Search Results (91)

Accurate and robust information on land cover changes in urban and coastal areas is essential for effective urban land management, ecosystem monitoring, and urban planning. This paper details the methodology and results of a pixel-level classification and change detection analysis, leveraging 1945 Royal Air Force (RAF) aerial imagery and 2011 Very High-Resolution (VHR) multispectral WorldView-2 satellite imagery from the broader area of Patras, Greece. Our attention is mainly focused on the changes in the coastline from the city of Patras to the northeast direction and the two major rivers, Charadros and Selemnos. The methodology involves preprocessing steps such as registration, denoising, and resolution adjustments to ensure computational feasibility for both coastal and riverbed change detection procedures while maintaining critical spatial features. For change detection at coastal areas over time, the Normalized Difference Water Index (NDWI) was applied to the new imagery to mask out the sea from the coastline and manually archive imagery from 1945. To determine the differences in the coastline between 1945 and 2011, we perform image differencing by subtracting the 1945 image from the 2011 image. This highlights the areas where changes have occurred over time. To conduct riverbed change detection, feature extraction using the Gray-Level Co-occurrence Matrix (GLCM) was applied to capture spatial characteristics. A Support Vector Machine (SVM) classification model was trained to distinguish river pixels from non-river pixels, enabling the identification of changes in riverbeds and achieving 92.6% and 92.5% accuracy for new and old imagery, respectively. Post-classification processing included classification maps to enhance the visualization of the detected changes. This approach highlights the potential of combining historical and modern imagery with supervised machine learning methods to effectively assess coastal erosion and riverbed alterations. Full article

Accurate understanding of urban land use change information is of great significance for urban planning, urban monitoring, and disaster assessment. The use of Very-High-Resolution (VHR) remote sensing images for change detection on urban land features has gradually become mainstream. However, most existing transfer learning-based change detection models compute multiple deep image features, leading to feature redundancy. Therefore, we propose a Transfer Learning Change Detection Model Based on Change Feature Selection (TL-FS). The proposed method involves using a pretrained transfer learning model framework to compute deep features from multitemporal remote sensing images. A change feature selection algorithm is then designed to filter relevant change information. Subsequently, these change features are combined into a vector. The Change Vector Analysis (CVA) is employed to calculate the magnitude of change in the vector. Finally, the Fuzzy C-Means (FCM) classification is utilized to obtain binary change detection results. In this study, we selected four VHR optical image datasets from Beijing-2 for the experiment. Compared with the Change Vector Analysis and Spectral Gradient Difference, the TL-FS method had maximum increases of 26.41% in the F1-score, 38.04% in precision, 29.88% in recall, and 26.15% in the overall accuracy. The results of the ablation experiments also indicate that TL-FS could provide clearer texture and shape detections for dual-temporal VHR image changes. It can effectively detect complex features in urban scenes. Full article

High- and very high-resolution (HR, VHR) remote sensing (RS) images can provide comprehensive and intricate spatial information for land cover classification, which is particularly crucial when analyzing complex built-up environments. However, the application of HR and VHR images to large-scale and detailed land cover mapping is always constrained by the intricacy of land cover classification models, the exorbitant cost of collecting training samples, and geographical changes or acquisition conditions. To overcome this limitation, we propose an unsupervised domain adaptation (UDA) with contrastive learning-based discriminative feature augmentation (CLDFA) for RS image classification. In detail, our method first utilizes contrastive learning (CL) through a memory bank in order to memorize sample features and improve model performance, where the approach employs an end-to-end Siamese network and incorporates dynamic pseudo-label assignment and class-balancing strategies for adaptive domain joint learning. By transferring classification models trained on a source domain (SD) to an unlabeled target domain (TD), our proposed UDA method enables large-scale land cover mapping. We conducted experiments using a massive five billion-pixels dataset as the SD and tested the HR and VHR RS images of five typical Chinese cities as the TD and applied the method on the completely unlabeled world view 3 (WV3) image of Urumqi city. The experimental results demonstrate that our method excels in large-scale HR and VHR RS image classification tasks, highlighting the advantages of semantic segmentation based on end-to-end deep convolutional neural networks (DCNNs). Full article

Solar radiation reflected by the Earth’s surface to satellite sensors is modified by its interaction with the atmosphere. The application of atmospheric correction of optical satellite imagery is an essential and needed pre-processing tool for modeling biophysical variables, multi-temporal analysis, and digital classification processes. As a result, true surface reflectance values are obtained without atmosphere influence. To assess this process, GEOSAT (part of the ESA’s Third-Party Mission Programme) performs an optimization of the GEOSAT 2 very high resolution (VHR) multispectral imagery adapting the well-known 6S model to the different wavelengths covered by the GEOSAT 2 spectral bands (VHR, PAN). The 6S model predicts surface reflectance (BOA) using information from the apparent reflectance (TOA) captured by the satellite sensor and the corresponding atmospheric conditions. To perform the atmospheric correction (AC), both the configuration of the atmosphere at the time of capture and the conditions of scene pointing and luminosity, must be considered. The first is mainly determined by three values: water vapor, ozone, and the number of air-suspended particles (aerosols). For the latter, the geometry of the scene, as well as the respective sun and sensor observation positions are the values to be considered. To validate the resultant GEOSAT 2 AC images, obtained from applying the GEOSAT atmospheric correction algorithm, different common areas between these and Sentinel-2 L2A products have been selected. Then, band-by-band (R, G, B, and NIR) operations were performed, such as the calculation of the mean square error (RMSE) and a regression analysis. Then, spectral profiles for the three generic land coverages (vegetation, soil, and water) were also gathered over the spectral range of GEOSAT 2 and S2 corresponding bands. The outcomes, once analyzed, lead us to conclude that the results obtained by applying the promising GEOSAT AC algorithm are satisfactory and seem to correctly estimate BOA reflectance values for vegetation and water coverages. To extend the study and improve the result, ground reflectance values will be required. Full article

As a medium-resolution multi-temporal data source, Sentinel-2 data has the potential to match the performance of using very-high-resolution (VHR) images in deep learning applications. To fully leverage the multi-temporal nature of Sentinel-2 data, we introduce the Deep Seasonal Network (DeepSN). This composite architecture combines a pre-trained deep convolutional neural network (DCNN) for visual feature extraction with a long short-term memory (LSTM) model to capture temporal information and make classification predictions. We evaluate the effectiveness of DeepSN on a Maasai Boma classification task in the Tanzania region. The DeepSN takes a sequence of four seasonal data, each spanning three months, for Boma prediction. Through cross-season validation experiments, we compare various advanced DCNNs and select EfficientNet as the backbone for DeepSN, as it performs the best. DeepSN with an EfficientNet backbone achieves a significant 19% improvement in the F1 score compared to plain EfficientNet for the Boma classification task. This work introduces a versatile composite architecture capable of handling multi-temporal data efficiently, providing flexibility in choosing the most suitable feature extraction backbone. The performance of DeepSN demonstrates the viability of utilizing medium-resolution multi-temporal data instead of high-resolution images for diverse tasks. Full article

Coldwater streams are crucial habitats for many biota including Salmonidae and Cottidae species that are unable to tolerate warmer water temperatures. Accurate classification of coldwater streams is essential for their conservation, restoration, and management, especially in light of increasing human disturbance and climate change. Coldwater streams receive cooler groundwater inputs and, as a result, typically remain ice-free during the winter. Based on this empirical thermal evidence, we examined the potential of very high-resolution (VHR) satellite and uncrewed aerial system (UAS) imagery to (i) detect coldwater streams using semi-automatic classification versus visual interpretation approaches, (ii) examine the physical factors that contribute to inaccuracies in detecting coldwater habitats, and (iii) use the results to identify inaccuracies in existing thermal stream classification datasets and recommend coverage updates. Due to complex site conditions, semi-automated classification was time consuming and produced low mapping accuracy, while visual interpretation produced better results. VHR imagery detected only the highest quality coldwater streams while lower quality streams that still met the thermal and biological criteria to be classified as coldwater remained undetected. Complex stream and site variables (narrow stream width, canopy cover, terrain shadow, stream covered by ice and drifting snow), image quality (spatial resolution, solar elevation angle), and environmental conditions (ambient temperature prior to image acquisition) make coldwater detection challenging; however, UAS imagery is uniquely suited for mapping very narrow streams and can bridge the gap between field data and satellite imagery. Field-collected water temperatures and stream habitat and fish community inventories may be necessary to overcome these challenges and allow validation of remote sensing results. We detected >30 km of coldwater streams that are currently misclassified as warmwater. Overall, visual interpretation of VHR imagery it is a relatively quick and inexpensive approach to detect the location and extent of coldwater stream resources and could be used to develop field monitoring programs to confirm location and extent of coldwater aquatic resources. Full article

Land Use/Land Cover (LULC) mapping is the first step in monitoring urban sprawl and its environmental, economic and societal impacts. While satellite imagery and vegetation indices are commonly used for LULC mapping, the limited resolution of these images can hamper object recognition for Geographic Object-Based Image Analysis (GEOBIA). In this study, we utilize very high-resolution (VHR) optical imagery with a resolution of 50 cm to improve object recognition for GEOBIA LULC classification. We focused on the city of Nice, France, and identified ten LULC classes using a Random Forest classifier in Google Earth Engine. We investigate the impact of adding Gray-Level Co-Occurrence Matrix (GLCM) texture information and spectral indices with their temporal components, such as maximum value, standard deviation, phase and amplitude from the multi-spectral and multi-temporal Sentinel-2 imagery. This work focuses on identifying which input features result in the highest increase in accuracy. The results show that adding a single VHR image improves the classification accuracy from 62.62% to 67.05%, especially when the spectral indices and temporal analysis are not included. The impact of the GLCM is similar but smaller than the VHR image. Overall, the inclusion of temporal analysis improves the classification accuracy to 74.30%. The blue band of the VHR image had the largest impact on the classification, followed by the amplitude of the green-red vegetation index and the phase of the normalized multi-band drought index. Full article

As one of the most important agricultural production types in the world, orchards have high economic, ecological, and cultural value, so the accurate and timely mapping of orchards is highly demanded for many applications. Selecting a remote-sensing (RS) data source is a critical step in efficient orchard mapping, and it is hard to have a RS image with both rich temporal and spatial information. A trade-off between spatial and temporal resolution must be made. Taking grape-growing regions as an example, we tested imagery at different spatial and temporal resolutions as classification inputs (including from Worldview-2, Landsat-8, and Sentinel-2) and compared and assessed their orchard-mapping performance using the same classifier of random forest. Our results showed that the overall accuracies improved from 0.6 to 0.8 as the spatial resolution of the input images increased from 58.86 m to 0.46 m (simulated from Worldview-2 imagery). The overall accuracy improved from 0.7 to 0.86 when the number of images used for classification was increased from 2 to 20 (Landsat-8) or approximately 60 (Sentinel-2) in one year. The marginal benefit of increasing the level of details (LoD) of temporal features on accuracy is higher than that of spatial features, indicating that the classification ability of temporal information is higher than that of spatial information. The highest accuracy of using a very high-resolution (VHR) image can be exceeded only by using four to five medium-resolution multi-temporal images, or even two to three growing season images with the same classifier. Combining the spatial and temporal features from multi-source data can improve the overall accuracies by 5% to 7% compared to using only temporal features. It can also compensate for the accuracy loss caused by missing data or low-quality images in single-source input. Although selecting multi-source data can obtain the best accuracy, selecting single-source data can improve computational efficiency and at the same time obtain an acceptable accuracy. This study provides practical guidance on selecting data at various spatial and temporal resolutions for the efficient mapping of other types of annual crops or orchards. Full article

Urban tree canopy (UTC) area is an important index for evaluating the urban ecological environment; the very high resolution (VHR) images are essential for improving urban tree canopy survey efficiency. However, the traditional image classification methods often show low robustness when extracting complex objects from VHR images, with insufficient feature learning, object edge blur and noise. Our objective was to develop a repeatable method—superpixel-enhanced deep neural forests (SDNF)—to detect the UTC distribution from VHR images. Eight data expansion methods was used to construct the UTC training sample sets, four sample size gradients were set to test the optimal sample size selection of SDNF method, and the best training times with the shortest model convergence and time-consumption was selected. The accuracy performance of SDNF was tested by three indexes: F1 score (F1), intersection over union (IoU) and overall accuracy (OA). To compare the detection accuracy of SDNF, the random forest (RF) was used to conduct a control experiment with synchronization. Compared with the RF model, SDNF always performed better in OA under the same training sample size. SDNF had more epoch times than RF, converged at the 200 and 160 epoch, respectively. When SDNF and RF are kept in a convergence state, the training accuracy is 95.16% and 83.16%, and the verification accuracy is 94.87% and 87.73%, respectively. The OA of SDNF improved 10.00%, reaching 89.00% compared with the RF model. This study proves the effectiveness of SDNF in UTC detection based on VHR images. It can provide a more accurate solution for UTC detection in urban environmental monitoring, urban forest resource survey, and national forest city assessment. Full article

Fundamental image processing methods, such as atmospheric corrections and pansharpening, influence the signal of the pixel. This morphs the spectral signature of target features causing a change in both the final spectra and the way different mapping methods may assign thematic classes. In the current study, we aim to identify the variations induced by popular image processing methods in the spectral reflectance and final thematic maps of facies. To this end, we have tested three different atmospheric corrections: (a) Quick Atmospheric Correction (QUAC), (b) Dark Object Subtraction (DOS), and (c) Fast Line-of-Sight Atmospheric Analysis of Hypercubes (FLAASH), and two pansharpening methods: (a) Hyperspherical Color Sharpening (HCS) and (b) Gram–Schmidt (GS). WorldView-2 and WorldView-3 satellite images over Chandra-Bhaga Basin, Himalaya, and Ny-Ålesund, Svalbard are tested via spectral subsets in traditional (BGRN1), unconventional (CYRN2), visible to near-infrared (VNIR), and the complete available spectrum (VNIR_SWIR). Thematic mapping was comparatively performed using 12 pixel-based (PBIA) algorithms and 3 object-based (GEOBIA) rule sets. Thus, we test the impact of varying image processing routines, effectiveness of specific spectral bands, utility of PBIA, and versatility of GEOBIA for mapping facies. Our findings suggest that the image processing routines exert an extreme impact on the end spectral reflectance. DOS delivers the most reliable performance (overall accuracy = 0.64) averaged across all processing schemes. GEOBIA delivers much higher accuracy when the QUAC correction is employed and if the image is enhanced by GS pansharpening (overall accuracy = 0.79). SWIR bands have not enhanced the classification results and VNIR band combination yields superior performance (overall accuracy = 0.59). The maximum likelihood classifier (PBIA) delivers consistent and reliable performance (overall accuracy = 0.61) across all processing schemes and can be used after DOS correction without pansharpening, as it deteriorates spectral information. GEOBIA appears to be robust against modulations in atmospheric corrections but is enhanced by pansharpening. When utilizing GEOBIA, we find that a combination of spatial and spectral object features (rule set 3) delivers the best performance (overall accuracy = 0.86), rather than relying only on spectral (rule set 1) or spatial (rule set 2) object features. The multiresolution segmentation parameters used here may be transferable to other very high resolution (VHR) VNIR mapping of facies as it yielded consistent objects across all processing schemes. Full article

Historical land cover (LC) maps are an essential instrument for studying long-term spatio-temporal changes of the landscape. However, manual labelling on low-quality monochromatic historical orthophotos for semantic segmentation (pixel-level classification) is particularly challenging and time consuming. Therefore, this paper proposes a methodology for the automated extraction of very-high-resolution (VHR) multi-class LC maps from historical orthophotos under the absence of target-specific ground truth annotations. The methodology builds on recent evolutions in deep learning, leveraging domain adaptation and transfer learning. First, an unpaired image-to-image (I2I) translation between a source domain (recent RGB image of high quality, annotations available) and the target domain (historical monochromatic image of low quality, no annotations available) is learned using a conditional generative adversarial network (GAN). Second, a state-of-the-art fully convolutional network (FCN) for semantic segmentation is pre-trained on a large annotated RGB earth observation (EO) dataset that is converted to the target domain using the I2I function. Third, the FCN is fine-tuned using self-annotated data on a recent RGB orthophoto of the study area under consideration, after conversion using again the I2I function. The methodology is tested on a new custom dataset: the ‘Sagalassos historical land cover dataset’, which consists of three historical monochromatic orthophotos (1971, 1981, 1992) and one recent RGB orthophoto (2015) of VHR (0.3–0.84 m GSD) all capturing the same greater area around Sagalassos archaeological site (Turkey), and corresponding manually created annotations (2.7 km² per orthophoto) distinguishing 14 different LC classes. Furthermore, a comprehensive overview of open-source annotated EO datasets for multiclass semantic segmentation is provided, based on which an appropriate pretraining dataset can be selected. Results indicate that the proposed methodology is effective, increasing the mean intersection over union by 27.2% when using domain adaptation, and by 13.0% when using domain pretraining, and that transferring weights from a model pretrained on a dataset closer to the target domain is preferred. Full article

Recent developments in remote sensing technology have allowed us to observe the Earth with very high-resolution (VHR) images. VHR imagery scene classification is a challenging problem in the field of remote sensing. Vision transformer (ViT) models have achieved breakthrough results in image recognition tasks. However, transformer–encoder layers encode different levels of features, where the latest layer represents semantic information, in contrast to the earliest layers, which contain more detailed data but ignore the semantic information of an image scene. In this paper, a new deep framework is proposed for VHR scene understanding by exploring the strengths of ViT features in a simple and effective way. First, pre-trained ViT models are used to extract informative features from the original VHR image scene, where the transformer–encoder layers are used to generate the feature descriptors of the input images. Second, we merged the obtained features as one signal data set. Third, some extracted ViT features do not describe well the image scenes, such as agriculture, meadows, and beaches, which could negatively affect the performance of the classification model. To deal with this challenge, we propose a new algorithm for feature- and image selection. Indeed, this gives us the possibility of eliminating the less important features and images, as well as those that are abnormal; based on the similarity of preserving the whole data set, we selected the most informative features and important images by dropping the irrelevant images that degraded the classification accuracy. The proposed method was tested on three VHR benchmarks. The experimental results demonstrate that the proposed method outperforms other state-of-the-art methods. Full article

The current deep convolutional neural networks for very-high-resolution (VHR) remote-sensing image land-cover classification often suffer from two challenges. First, the feature maps extracted by network encoders based on vanilla convolution usually contain a lot of redundant information, which easily causes misclassification of land cover. Moreover, these encoders usually require a large number of parameters and high computational costs. Second, as remote-sensing images are complex and contain many objects with large-scale variances, it is difficult to use the popular feature fusion modules to improve the representation ability of networks. To address the above issues, we propose a dynamic convolution self-attention network (DCSA-Net) for VHR remote-sensing image land-cover classification. The proposed network has two advantages. On one hand, we designed a lightweight dynamic convolution module (LDCM) by using dynamic convolution and a self-attention mechanism. This module can extract more useful image features than vanilla convolution, avoiding the negative effect of useless feature maps on land-cover classification. On the other hand, we designed a context information aggregation module (CIAM) with a ladder structure to enlarge the receptive field. This module can aggregate multi-scale contexture information from feature maps with different resolutions using a dense connection. Experiment results show that the proposed DCSA-Net is superior to state-of-the-art networks due to higher accuracy of land-cover classification, fewer parameters, and lower computational cost. The source code is made public available. Full article

Surface water is a crucial resource and environmental element for human survival and ecosystem stability; therefore, accurate information on the distribution of surface water bodies is essential. Extracting this information on a large scale is commonly implemented using moderate- and low-resolution satellite images. However, the detection and analysis of more detailed surface water structures and small water bodies necessitate the use of very high-resolution (VHR) satellite images. The large-scale application of VHR images for water extraction requires convenient and accurate methods. In this paper, a method combining a pixel-level water index and image object detection is proposed. The method was tested using 2018/2019 multispectral 4-m resolution images obtained from the Chinese satellite Gaofen-2 across Beijing, China. Results show that the automatic extraction of water body information over large areas using the proposed method and VHR images is feasible. Kappa coefficient and overall accuracy of 0.96 and 99.8% after post-classification improvement were obtained for testing images inside the Beijing area. The Beijing water body dataset obtained included a total of 489.53 km² of surface water in 2018/2019, 108.01 km² of which were ponds with an area smaller than 2 km². This study can be applied for water body extraction and mapping in other large regions and provides a reference for other methods for using VHR images to extract water body information on a large scale. Full article

Advancements in remote sensing have led to the development of Geographic Object-Based Image Analysis (GEOBIA). This method of information extraction focuses on segregating correlated pixels into groups for easier classification. This is of excellent use in analyzing very-high-resolution (VHR) data. The application of GEOBIA for glacier surface mapping, however, necessitates multiple scales of segmentation and input of supportive ancillary data. The mapping of glacier surface facies presents a unique problem to GEOBIA on account of its separable but closely matching spectral characteristics and often disheveled surface. Debris cover can induce challenges and requires additions of slope, temperature, and short-wave infrared data as supplements to enable efficient mapping. Moreover, as the influence of atmospheric corrections and image sharpening can derive variations in the apparent surface reflectance, a robust analysis of the effects of these processing routines in a GEOBIA environment is lacking. The current study aims to investigate the impact of three atmospheric corrections, Dark Object Subtraction (DOS), Quick Atmospheric Correction (QUAC), and Fast Line-of-Sight Atmospheric Analysis of Hypercubes (FLAASH), and two pansharpening methods, viz., Gram–Schmidt (GS) and Hyperspherical Color Sharpening (HCS), on the classification of surface facies using GEOBIA. This analysis is performed on VHR WorldView-2 imagery of selected glaciers in Ny-Ålesund, Svalbard, and Chandra–Bhaga basin, Himalaya. The image subsets are segmented using multiresolution segmentation with constant parameters. Three rule sets are defined: rule set 1 utilizes only spectral information, rule set 2 contains only spatial and contextual features, and rule set 3 combines both spatial and spectral attributes. Rule set 3 performs the best across all processing schemes with the highest overall accuracy, followed by rule set 1 and lastly rule set 2. This trend is observed for every image subset. Among the atmospheric corrections, DOS displays consistent performance and is the most reliable, followed by QUAC and FLAASH. Pansharpening improved overall accuracy and GS performed better than HCS. The study reports robust segmentation parameters that may be transferable to other VHR-based glacier surface facies mapping applications. The rule sets are adjusted across the processing schemes to adjust to the change in spectral characteristics introduced by the varying routines. The results indicate that GEOBIA for glacier surface facies mapping may be less prone to the differences in spectral signatures introduced by different atmospheric corrections but may respond well to increasing spatial resolution. The study highlighted the role of spatial attributes for mapping fine features, and in combination with appropriate spectral features may enhance thematic classification. Full article