Search Results (3)

For the last decade, Gaussian process regression (GPR) proved to be a competitive machine learning regression algorithm for Earth observation applications, with attractive unique properties such as band relevance ranking and uncertainty estimates. More recently, GPR also proved to be a proficient time series processor to fill up gaps in optical imagery, typically due to cloud cover. This makes GPR perfectly suited for large-scale spatiotemporal processing of satellite imageries into cloud-free products of biophysical variables. With the advent of the Google Earth Engine (GEE) cloud platform, new opportunities emerged to process local-to-planetary scale satellite data using advanced machine learning techniques and convert them into gap-filled vegetation properties products. However, GPR is not yet part of the GEE ecosystem. To circumvent this limitation, this work proposes a general adaptation of GPR formulation to parallel processing framework and its integration into GEE. To demonstrate the functioning and utility of the developed workflow, a GPR model predicting green leaf area index (LAI${}_G$) from Sentinel-2 imagery was imported. Although by running this GPR model into GEE any corner of the world can be mapped into LAI${}_G$ at a resolution of 20 m, here we show some demonstration cases over western Europe with zoom-ins over Spain. Thanks to the computational power of GEE, the mapping takes place on-the-fly. Additionally, a GPR-based gap filling strategy based on pre-optimized kernel hyperparameters is also put forward for the generation of multi-orbit cloud-free LAI${}_G$ maps with an unprecedented level of detail, and the extraction of regularly-sampled LAI${}_G$ time series at a pixel level. The ability to plugin a locally-trained GPR model into the GEE framework and its instant processing opens up a new paradigm of remote sensing image processing. Full article

Networking services may be broken down in a disaster situation while people in the disaster area(s) strongly demand networking services for both communication and information sharing among users. This requires the fast restoration of networking services to fulfil the demand–supply gap. Although there are a number of studies on restoring communication and networking in disasters, few studies have explicitly examined the service availability during the temporary and partial recovery process of network restoration. From the perspective of users in the disaster area, it is important to be able to communicate or share information with people whenever they want/need to do so. Therefore, partial and local recovery of the networking services also plays an important role for improving service availability in the disaster situations. To assess the restoration effectiveness of networking services with a measure of user satisfaction level, we propose to use instant networking service availability (I-NSA), a novel metric, and we examine the effectiveness of networking service restoration solutions using the metric. I-NSA allows us to clearly express the instant availability of networking services that drastically changes with the elapsed time from the disaster occurrence in disaster areas. This paper examines the effective improvement of I-NSA when Local-and-Instant Clouds (LI-Clouds) are applied to the disaster situation. LI-Cloud has been designed and practically developed to provide deployable networking services to users. We verify that LI-Cloud enables significant improvement on the I-NSA performance in the fast restoration of networking services. Full article

This paper discusses a high-performance similarity measurement method based on known map information named the cross mean absolute difference (CMAD) method. Applying the conventional normalized cross-correlation (NCC) feature registration method requires sufficient numbers of feature points, which must also exhibit near-normal distribution. However, Light Detection and Ranging (LiDAR) ranging point cloud data scanned and collected on-site are scarce and do not fulfill near-normal distribution. Consequently, considerable localization errors occur when NCC features are registered with map features. Thus, the CMAD method was proposed to effectively improve the NCC algorithm and localization accuracy. Because uncertainties in localization sensors cause deviations in the localization processes, drivable moving regions (DMRs) were established to restrict the range of location searches, filter out unreasonable trajectories, and improve localization speed and performance. An error comparison was conducted between the localization results of the window-based, DMR–CMAD, and DMR–NCC methods, as well as those of the simultaneous localization and mapping methods. The DMR–CMAD method did not differ considerably from the window-based method in its accuracy: the root mean square error in the indoor experiment was no higher than 10 cm, and that of the outdoor experiment was 10–30 cm. Additionally, the DMR–CMAD method was the least time-consuming of the three methods, and the DMR–NCC generated more localization errors and required more localization time than the other two methods. Finally, the DMR–CMAD algorithm was employed for the successful on-site instant localization of a car. Full article