Search

Regular inspection and maintenance of infrastructure facilities are crucial to ensure their functionality and safety for users. However, current inspection methods are labor-intensive and can vary depending on the inspector. To improve this process, modern sensor systems and machine learning algorithms can be deployed to detect defects based on rapidly acquired data, resulting in lower downtime. A quality-controlled processing chain allows to provide hence informed uncertainty assessments to inspection operators. In this study, we present several Deeplab V3+ models optimized to predict corroded segments of the quay wall at JadeWeserPort, Germany, which is a dataset from the 3D HydroMapper research project. Our models achieve generally high accuracy in detecting this damage type. Therefore, we examine the use of a Region Growing-based weakly supervised approach to efficiently extend our model to other common types in the future. This approach achieves about 90 % of the results compared to corresponding fully supervised networks, of which a ResNet-50 variant peaks at 55.6 % Intersection-over-Union regarding the test set’s corrosion class.

In recent years, remote sensing images have been used for many different applications that require visual analysis and interpretation. In this paper, reducing/removing noise is the basic approach, as it causes loss of information and therefore affects the accuracy of the analyses. Within the scope of the study, two different test areas of land cover/use were applied to examine the effects of noise on optical satellite images. In this context, Landsat 8 and Sentinel 2 satellites were used to study the effects of denoising methods on different spatial resolutions. Due to the lack of raw images of the selected satellites, two different types of noise (i.e. Gaussian and Stripe) were added to the images. In this context, four different denoising methods were compared by using conventional filter techniques commonly used in the spatial domain, while also different methods that used different threshold values in the frequency domain. The first approach is Median, Block Matching and 3D Filtering methods in the spatial domain, applications that depend mainly on the neighborhood relationship of pixels in the image. The second approach is wavelet-based Contourlet and Curvelet methods in the frequency domain. The quality analysis of denoised images were evaluated as qualitative (statistical methods Peak Signal to Noise Ratio, Mean Square Error, standard deviation, min/max value), and quantitative. Finally, Curvelet hard thresholding transform was the selected method as the best algorithm after quality analysis additionally, the method also effectively preserves edges in homogeneous test area and other fine details in the heterogeneous test area.

Deep Learning (DL) algorithms provide numerous benefits in different applications, and they usually yield successful results in scenarios with enough labeled training data and similar class proportions. However, the labeling procedure is a cost and time-consuming task. Furthermore, numerous real-world classification problems present a high level of class imbalance, as the number of samples from the classes of interest differ significantly. In various cases, such conditions tend to promote the creation of biased systems, which negatively impact their performance. Designing unbiased systems has been an active research topic, and recently some DL-based techniques have demonstrated encouraging results in that regard. In this work, we introduce an extension of the Debiasing Variational Autoencoder (DB-VAE) for semantic segmentation. The approach is based on an end-to-end DL scheme and employs the learned latent variables to adjust the individual sampling probabilities of data points during the training process. For that purpose, we adapted the original DB-VAE architecture for dense labeling in the context of deforestation mapping. Experiments were carried out on a region of the Brazilian Amazon, using Sentinel-2 data and the deforestation map from the PRODES project. The reported results show that the proposed DB-VAE approach is able to learn and identify under-represented samples, and select them more frequently in the training batches, consequently delivering superior classification metrics.

Structural disaster damage detection and characterisation is one of the oldest remote sensing challenges, and the utility of virtually every type of active and passive sensor deployed on various air- and spaceborne platforms has been assessed. The proliferation and growing sophistication of UAV in recent years has opened up many new opportunities for damage mapping, due to the high spatial resolution, the resulting stereo images and derivatives, and the flexibility of the platform. We have addressed the problem in the context of two European research projects, RECONASS and INACHUS. In this paper we synthesize and evaluate the progress of 6 years of research focused on advanced image analysis that was driven by progress in computer vision, photogrammetry and machine learning, but also by constraints imposed by the needs of first responder and other civil protection end users. The projects focused on damage to individual buildings caused by seismic activity but also explosions, and our work centred on the processing of 3D point cloud information acquired from stereo imagery. Initially focusing on the development of both supervised and unsupervised damage detection methods built on advanced texture features and basic classifiers such as Support Vector Machine and Random Forest, the work moved on to the use of deep learning. In particular the coupling of image-derived features and 3D point cloud information in a Convolutional Neural Network (CNN) proved successful in detecting also subtle damage features. In addition to the detection of standard rubble and debris, CNN-based methods were developed to detect typical façade damage indicators, such as cracks and spalling, including with a focus on multi-temporal and multi-scale feature fusion. We further developed a processing pipeline and mobile app to facilitate near-real time damage mapping. The solutions were tested in a number of pilot experiments and evaluated by a variety of stakeholders.

Height models are basic information required for disaster Management. Not in any case satisfying and actual height models are available, but they can be generated by satellite stereo pairs being more precise as InSAR. The Korean Kompsat-3 has a ground sampling distance of 0.71m. A stereo combination covering the main part of Istanbul has been used for the generation of height models. Kompsat-3 images are available as L1R images, corresponding close to the original image geometry, and as L1G, being projected to the SRTM 3 arcsec height model. For use of Semi Global Matching quasi epipolar images are required. They can be produced by just rotating the L1G-images to the stereo base, while with L1R-images requires at first a projection to a constant height level. The projection of L1G to the SRTM height models leads to height differences against the SRTM heights. The orientation of the L1R images with 71 ground control points (GCP) was possible in X and Y with 0.6 GSD and in Z with 1.1 GSD, while with L1G images only 1.2 GSD respectively 2.9 GSD have been reached. A standard deviation of 0.6 GSD for X and Y and 1.1 GSD for Z is satisfying and a usual accuracy for satellite images. A comparison of the generated height model based on the L1G-images with airborne LiDAR data (ALS) showed clear local systematic height errors of the height model based on L1G-images which could not be seen with L1R-images. The area based least squares matching leads to good results in open areas while in build up areas no accurate building determination is possible. Here SGM has a clear advantage with accurate roof structures corresponding to the 0.71 m GSD. For the relative accuracy, that means the building height and the roof structure, it does not matter if L1G or L1R images are used.

To decrease the dependency on fossil fuels, more renewable energy sources need to be explored. Over the last years, the consumption of biomass has risen steadily and it has become a major source for re-growing energy. Besides the most common sources of biomass (forests, agriculture etc.) there are smaller supplies available in mostly unused areas like hedges, vegetation along streets, railways, rivers and field margins. However, these sources are not mapped and in order to obtain their potential for usage as a renewable energy, a method to quickly assess their spatial distribution and their volume is needed. We use a range of data sets including satellite imagery, GIS and elevation data to evaluate these parameters. With the upcoming Sentinel missions, our satellite data is chosen to match the spatial resolution of Sentinel-2 (10-20m) as well as its spectral characteristics. To obtain sub-pixel information from the satellite data, we use a spectral unmixing approach. Additional GIS data is provided by the German Digital Landscape Model (ATKIS Base-DLM). To estimate the height (and derive the volume) of the vegetation, we use LIDAR data to produce a digital surface model. These data sets allow us to map the extent of previously unused biomass sources. This map can then be used as a starting point for further analyses about the feasibility of the biomass extraction and their usage as a renewable energy source. BMWi/DLR/50EE1333 BMWi/DLR/50EE1334 BMWi/DLR/50EE1335

This paper will present the project RAPIDMAP. The project is part of CONCERT-Japan, an ERA-NET initiative funded through the FP7 INCO project frame for enhancing research cooperation between European countries and Japan on two topics, one of which is Resilience Against Disasters. The project started in June 2013 and has a duration of 2 years. In the paper, we will outline the aims of the project, methodologies and techniques to be developed and some test data. Remote Sensing (RS) and Geographic Information System (GIS) are powerful technologies for collecting useful information on the damages of disasters in short time. However, since many different types of RS data are available (satellite, aerial, UAV, terrestrial), data co-registration, information integration and feature extraction need reliable and advanced methodologies. In the RAPIDMAP project, we will develop practical ways to integrate RS data processing tools in near-real-time and allow users to use this data soon after the disasters by means of WebGIS tools. This will help not only decision makers but also end-users in the disaster area. The key components of this project are: (1) Near-real-time monitoring: the procedure of near-real-time monitoring with satellites as well as Unmanned Airborne Vehicles (UAV) will be set up and demonstrated. (2) Data co-registration: in disasters, various images as well as maps come from different sources. The co-registration of multiple images is a key technology for information integration. In this project, a system to co-register multiple images in near-real-time will be developed. (3) Data fusion and change detection: one of the advantages of RS is to collect information with multiple sensors. Various methods for fusing optical with active microwave (SAR) sensor data for information extraction and change detection will be developed. (4) Decision Support System (DSS) based on WebGIS technologies: the collected and integrated information has to be easily accessible and visible by decision makers and end-users in near-real-time and worldwide. By using WebGIS technologies, wireless networks and portable terminals, a DSS will allow easy access, retrieval and visualization of all information (fused data, images, maps, etc.) in very short time after data collection and processing. The project will be practically tested and demonstrated at the Tohoku area in Japan and another test site, which were recently affected by large disasters.

High density point clouds of urban scenes are used to identify object classes like buildings, vegetation, vehicles, ground, and water. Point cloud segmentation can support classification and further feature extraction provided that the segments are logical groups of points belonging to the same object class. A single segmentation method will typically not provide a satisfactory segmentation for a variety of classes. This paper explores the combination of various segmentation and post-processing methods to arrive at useful point cloud segmentations. A feature based on the normal vector and flatness of a point neighbourhood is used to group cluttered points in trees as well as points on surfaces in areas where the extraction of planes was not successful. Combined with segment merging and majority filtering large segments can be obtained allowing the derivation of accurate segment feature values. Results are presented and discussed for a 70 million point dataset over a part of Rotterdam.

River restoration projects, which aim at improved flood safety and increased ecological value, have resulted in more heterogeneous vegetation. However, they also resulted in increasing hydraulic roughness, which leads to higher flood water levels during peak discharges. Due to allowance of vegetation development and succession, both ecological and hydraulic characteristics of the floodplain change more rapidly over time. Monitoring of floodplain vegetation has become essential to document and evaluate the changing floodplain characteristics and associated functioning. Extraction of characteristics of low vegetation using single-epoch remote sensing data, however, remains challenging. The aim of this study was to (1) evaluate the performance of multi-temporal, high-spatial-resolution UAV imagery for extracting temporal vegetation height profiles of grassland and herbaceous vegetation in floodplains and (2) to assess the relation between height development and NDVI changes. Vegetation height was measured six times during one year in 28 field plots within a single floodplain. UAV true-colour and false-colour imagery of the floodplain were recorded coincidently with each field survey. We found that: (1) the vertical accuracy of UAV normalized digital surface models (nDSMs) is sufficiently high to obtain temporal height profiles of low vegetation over a growing season, (2) vegetation height can be estimated from the time series of nDSMs, with the highest accuracy found for combined imagery from February and November (RMSE = 29-42 cm), (3) temporal relations between NDVI and observed vegetation height show different hysteresis behaviour for grassland and herbaceous vegetation. These results show the high potential of using UAV imagery for increasing grassland and herbaceous vegetation classification accuracy.

Matching Persistent Scatterers (PS) to airborne optical imagery is one possibility to augment applications and deepen the understanding of SAR processing and products. While recently this data registration task was done with PS and optical nadir images the alternatively available optical oblique imagery is mostly neglected. Yet, the sensing geometry of oblique images is very similar in terms of viewing direction with respect to SAR.We exploit the additional information coming with these optical sensors to assign individual PS to single parts of buildings. The key idea is to incorporate topology information which is derived by grouping regularly aligned PS at facades and use it together with a geometry based measure in order to establish a consistent and meaningful matching result. We formulate this task as an optimization problem and derive a graph matching based algorithm with guaranteed convergence in order to solve it. Two exemplary case studies show the plausibility of the presented approach.

Supervised classification of remotely sensed images is a classical method to update topographic geospatial databases. The task requires training data in the form of image data with known class labels, whose generation is time-consuming. To avoid this problem one can use the labels from the outdated database for training. As some of these labels may be wrong due to changes in land cover, one has to use training techniques that can cope with wrong class labels in the training data. In this paper we adapt a label noise tolerant training technique to the problem of database updating. No labelled data other than the existing database are necessary. The resulting label image and transition matrix between the labels can help to update the database and to detect changes between the two time epochs. Our experiments are based on different test areas, using real images with simulated existing databases. Our results show that this method can indeed detect changes that would remain undetected if label noise were not considered in training.

In coastal areas morphological changes of various kinds are caused by tidal flows, storms, climate change, and human activities. For these reasons a recurrent monitoring becomes necessary in order to detect undesired changes at early stages enabling rapid countermeasures to mitigate or minimize potential harm or hazard. The morphology of the terrain can be represented by highly precise digital terrain models (DTM). Airborne lidar (light detection and ranging) has become a standard method for DTM generation in coastal zones like Wadden Sea areas. In comparison to echo sounding systems, lidar is feasible for data acquisition of large areas. However, only the eulittoral zone can be covered by standard laser because the near-infrared laser pulses are not able to penetrate water which remains, for example, in some tidal channels even during low tide. In the framework of a German research project, we analyse the spatial and temporal variability of Wadden Sea areas in the North Sea. For a systematic monitoring and the detection of morphological changes we compare terrain models of two different epochs in order to determine height differences which can be caused by natural influences or human activities. We focus especially on the analysis of morphological changes near to tidal channels. In order to detect changes we compare the location of edges derived from each DTM based on the gray values' gradients. Our results for a test site in the German Wadden Sea show height differences up to 1 m due to the shifting of tidal channels and relocations of the channels up to 55 m within a period of two years.

The German Aerospace Center (DLR, Deutsches Zentrum für Luft- und Raumfahrt) currently conducts the bistatic interferometric synthetic aperture radar (SAR) Mission TanDEM-X, which shall result in a DEM of global coverage in an unprecedented resolution and accuracy according to DTED level 3 standard. The mission is based on the two SAR satellites TerraSAR-X and TanDEM-X that have been launched in June 2007 and 2010, respectively. After the commissioning phase of TanDEM satellite and the orbital adjustment the bistatic image acquisition in close formation began end of 2010. The data collection for the mission is scheduled to last about three years, i.e., the bigger part of the required data have been already gathered. Based on this data DLR will conduct several processing steps in order to come up finally with a global and seamless DEM of the Earth's landmass which shall meet the envisaged specifications. Since the entire mission is an endeavor in the framework of a private-public-partnership, the private partner, Astrium, will eventually commercialize the DEM product. In this paper, we will provide an overview of the data collection and the deliverables that will come along with TanDEM-X mission. Furthermore, we will analyze a DEM derived from early stage immediate products of the mission.

The paper presents the research carried out in the on-going 3DSolarWeb project to test and implement a complete pipeline for the generation of a solar cadastre of building roofs located in alpine areas. The project aims at providing reliable results in a costeffective way, using (low resolution) available data and new aerial imagery acquisitions as input. The environmental context is digitally represented using already existing low resolution LiDAR data (1–2 m resolution), while the urban area is modelled using high resolution aerial images (10–20 cm GSD) and photogrammetric DSM. Reliable models and algorithms for the estimation of the incoming sun radiance are then adopted and a WebGIS is set up for the interactive calculation of the photovoltaic (PV) potential in a raster-based form. The paper summarizes the entire pipeline and the results (Figure 1) achieved on the test areas to show the potentialities of the method and the web-based service.

The General Directorate of Surveying and Mapping (GDSM), under the Ministry of Municipal and Rural Affairs (MOMRA) is responsible for the production and dissemination of accurate geospatial data for all the metropolitan cities, towns and rural settlements in the Kingdom of Saudi Arabia. GDSM maintains digital geospatial databases that support the production of conventional line and orthophoto maps at scales ranging from 1:1,000 to 1:20,000. The current procedures for the acquisition of new aerial imagery cover a long time cycle of three or more years. Consequently, the availability of recently acquired High Resolution Satellite Imagery (HRSI) presents an attractive alternative image data source for rapid response to updated geospatial data needs. The direct sensor orientation of HRSI is not accurate enough requiring ground control points (GCP). A field survey of GCP is time consuming and costly. Seeking an alternative approach, a research study has recently been completed to use existing image and data base information instead of traditional ground control for the orthoprojection of HRSI in order to automate and speed up as much as possible the whole process. Based on a series of practical experiments, the ability for automated matching of aerial and satellite images by using the Speeded-Up Robust Features (SURF) algorithm is demonstrated to be useful for this task. Practical results from matching with SURF validate the ability for multi-scale, multi-sensor and multi-season matching of aerial and satellite images. The matched tie points are then used to transform the satellite orthophoto to the aerial orthophoto through a 2D affine coordinate transformation. GeoEye-1 and IKONOS imagery, when geo-referenced through SURF-based matching and transformed meet the MOMRA Map Accuracy Standards for 1:10,000 and 1:20,000 scale. However, a similarly processed SPOT-5 image does not meet these standards. This research has led to the development of a simple and efficient tool for the geo-referencing of HRSI of 0,5m to 1m ground sampling distance (GSD) that can be used for updating map information. The process completely eliminates the need for any ground control as well as image measurements by human operators.

Traditional ship-based bathymetric surveys based on echo sounding are expensive and time consuming. This paper presents a project with the aim of identifying the opportunities of airborne laser bathymetry for the monitoring of the sea bed at the German Baltic Sea coast. Such devices operate with laser signal in the green part of the visible spectrum which is capable to penetrate the water. The depth is determined from the two-way runtime between the water surface and reflections from the ground underneath. Several flight campaigns in representative test areas will be carried out in order to analyze the reachable depths, the accuracies of the acquired points, and the detection of obstacles depending on different water conditions (e.g. turbidity). We discuss some preliminary results of a pilot project and the first campaign of a study area close to the island of Poel, Germany.

21st Congress of the International Society for Photogrammetry and Remote Sensing, ISPRS 2008 --3 July 2008 through 11 July 2008 -- -- In this paper, it is aimed to investigate the capabilities of boosting classification approach for forest fire detection using SPOT-4 imagery. The study area, Bodrum in the province of Mugla, is located at the south-western Mediterranean coast of Turkey where recent largest forest fires occurred in July 2007. Boosting method is one of the recent advanced classifiers proposed in the machine learning community, such as neural networks classifiers based on multilayer perceptron (MLP), radial basis function and learning vector quantization. The Adaboost (AB) and Logitboost (LB) algorithms which are the most common boosting methods were used for binary and multiclass classifications. The effectiveness of boosting algorithms was shown through comparison with Bayesian maximum likelihood (ML) classifier, neural network classifier based on multilayer perceptron (MLP) and regression tree (RT) classifiers. The pre and post SPOT images were corrected atmospherically and geometrically. Binary classification comprised burnt and non-burnt classes. In addition to the pixel based classification, textural measures including, gray level co-occurrence matrix such as entropy, homogeneity, second angular moment, etc. were also incorporated. Instead of the traditional boosting weak (base) classifiers such as decision tree builder or perceptron learning rule, neural network classifier based on multilayer perceptron were adapted as a weak classifier. The accuracy of the MLP was greater than that of ML, AB, LB and RT both using spectral data alone and textural data. The use of texture measures alone was found to increase classification accuracy of binary and multi-class classifications. The accuracy of the land cover classifications based on either binary or multi-class was maximised using a MLP approach. This was slightly greater than the accuracy achieved using AB and LB classifications. However, it was shown that AB and LB classifications hold great potential as an alternative to conventional techniques.

21st International Congress for Photogrammetry and Remote Sensing, ISPRS 2008 --3 July 2008 through 11 July 2008 -- -- The aim of this study was to compare percent tree cover products of Envisat MERIS and MODIS data of Seyhan River Basin at the Eastern Mediterranean Region of Turkey. In this study, Regression Tree (RT) algorithm was used to estimate percent tree cover maps. This technique is well suited for percentage tree cover mapping because, as a non-parametric classifier, it requires no prior assumptions about the distribution of the training data. This model also allows for the calibration of the model along the entire continuum of tree cover, avoiding the problems of using only end members for calibration.The medium resolution Envisat MERIS with a 300 m and MODIS with a 500 m pixel representation data were used as predictor variables. Three scenes of high resolution IKONOS images were employed as a training data, and testing the accuracy of model. The regression tree method for this study consisted of six steps: i) generate reference percentage tree cover data, ii) derive metrics from Envisat MERIS and MODIS data, iii) select predictor variables, iv) fit RT model, v) undertake accuracy assessment and produce final model and map, vi) compare results. The training data set was derived supervised land cover classification of IKONOS imagery to generate reference percent tree cover data. Specifically, this classification was aggregated to estimate percent tree cover at the MERIS and MODIS spatial resolution.The predictor variables incorporated the MERIS and MODIS wavebands in addition to biophysical variables estimated from the MERIS and MODIS data. Percent tree cover maps were derived from MERIS and MODIS data for Seyhan upper Basin as final outputs. These final outputs consisted of spatially distributed estimates of percent tree cover at 300 m and 500 m spatial resolution and error estimates obtained through validation. This study showed that Envisat MERIS data can be used to predict percentage tree cover with greater spatial detail than using MODIS data. This finer-scale depiction should be of great utility for environmental monitoring purposes at the regional scale. © 2008 International Society for Photogrammetry and Remote Sensing. All rights reserved.