Detecting wild animals in satellite imagery is influenced by body size, background complexity and contrast between species and surrounding habitat. We used the TensorFlow Object Detection API [https://github.com/tensorflow/models/tree/master/research/object_detection] to build our model (Huang et al., 2017). camera trap images, aerial survey images and unmanned aerial vehicle (UAV) images (Bruijning et al., 2018; Chabot & Bird, 2015; Ferreira et al., 2020; Petersen et al., 2019; Torney et al., 2019; Weinstein, 2018). Now SIIS is the exclusive distributor of the satellite imagery obtained by KOMPSAT-2, KOMPSAT-3, KOMPSAT-3A and KOMPSAT-5 satellites. The last layer of the network is fully connected and performs classification (Schmidhuber, 2015). Pléiades (1A and 1B) operate in the same sun-synchronous orbit at 694 km with global coverage in 26 days and daily revisit to any place on Earth. segmentation of very high spatial resolution satellite image using active contours. A Convolutional Neural Network (CNN) is a deep learning artificial neural network architecture that has been extensively used for object detection and recognition in recent years. Neural networks models are commonly run as many times as time and availability of computational resources allow. Semantic Classification of Urban Trees Using Very High Resolution Satellite Imagery Abstract: There is an urgent need for urban tree classification, in order to assist with ecological environment protection and provide sustainable development guidance for urban planners. This technique is effective in homogeneous environments where species have strong spectral separability from background context. This is particularly relevant in cross‐border areas where multiple national permissions are required. temporal resolution) gibt die Wie-derholungsrate an, d.h. die Zeitdauer zwischen zwei Überflügen von einem Gebiet. Our vast archive includes imagery from all leading providers, such as DigitalGlobe, Airbus, and Satrec Imaging. 3D change detection from high and very high resolution satellite stereo imagery . The major unknown is what was behind the VHR requirement—why did the satellite have to be so powerful? In addition, an image from the Maasai Mara in Kenya was used to test the generalizability of the CNN for broader image conditions ‐ no additional training data were included for this test. Very high resolution satellite (VHRS) imaging sensors provide stereo images at up to 0.3m ground sampling distance (GSD). Several studies have relied on environmental proxies and indirect ecological signs of animal presence e.g. If satellite monitoring is applied at scale then developing methods to ensure standardized and occasional ground‐truthing will be required to ensure image interpretation is accurate (LaRue et al., 2017). Several studies have applied a form of supervised or semi‐supervised classification approaches to detect species in satellite imagery. If these point detections were inside true bounding boxes, they were counted as true positives. The last few years have seen an accelerated provision of very high resolution satellite data* by a number of providers in the commercial satellite data supply world. Very high resolution (VHR) Ikonos images were analysed to assess the state of activity of a diamond mine extraction site in Africa. A neural network is typically comprised of multiple layers connected by a set of learnable weights and biases (Romero et al., 2013). However, no research to date has detected animals in complex heterogeneous environments or detected elephants from space using very‐high‐resolution satellite imagery and deep learning. aus . No study has yet, to the best of our knowledge, detected species in complex heterogeneous landscapes from space. A CNN is a feed‐forward neural network designed to process large‐scale images by considering their local and global characteristics (LeCun et al., 2015). For example, in the case of the Emperor penguin, new colony locations were detected in a pan‐continental survey of the Antarctic coast (Abileah, 2002; Smit et al., 2017). The benefits of this monitoring technique are numerous; large spatial extents can be covered in short time periods making repeat surveys and reassessments possible at short intervals. Some of the more popular programs are listed below, recently followed by the European Union's Sentinel constellation. Through creation of a customized training dataset and application of a Convolutional Neural Network, we have automated the detection of elephants in satellite imagery with accuracy as high as human detection capabilities. Designed for civil and military users, the system is especially suited for emergency response and change detection. Several studies have shown that observers on aerial surveys often miscount due to fatigue and visibility issues resulting in over‐estimates (Caughley et al., 1976; Jachmann, 2002; Koneff et al., 2008). Various methods have been used to detect species in satellite imagery. We restricted the search to images that contain less than 20% cloud cover and acquired less than 25% off‐nadir (degree off centre of image captured). des Doktorgrades der Naturwissenschaften (Dr. rer. To test this, we use a population in Addo Elephant National Park, South Africa where herds move between open savannah habitat and closed heterogeneous woodland and thicket. Check Pages 1 - 6 of HIGH-RESOLUTION SATELLITE IMAGERY ANALYSIS BASED ON ... in the flip PDF version. The images were labelled by the volunteers using the VGG Annotation Tool [http://www.robots.ox.ac.uk/~vgg/software/via/] (Dutta & Zisserman, 2019). The bright morning light improves image clarity as elephants gather at water holes in the morning which makes them easy to identify (Figure 1). Satellites allow large-scale surveys to be conducted in short time periods with repeat surveys possible <24hrs. Moreover, they enable observations of parameters of our environment at greater broader spatial and finer temporal scales than those allowed through field observation alone. The technique benefits many industries and is widely used in farming software. The image archive for all WorldView 3 & 4 satellite images from Maxar Technologies (formerly DigitalGlobe) was searched via the Secure Watch Platform [https://www.digitalglobe.com/products/securewatch]. Not only is the detection accuracy we achieve for elephants as high as that of humans, but there is less variation in the consistency of detection for the CNN compared to human detection performance (as shown in Figure 6). very high resolution satellite imagery - Deutsch-Übersetzung – Linguee Wörterbuch It was established in April 2014 as a subsidiary of Satrec Initiative (SI) and started operations as a satellite image and service provider. I.D. This compares with the detection accuracy of the human labels with an averaged F2 score 0.77 in heterogeneous areas and 0.80 in homogenous areas. Elephant calves were accurately detected, despite their absence in the training dataset. Dissertation . Human volunteer annotators provide point detections for elephants. Please check your email for instructions on resetting your password. Very High Resolution (50cm) Satellite. These techniques can now leverage massive image datasets e.g. In this study, we apply a Convolution Neural Network (CNN) model to automatically detect and count African elephants in a woodland savanna ecosystem in South Africa. This API provides implementations of different deep learning object detection algorithms. Each participant was provided with a detailed training sheet and an example of how elephants look in satellite images prior to labelling. For a number of species remote sensing via satellite imagery is already a viable monitoring technique. Very high-resolution Earth observation satellite imagery provides a critical baseline for creating effective GIS strategy and the foundation for a modern National Spatial Data Infrastructure. Specifically, we produce the first high-resolution wetland map of the City of St. John's using advanced machine learning algorithms, very high-resolution satellite imagery, and airborne LiDAR. Our results show that the CNN performs with high accuracy, comparable to human detection capabilities. Concern about climate change, biodiversity loss, habitat degradation and landscape change: Embedded in different packages of environmental concern? Inventorying (un)built areas based on very high resolution satellite imagery. Imagery at high zoom levels is provided by Microsoft. View more >. We compare the accuracy of detections from the human volunteer annotators and CNN against our count which we deem as the baseline i.e. Using camera traps to study the age–sex structure and behaviour of crop‐using elephants Loxodonta africana in Udzungwa Mountains National Park, Tanzania, Accelerometers and simple algorithms identify activity budgets and body orientation in African elephants Loxodonta africana, Polar bears from space: assessing satellite imagery as a tool to track Arctic wildlife, Performance and Scalability of GPU‐Based Convolutional Neural Networks, Inception‐v4, Inception‐ResNet and the Impact of Residual Connections on Learning, Spatial and temporal changes in group dynamics and range use enable anti‐predator responses in African buffalo, Mapping inundation extent, frequency and duration in the Okavango Delta from 2001 to 2012, A comparison of deep learning and citizen science techniques for counting wildlife in aerial survey images, Autonomous Tracking of Sea Turtles based on Multibeam Imaging Sonar: Toward Robotic Observation of Marine Life, Rising temperatures and changing rainfall patterns in South Africa's national parks, A Quickbird’s eye view on marmots, in International Institute for Geo‐information science and Earth Observation, Aimed Object‐throwing by a Wild African Elephant in an interspecific encounter, Identifying animal species in camera trap images using deep learning and citizen science, Benchmarking of data fusion algorithms in support of earth observation based Antarctic wildlife monitoring, Automatic Counting of Large Mammals from Very‐high Resolution Panchromatic Satellite Imagery, Spotting East African mammals in open savannah from space, https://www.digitalglobe.com/products/securewatch, http://www.robots.ox.ac.uk/~vgg/software/via/, https://github.com/tensorflow/models/tree/master/research/object_detection. nat.) Many wildlife species are under threat across their geographical range as we are currently undergoing the sixth‐mass extinction (Barnosky et al., 2014; Cardinale, 2012; Skogen et al., 2018). A detection process that would formally have taken weeks can thus be completed in a matter of hours. Satellite remote sensing has recently emerged as a new viable monitoring technique for detecting wildlife. The sensor delivers panchromatic as well as four-band multispectral imagery with a resolution of 1,0 m/pxl. GF-1 satellite is the first low earth orbit remote sensing satellite of China’s high-resolution earth observation system, which breaks through the key technologies of optical remote sensing for high spatial resolution and multispectral and wide coverage. Zooniverse [https://www.zooniverse.org/], Amazon Mechanical Turk [https://www.mturk.com/] can help in the creation of these bespoke training datasets using the ‘Wisdom of the crowd’ (Kao et al., 2018; Mierswa, 2016). EOS.com uses cookies which are necessary for this site to operate properly, and some of which are used for improving your experience with us. Simply enter any location or even your address to view easy satellite & aerial maps. Twenty-First Century Aerospace Technology Co. Ltd (21AT), Landsat 8 Bands: Combinations For Satellite Images, High-Resolution Images Are Close As Never Before, Tips For Improving Satellite Imagery Search, How To Get High-Resolution Satellite Images, Free Satellite Imagery Sources: Zoom In Our Planet. The test dataset used to test the CNN against human annotator performance contains 164 elephants across seven different satellite images. Satellite image (c) 2020 Maxar Technologies, Example of CNN detections in Maasai Mara, Kenya from Geoeye‐1 Satellite. Figure 5 provides visualization of some example CNN detections. High-resolution satellite imagery tracks the changing human footprint across the globe, including rapidly growing cities, urban sprawl, and informal settlements. Since 2015 satellites produced by SSTL in the UK are part of this family, with the launch of 3 identical DMC3 satellites providing multispectral images with a panchromatic channel better than 1 metre spatial resolution. While the majority of such work has used medium-to low-spatial resolution imagery, increasing availability of high- and very high-resolution (VHR) imagery has resulted in wetland maps being produced at high levels of spatial and qualitative detail. led the writing of the manuscript with input from O.I., S.R., D.M. The boxplots show the human results (circles represent outliers); lines are results from the four different CNN models, Remote Sensing in Ecology and Conservation, I have read and accept the Wiley Online Library Terms and Conditions of Use, Marine mammal census using space satellite imagery, Estimating the relative abundance of emperor penguins at inaccessible colonies using satellite imagery, Estimating forest elephant numbers with dung counts and a geographic information system, Introducing the Scientific Consensus on Maintaining Humanity’s Life Support Systems in the 21st Century: Information for Policy Makers, Aerial‐trained deep learning networks for surveying cetaceans from satellite imagery, Using deep learning to count albatrosses from space: Assessing results in light of ground truth uncertainty, Detecting Wildlife in Unmanned Aerial Systems Imagery Using Convolutional Neural Networks Trained with an Automated Feedback Loop, trackdem: Automated particle tracking to obtain population counts and size distributions from videos in r, Multispectral image‐based estimation of drought patterns and intensity around Lake Chad, Africa, An Improved Faster R‐CNN for Small Object Detection. Big variety of data and moderate pricing policy makes Airbus imagery irreplaceable for the following purposes: Fields analytics based on high-resolution satellite images to track all the changes on-the-spot! Now, you are able to order a night shooting of the city from Chang Guang Satellite, that delivers not only hyperspectral satellite imagery (0.92 m/pxl), but also a unique product for GIS market – night-time optical imagery with a resolution of 0.92 m/pxl along with a night-time video from space. These training images contain 1270 elephant labels of which 1125 are unique elephants. Training a CNN requires images to be split into training, validation and test sets. Images can be acquired via targeted satellite tasking on specific days; however, this is more costly than using archive imagery. Worldview‐3 costs $17.50 per km2 for archive imagery and tasking new imagery costs $27.50 per km2, with a minimum order of 100 km2(2020 pricing). Aerial surveys are conducted either as total counts – flying closely spaced transects, or sample counts, covering 5‐20% of an area and extrapolating to a larger area. Different areas of the park have been sectioned off for conservation purposes‐ elephants were identified in the Main Camp section of the park surrounding Hapoor Dam (Figure 1).The Main Camp is a combination of dense shrubland and low forest e.g.porkbush (Portulcariaafra), White milkwood (Sideroxyloninerme), Cape leadwort(Plumbago auriculate) and open grassland (Kakembo et al., 2015; Tambling et al., 2012). HIGH-RESOLUTION SATELLITE IMAGERY ANALYSIS BASED ON ... was published by on 2015-06-04. To the best of our knowledge, only three studies have applied a CNN to satellite imagery in the case of albatross (Bowler et al., 2019), whales (Borowicz et al., 2019; Guirado et al., 2019) and pack‐ice seals (Gonçalves et al., 2020). Over six hundred elephants move between these habitats (Du Toit & O’Connor, 2014; Wilgen et al., 2016). Areas of future research to expand this technique include testing whether performance improvements for detecting elephants can be achieved by including the near infrared band and testing to discover for which other species this is already a viable monitoring technique. A new high-resolution sensor offers monitoring of any place on Earth revealing details as small as 80 centimetres. An object-based random forest algorithm is applied to features extracted from WorldView-4, GeoEye-1, and LiDAR data to characterize five wetland classes, namely bog, fen, marsh, swamp, and open water, … des Fachbereichs Mathematik / Informatik . der Universität Osnabrück . In this paper we present re- sults using TerraSAR-X and Radarsat-2 data for reliable and robust glacial lake mapping over test sites in the Alps, Pamir and Himalaya. We tested several pan‐sharpening algorithms using visual inspection method‐ the Gram‐Schmidt pan‐sharpening algorithm provided the cleanest visual result in terms of spectral and spatial fidelity and was applied to all images. e.g. The whole constellation is planned to be completed by 2022. Contact Also, the satellite was enhanced with an on-board High – Resolution Stereoscopic viewing instrument to obtain large-area along-track stereoscopic panchromatic imagery with high altimetric accuracy (5-10 m relative and 10-15 m absolute). The baseline we deem as the true number of elephants is a labelled dataset doubled screened by two annotators – an Ecologist and Machine Learning Scientist. In this study, we investigate the feasibility of using very‐high‐resolution satellite imagery to detect wildlife species in heterogenous environments with deep learning. Designed as a dual civil/military system, Pléiades will meet the space imagery requirements of European defence as well as civil and commercial needs. The test subimages cover both heterogeneous and homogeneous areas of the park from different seasons and years (see Table 1). In this context, very high resolution (VHR) satellite-based optical, infrared, and radar imaging instruments provide reliable information to implement spatially-based conservation actions. This image allowed us to test the generalizability of our algorithm to a different environment and satellite. ran the code in Tensorflow with input from S.R. High Resolution Maps. Benefits for situation awareness, tracing changes and prognosis. Easy To calculate the deviation from this baseline we generate an F2 score. Zoom into recent high-resolution maps. European Space Imaging partnered with reseller Geo4i to investigate the current ongoing Libyan Civil War. The imagery collected serves multiple purposes: Airbus Defense and Space is the second largest space company and one of the biggest manufacturers of satellites for Earth observation, navigation and telecommunication purposes. Addo Elephant National Park in South Africa was chosen as the study site. Satellite image (c) 2020 Maxar Technologies, F2 score obtained by each of the four models considered over training steps on the validation dataset. SPOT 6 and SPOT 7 satellites were launched to continue the SPOT 5 mission.