Search

In recent history, normalized digital surface models (nDSMs) have been constantly gaining importance as a means to solve large-scale geographic problems. High-resolution surface models are precious, as they can provide detailed information for a specific area. However, measurements with a high resolution are time consuming and costly. Only a few approaches exist to create high-resolution nDSMs for extensive areas. This article explores approaches to extract high-resolution nDSMs from low-resolution Sentinel-2 data, allowing us to derive large-scale models. We thereby utilize the advantages of Sentinel 2 being open access, having global coverage, and providing steady updates through a high repetition rate. Several deep learning models are trained to overcome the gap in producing high-resolution surface maps from low-resolution input data. With U-Net as a base architecture, we extend the capabilities of our model by integrating tailored multiscale encoders with differently sized kernels in the convolution as well as conformed self-attention inside the skip connection gates. Using pixelwise regression, our U-Net base models can achieve a mean height error of approximately 2 m. Moreover, through our enhancements to the model architecture, we reduce the model error by more than 7%.

In this article, we present a novel hybrid framework, which integrates spatial–temporal semantic segmentation with postclassification relearning, for multitemporal land use and land cover (LULC) classification based on very high resolution (VHR) satellite imagery. To efficiently obtain optimal multitemporal LULC classification maps, the hybrid framework utilizes a spatial–temporal semantic segmentation model to harness temporal dependency for extracting high-level spatial–temporal features. In addition, the principle of postclassification relearning is adopted to efficiently optimize model output. Thereby, the initial outcome of a semantic segmentation model is provided to a subsequent model via an extended input space to guide the learning of discriminative feature representations in an end-to-end fashion. Last, object-based voting is coupled with postclassification relearning for coping with the high intraclass and low interclass variances. The framework was tested with two different postclassification relearning strategies (i.e., pixel-based relearning and object-based relearning) and three convolutional neural network models, i.e., UNet, a simple Convolutional LSTM, and a UNet Convolutional-LSTM. The experiments were conducted on two datasets with LULC labels that contain rich semantic information and variant building morphologic features (e.g., informal settlements). Each dataset contains four time steps from WorldView-2 and Quickbird imagery. The experimental results unambiguously underline that the proposed framework is efficient in terms of classifying complex LULC maps with multitemporal VHR images.

Spectral-spatial classification for hyperspectral imagery has been receiving much attention, since the detailed spectral and rich spatial information of hyperspectral images can be fully exploited to improve the classification accuracy. However, when the original hyperspectral images have very noisy bands, these bands may have an unfavorable impact on the classification, and are often discarded in advance based on expert knowledge. In this study, a spectral-spatial conditional random field classification algorithm integrating band selection (CRFBS) is developed for hyperspectral imagery with severe noise bands. The proposed algorithm integrates band selection based on the relative utility of the spectral bands for classification. Consequently, negative effects of severe noise bands are eliminated and the need for high-quality image data is substantially reduced. In addition, the CRFBS algorithm makes comprehensive use of both the spectral and the spatial cues to improve the classification performance. The spectral cues are formulated by integrating the support vector machine and random forest algorithms to improve the spectral discriminative ability in the unary potentials, and the spatial information are modeled to consider the interactions between pixels in pairwise potentials. The experiments using different airborne and UAV-borne hyperspectral data verified the effectiveness of the CRFBS method. The CRFBS algorithm can achieve accurate interpretation of the various classification categories and a more than 3% improvement in classification accuracy, compared with the method using the original hyperspectral image with severe noise bands.

Large-area urban ecology studies often miss information on vertical parameters of vegetation, even though they represent important constituting properties of complex urban ecosystems. The new globally available digital elevation model (DEM) of the spaceborne TanDEM-X mission has an unprecedented spatial resolution (12 × 12 m) that allows us to derive such relevant information. So far, suitable approaches using a TanDEM-X DEM for the derivation of a normalized canopy model (nCM) are largely absent. Therefore, this paper aims to obtain digital terrain models (DTMs) for the subsequent computation of two nCMs for urban-like vegetation (e.g., street trees) and forest-like vegetation (e.g., parks), respectively, in Berlin, Germany, using a TanDEM-X DEM and a vegetation mask derived from UltraCam-X data. Initial comparisons between morphological DTM-filter confirm the superior performance of a novel disaggregated progressive morphological filter (DPMF). For improved assessment of a DTM for urban-like vegetation, a modified DPMF and image enhancement methods were applied. For forest-like vegetation, an interpolation and a weighted DPMF approach were compared. Finally, all DTMs were used for nCM calculation. The nCM for urban-like vegetation revealed a mean height of 4.17 m compared to 9.61 m of a validation nCM. For forest-like vegetation, the mean height for the nCM of the weighted filtering approach (9.16 m) produced the best results (validation nCM: 13.55 m). It is concluded that an nCM from TanDEM-X can capture vegetation heights in their appropriate dimension, which can be beneficial for automated height-related vegetation analysis such as comparisons of vegetation carbon storage between several cities.