Solar radiation, the radiant energy from the sun, is a driving variable for numerous ecological, physiological, and other life-sustaining processes in the environment. Traditional methods to quantify solar radiation are done either directly (e.g., quantum sensors), or indirectly (e.g., hemispherical photography). This study, however, evaluates literature which utilized remote sensing (RS) technologies to estimate various forms of solar radiation or components, thereof under or within forest canopies. Based on the review, light detection and ranging (LiDAR) has, so far, been preferably used for modeling light under tree canopies. Laser system’s capability of generating 3D canopy structure at high spatial resolution makes it a reasonable choice as a source of spatial information about light condition in various parts of forest ecosystem. The majority of those using airborne laser system (ALS) commonly adopted the volumetric-pixel (voxel) method or the laser penetration index (LPI) for modeling the radiation, while terrestrial laser system (TLS) is preferred for canopy reconstruction and simulation. Furthermore, most of the studies focused only on global radiation, and very few on the diffuse fraction. It was also found out that most of these analyses were performed in the temperate zone, with a smaller number of studies made in tropical areas. Nonetheless, with the continuous advancement of technology and the RS datasets becoming more accessible and less expensive, these shortcomings and other difficulties of estimating the spatial variation of light in the forest are expected to diminish.
1 July 2019 in International Journal of Applied Earth Observation and Geoinformation
Generative adversarial networks (GANs) are a type of neural network that are characterized by their unique construction and training process. Utilizing the concept of the latent space and exploiting the results of a duel between different GAN components opens up interesting opportunities for computer vision (CV) activities, such as image inpainting, style transfer, or even generative art. GANs have great potential to support aerial and satellite image interpretation activities. Carefully crafting a GAN and applying it to a high-quality dataset can result in nontrivial feature enrichment. In this study, we have designed and tested an unsupervised procedure capable of engineering new features by shifting real orthophotos into the GAN’s underlying latent space. Latent vectors are a low-dimensional representation of the orthophoto patches that hold information about the strength, occurrence, and interaction between spatial features discovered during the network training. Latent vectors were combined with geographical coordinates to bind them to their original location in the orthophoto. In consequence, it was possible to describe the whole research area as a set of latent vectors and perform further spatial analysis not on RGB images but on their lower-dimensional representation. To accomplish this goal, a modified version of the big bidirectional generative adversarial network (BigBiGAN) has been trained on a fine-tailored orthophoto imagery dataset covering the area of the Pilica River region in Poland. Trained models, precisely the generator and encoder, have been utilized during the processes of model quality assurance and feature engineering, respectively. Quality assurance was performed by measuring model reconstruction capabilities and by manually verifying artificial images produced by the generator. The feature engineering use case, on the other hand, has been presented in a real research scenario that involved splitting the orthophoto into a set of patches, encoding the patch set into the GAN latent space, grouping similar patches latent codes by utilizing hierarchical clustering, and producing a segmentation map of the orthophoto.
Generative adversarial networks (GAN) opened new possibilities for image processing and analysis. In- painting, dataset augmentation using artificial samples, or increasing spatial resolution of aerial imagery are only a few notable examples of utilising GANs in remote sensing (RS). The normalised difference vegetation index (NDVI) ground-truth labels were prepared by combining RGB and NIR orthophotos. The dataset was then utilised as input for a conditional generative adversarial network (cGAN) to perform an image-to-image translation. The main goal of the neural network was to generate an artificial NDVI image for each processed 256 px × 256 px patch using only in- formation available in the panchromatic input. The network achieved a structural similarity index measure (SSIM) of 0.7569 ± 0.1083, a peak signal-to-noise ratio (PSNR) of 26.6459 ± 3.6577 and a root-mean-square error (RSME) of 0.0504 ± 0.0193 on the test set, which should be considered high. The perceptual evaluation was performed to verify the meth- od’s usability when working with a real-life scenario. The research confirms that the structure and texture of the pan- chromatic aerial RS image contain sufficient information for NDVI estimation for various objects of urban space. Even though these results can highlight areas rich in vegetation and distinguish them from the urban background, there is still room for improvement regarding the accuracy of the estimated values. The research aims to explore the possibility of utilising GAN to enhance panchromatic images (PAN) with information related to vegetation. This opens exciting opportunities for historical RS imagery processing and analysis.
14 December 2022 in Acta Scientiarum Polonorum Administratio Locorum
Motives: According to public statistics guidelines, areas officially classified in Lodz city as urban greenery include only forests, parks, lawns, squares and cemeteries. Areas of so-called unsealed greenery are omitted, which, however, have a great positive impact on improving the living conditions of the population. By taking information from satellite images and comparing them with official data, we have received a closer to the reality picture of the city, which is much more better than it would appear from official statistical data. Another dimension which the study addresses is the uneven distribution of greenery of a certain quality in individual units of the city. Aim: Comparing these data with the fact that the distribution of places of residence is also uneven, an attempt was made to assess the accessibility of green areas for the inhabitants of Lodz city. Results: The results show that there are much more green spaces, similar in terms of vegetation abundance to the official green spaces. That means the city is underestimated when talking about the degree of greenery.
Rapid changes in the human-changed and natural environment have intensified the need for efficient, scalable, and accurate topographic map updates. Traditional methods, which rely on labor-intensive and costly field surveys, are increasingly inadequate due to time and resource constraints. In parallel, advances in remote sensing have greatly improved the availability of high-resolution, multimodal data–though often at the cost of increased complexity and heterogeneity. This study demonstrates that such data, combined with state-of-the-art deep learning techniques, can effectively support the automation of topographic map updates through bi-temporal change detection. Focusing on urban areas where changes are most frequent, we propose a framework that integrates a panoptic segmentation model with a novel change detection algorithm designed to process VNIR orthophotos and classified LiDAR data. The algorithm combines three complementary metrics–semantic class comparison, geometric overlap, and histogram similarity–to categorize detected changes as added, removed, or modified. Experimental validation across 40 synthetic and real-world scenarios confirms the robustness of the approach, with an average processing time of 20 ms per 500 $$\text {m}^{2}$$ tile using GPU acceleration. Based on the Mask2Former architecture, the segmentation model generates high-quality semantic and instance masks for diverse object classes, achieving 0.86–0.87 mIoU for buildings and roads and 0.94 mIoU for forests and cultivated land. Incorporating LiDAR data improves segmentation performance by 5.9 percentage points. The framework enables computer-aided workflows with user-adjustable thresholds and produces georeferenced raster outputs compatible with GIS analysis. Future work will address segmentation challenges and explore soft classification strategies to enhance interpretability.