameters, i.e., icing area and volume, are generally estimated with low accuracy. Only scarce field observational data on icing volume and seasonal development exist to date. Our study evaluates and compares performance of several widely used techniques of icing morphometric estimation, based on field data, collected on a giant Icing #2 in the Samokit River basin, southern Yakutia. Groundwater icing area was estimated by: (a) staking, (b) unmanned aerial vehicle (UAV) surveys, and (c) satellite imagery analysis. Icing #2 area in late February was between 1.38·106 m2 and 1.68·106 m2, icing volume, between 1.73·106 m3 and 4.20·106 m3, depending on the technique used. Staking is the least accurate, but also the only direct technique, which is hence used as a baseline tool in our study. Staking-based assessment of icing morphometry is the most conservative, while UAV-based estimates of icing area are higher by 14% to 17%, and of icing volume, by 74% to 142%, compared to staking. The latter appears, in our case, to be the least accurate method, although a direct one. It requires a sufficient number of staking points and transects, which should be set up to represent all icing zones, i.e., channel branches and alluvial islands. Photogrammetry based on UAV surveys has numerous advantages, i.e., higher precision of a per pixel icing volume calculation, based on an ice-free valley bottom digital surface model (DSM), and potential reusability of a resulting DSM. However, positioning precision suffers from the overlay of multiple flyovers required because of battery replacements, and, in our case, an insufficient number of ground control points. Satellite imagery along with B.L. Sokolov's empirical approach were used to estimate the annual maximum icing area and volume, and the empirical estimates tend to converge to satellite-based values. Finally, all thing being equal, UAV-based photogrammetry shows higher precision in estimating the icing morphometrical parameters. [ABSTRACT FROM AUTHOR]