Search

Since individual neural networks have limited deep expressiveness and effectiveness, many learning frameworks face difficulties in the availability and balance of sample selection. As a result, in change detection, it is difficult to upgrade the hit rate of a high-performance model on both positive and negative pixels. Therefore, supposing that the sacrificed components coincide perfectly with the important evaluation objectives, such as positives, it would lose more than gain. To address this issue, in this paper, we propose a multi-visual collaborative deep network (MV-CDN) served by three collaborative network members that consists of three subdivision approaches, the CDN with one collaborator (CDN-C), CDN with two collaborators (CDN-2C), and CDN with three collaborators (CDN-3C). The purpose of the collaborator is to re-evaluate the feature elements in the network transmission, and thus to translate the group-thinking into a more robust field of vision. We use three sets of public double-temporal hyperspectral images taken by the AVIRIS and HYPERION sensors to show the feasibility of the proposed schema. The comparison results have confirmed that our proposed schema outperforms the existing state-of-the-art algorithms on the three tested datasets, which demonstrates the broad adaptability and progressiveness of the proposal. [ABSTRACT FROM AUTHOR]

Atmospheric correction of satellite optical imagery over inland waters is a key remaining challenge in aquatic remote sensing. This is due to numerous confounding factors such as the complexity of water optical properties, the surface glint, the heterogeneous nature of atmospheric aerosols, and the proximity of bright land surfaces. This combination of factors makes it difficult to retrieve accurate information about the system observed. Moreover, the impact of radiance coming from adjacent land (adjacency effects) in complex geometries further adds to this challenge, especially for small lakes. In this study, ten atmospheric correction algorithms were evaluated for high-resolution multispectral imagery of Landsat-8 Operational Land Imager and Sentinel-2 MultiSpectral Instrument using in situ optical measurements from ~300 lakes across Canada. The results of the validation show that the performance of the algorithms varies by spectral band and evaluation metrics. The dark spectrum fitting algorithm had the best performance in terms of similarity angle (spectral shape), while the neural network-based models showed the lowest errors and bias per band. However, none of the tested atmospheric correction algorithms meet a 30% retrieval accuracy target across all the visible bands, likely due to uncorrected adjacency effects. To quantify this process, three-dimensional radiative transfer simulations were performed and compared to satellite observations. These simulations show that up to 60% of the top of atmosphere reflectance in the near-infrared bands over the lake was from the adjacent lands covered with green vegetation. The significance of these adjacency effects on atmospheric correction has been analyzed qualitatively, and potential efforts to improve the atmospheric correction algorithms are discussed. [ABSTRACT FROM AUTHOR]

In order to better constrain the utility of multispectral datasets in the characterization of surface materials, Landsat Thematic Mapper (TM) data were evaluated in the discrimination of geological classes in the Cape Smith Belt of Quebec, a greenstone belt that hosts Early Proterozoic units including those of the Purtuniq ophiolite. Ground-based measurements collected for the study area highlight the importance of chemical alteration in controlling the reflectance properties of key geological classes. The spatial distribution of exposed lithologies in the study area was determined through (1) image classification using a feedforward backpropagation neural network classifier; and (2) generation of fraction images for spectral end members using a linear unmixing algorithm and ground reflectance data. Despite some shortcomings, the database of surface cover generated by the neural network classifier is a useful representation of the spatial distribution of exposed geological materials in the study area, with an overall agreement with ground truth of 87.7%. In contrast, the fraction images generated through unmixing are poor representations of ground truth for several key lithological classes. These results underscore both the considerable utility and marked limitations of Landsat TM data in the mapping of igneous and metamorphic lithologies. [ABSTRACT FROM AUTHOR]