scholarly journals Evaluating Image Normalization via GANs for Environmental Mapping: A Case Study of Lichen Mapping Using High-Resolution Satellite Imagery

2021 ◽  
Vol 13 (24) ◽  
pp. 5035
Author(s):  
Shahab Jozdani ◽  
Dongmei Chen ◽  
Wenjun Chen ◽  
Sylvain G. Leblanc ◽  
Julie Lovitt ◽  
...  
Keyword(s):  
High Resolution ◽  
Auxiliary Information ◽  
Extreme Case ◽  
Training Data ◽  
Classification Model ◽  
Generative Adversarial Networks ◽  
Detection Accuracy ◽  
Large Area ◽  
Environmental Mapping ◽  
Sentinel 2

Illumination variations in non-atmospherically corrected high-resolution satellite (HRS) images acquired at different dates/times/locations pose a major challenge for large-area environmental mapping and monitoring. This problem is exacerbated in cases where a classification model is trained only on one image (and often limited training data) but applied to other scenes without collecting additional samples from these new images. In this research, by focusing on caribou lichen mapping, we evaluated the potential of using conditional Generative Adversarial Networks (cGANs) for the normalization of WorldView-2 (WV2) images of one area to a source WV2 image of another area on which a lichen detector model was trained. In this regard, we considered an extreme case where the classifier was not fine-tuned on the normalized images. We tested two main scenarios to normalize four target WV2 images to a source 50 cm pansharpened WV2 image: (1) normalizing based only on the WV2 panchromatic band, and (2) normalizing based on the WV2 panchromatic band and Sentinel-2 surface reflectance (SR) imagery. Our experiments showed that normalizing even based only on the WV2 panchromatic band led to a significant lichen-detection accuracy improvement compared to the use of original pansharpened target images. However, we found that conditioning the cGAN on both the WV2 panchromatic band and auxiliary information (in this case, Sentinel-2 SR imagery) further improved normalization and the subsequent classification results due to adding a more invariant source of information. Our experiments showed that, using only the panchromatic band, F1-score values ranged from 54% to 88%, while using the fused panchromatic and SR, F1-score values ranged from 75% to 91%.

scholarly journals Continental-Scale Land Cover Mapping at 10 m Resolution Over Europe (ELC10)

2021 ◽  
Vol 13 (12) ◽  
pp. 2301
Author(s):  
Zander Venter ◽  
Markus Sydenham
Keyword(s):  
Land Cover ◽  
Atmospheric Correction ◽  
Google Earth ◽  
Tree Planting ◽  
Training Data ◽  
Classification Model ◽  
Night Time ◽  
Continental Scale ◽  
Land Cover Maps ◽  
Sentinel 2

Land cover maps are important tools for quantifying the human footprint on the environment and facilitate reporting and accounting to international agreements addressing the Sustainable Development Goals. Widely used European land cover maps such as CORINE (Coordination of Information on the Environment) are produced at medium spatial resolutions (100 m) and rely on diverse data with complex workflows requiring significant institutional capacity. We present a 10 m resolution land cover map (ELC10) of Europe based on a satellite-driven machine learning workflow that is annually updatable. A random forest classification model was trained on 70K ground-truth points from the LUCAS (Land Use/Cover Area Frame Survey) dataset. Within the Google Earth Engine cloud computing environment, the ELC10 map can be generated from approx. 700 TB of Sentinel imagery within approx. 4 days from a single research user account. The map achieved an overall accuracy of 90% across eight land cover classes and could account for statistical unit land cover proportions within 3.9% (R2 = 0.83) of the actual value. These accuracies are higher than that of CORINE (100 m) and other 10 m land cover maps including S2GLC and FROM-GLC10. Spectro-temporal metrics that capture the phenology of land cover classes were most important in producing high mapping accuracies. We found that the atmospheric correction of Sentinel-2 and the speckle filtering of Sentinel-1 imagery had a minimal effect on enhancing the classification accuracy (< 1%). However, combining optical and radar imagery increased accuracy by 3% compared to Sentinel-2 alone and by 10% compared to Sentinel-1 alone. The addition of auxiliary data (terrain, climate and night-time lights) increased accuracy by an additional 2%. By using the centroid pixels from the LUCAS Copernicus module polygons we increased accuracy by <1%, revealing that random forests are robust against contaminated training data. Furthermore, the model requires very little training data to achieve moderate accuracies—the difference between 5K and 50K LUCAS points is only 3% (86 vs. 89%). This implies that significantly less resources are necessary for making in situ survey data (such as LUCAS) suitable for satellite-based land cover classification. At 10 m resolution, the ELC10 map can distinguish detailed landscape features like hedgerows and gardens, and therefore holds potential for aerial statistics at the city borough level and monitoring property-level environmental interventions (e.g., tree planting). Due to the reliance on purely satellite-based input data, the ELC10 map can be continuously updated independent of any country-specific geographic datasets.

scholarly journals MW-ACGAN: Generating Multiscale High-Resolution SAR Images for Ship Detection

Sensors ◽  
2020 ◽  
Vol 20 (22) ◽  
pp. 6673
Author(s):  
Lichuan Zou ◽  
Hong Zhang ◽  
Chao Wang ◽  
Fan Wu ◽  
Feng Gu
Keyword(s):  
High Resolution ◽  
Small Sample ◽  
Generative Adversarial Networks ◽  
Detection Accuracy ◽  
Data Set ◽  
Ship Detection ◽  
Multi Scale ◽  
Adversarial Networks ◽  
Sar Data ◽  
Composite Data

In high-resolution Synthetic Aperture Radar (SAR) ship detection, the number of SAR samples seriously affects the performance of the algorithms based on deep learning. In this paper, aiming at the application requirements of high-resolution ship detection in small samples, a high-resolution SAR ship detection method combining an improved sample generation network, Multiscale Wasserstein Auxiliary Classifier Generative Adversarial Networks (MW-ACGAN) and the Yolo v3 network is proposed. Firstly, the multi-scale Wasserstein distance and gradient penalty loss are used to improve the original Auxiliary Classifier Generative Adversarial Networks (ACGAN), so that the improved network can stably generate high-resolution SAR ship images. Secondly, the multi-scale loss term is added to the network, so the multi-scale image output layers are added, and multi-scale SAR ship images can be generated. Then, the original ship data set and the generated data are combined into a composite data set to train the Yolo v3 target detection network, so as to solve the problem of low detection accuracy under small sample data set. The experimental results of Gaofen-3 (GF-3) 3 m SAR data show that the MW-ACGAN network can generate multi-scale and multi-class ship slices, and the confidence level of ResNet18 is higher than that of ACGAN network, with an average score of 0.91. The detection results of Yolo v3 network model show that the detection accuracy trained by the composite data set is as high as 94%, which is far better than that trained only by the original SAR data set. These results show that our method can make the best use of the original data set, improve the accuracy of ship detection.

scholarly journals High-Resolution Urban Land Mapping in China from Sentinel 1A/2 Imagery Based on Google Earth Engine

2019 ◽  
Vol 11 (7) ◽  
pp. 752 ◽  
Author(s):  
Zhongchang Sun ◽  
Ru Xu ◽  
Wenjie Du ◽  
Lei Wang ◽  
Dengsheng Lu
Keyword(s):  
High Resolution ◽  
Vegetation Index ◽  
Urban Land ◽  
Google Earth ◽  
Economic Research ◽  
Optical Data ◽  
Large Area ◽  
Google Earth Engine ◽  
Sar Data ◽  
Sentinel 2

Accurate and timely urban land mapping is fundamental to supporting large area environmental and socio-economic research. Most of the available large-area urban land products are limited to a spatial resolution of 30 m. The fusion of optical and synthetic aperture radar (SAR) data for large-area high-resolution urban land mapping has not yet been widely explored. In this study, we propose a fast and effective urban land extraction method using ascending/descending orbits of Sentinel-1A SAR data and Sentinel-2 MSI (MultiSpectral Instrument, Level 1C) optical data acquired from 1 January 2015 to 30 June 2016. Potential urban land (PUL) was identified first through logical operations on yearly mean and standard deviation composites from a time series of ascending/descending orbits of SAR data. A Yearly Normalized Difference Vegetation Index (NDVI) maximum and modified Normalized Difference Water Index (MNDWI) mean composite were generated from Sentinel-2 imagery. The slope image derived from SRTM DEM data was used to mask mountain pixels and reduce the false positives in SAR data over these regions. We applied a region-specific threshold on PUL to extract the target urban land (TUL) and a global threshold on the MNDWI mean, and slope image to extract water bodies and high-slope regions. A majority filter with a three by three window was applied on previously extracted results and the main processing was carried out on the Google Earth Engine (GEE) platform. China was chosen as the testing region to validate the accuracy and robustness of our proposed method through 224,000 validation points randomly selected from high-resolution Google Earth imagery. Additionally, a total of 735 blocks with a size of 900 × 900 m were randomly selected and used to compare our product’s accuracy with the global human settlement layer (GHSL, 2014), GlobeLand30 (2010), and Liu (2015) products. Our method demonstrated the effectiveness of using a fusion of optical and SAR data for large area urban land extraction especially in areas where optical data fail to distinguish urban land from spectrally similar objects. Results show that the average overall, producer’s and user’s accuracies are 88.03%, 94.50% and 82.22%, respectively.

scholarly journals Super-Resolution of Sentinel-2 Images at 10m Resolution without Reference Images

2021 ◽  
Author(s):  
Yunhe Li ◽  
Bo Li
Keyword(s):  
High Resolution ◽  
Image Quality Assessment ◽  
Kernel Estimation ◽  
Super Resolution ◽  
Generative Adversarial Networks ◽  
Reference Image ◽  
Optical Remote Sensing ◽  
Learning Technology ◽  
Reference Images ◽  
Sentinel 2

Sentinel-2 can provide multi-spectral optical remote sensing images in RGBN bands with a spatial resolution of 10m, but the spatial details provided are not enough for many applications. WorldView can provide HR multi-spectral images less than 2m, but it is a commercial paid resource with relatively high usage costs. In this paper, without any available reference images, Sentinel-2 images at 10m resolution are improved to a resolution of 2.5m through super-resolution (SR) based on deep learning technology. Our model, named DKN-SR-GAN, uses degradation kernel estimation and noise injection to construct a dataset of near-natural low-high-resolution (LHR) image pairs, with only low-resolution (LR) images and no high-resolution (HR) prior information. DKN-SR-GAN uses the Generative Adversarial Networks (GAN) combined of ESRGAN-type generator, PatchGAN-type discriminator and the VGG-19-type feature extractor, using perceptual loss to optimize the network, so as to obtain SR images with clearer details and better perceptual effects. Experiments demonstrate that in the quantitative comparison of the non-reference image quality assessment (NR-IQA) metrics like NIQE, BRISQUE and PIQE, as well as the intuitive visual effects of the generated images, compared with state-of-the-art models such as EDSR8-RGB, RCAN and RS-ESRGAN, our proposed model has obvious advantages.

scholarly journals Comparative Research on Deep Learning Approaches for Airplane Detection from Very High-Resolution Satellite Images

2020 ◽  
Vol 12 (3) ◽  
pp. 458 ◽  
Author(s):  
Ugur Alganci ◽  
Mehmet Soydas ◽  
Elif Sertel
Keyword(s):  
Deep Learning ◽  
High Resolution ◽  
Object Detection ◽  
Satellite Images ◽  
Training Data ◽  
Object Localization ◽  
Detection Accuracy ◽  
Single Shot ◽  
High Resolution Satellite Images ◽  
Very High

Object detection from satellite images has been a challenging problem for many years. With the development of effective deep learning algorithms and advancement in hardware systems, higher accuracies have been achieved in the detection of various objects from very high-resolution (VHR) satellite images. This article provides a comparative evaluation of the state-of-the-art convolutional neural network (CNN)-based object detection models, which are Faster R-CNN, Single Shot Multi-box Detector (SSD), and You Look Only Once-v3 (YOLO-v3), to cope with the limited number of labeled data and to automatically detect airplanes in VHR satellite images. Data augmentation with rotation, rescaling, and cropping was applied on the test images to artificially increase the number of training data from satellite images. Moreover, a non-maximum suppression algorithm (NMS) was introduced at the end of the SSD and YOLO-v3 flows to get rid of the multiple detection occurrences near each detected object in the overlapping areas. The trained networks were applied to five independent VHR test images that cover airports and their surroundings to evaluate their performance objectively. Accuracy assessment results of the test regions proved that Faster R-CNN architecture provided the highest accuracy according to the F1 scores, average precision (AP) metrics, and visual inspection of the results. The YOLO-v3 ranked as second, with a slightly lower performance but providing a balanced trade-off between accuracy and speed. The SSD provided the lowest detection performance, but it was better in object localization. The results were also evaluated in terms of the object size and detection accuracy manner, which proved that large- and medium-sized airplanes were detected with higher accuracy.

scholarly journals Cascade Convolutional Neural Network Based on Transfer-Learning for Aircraft Detection on High-Resolution Remote Sensing Images

2017 ◽  
Vol 2017 ◽  
pp. 1-14 ◽  
Author(s):  
Bin Pan ◽  
Jianhao Tai ◽  
Qi Zheng ◽  
Shanshan Zhao
Keyword(s):  
Neural Network ◽  
Remote Sensing ◽  
High Resolution ◽  
Transfer Learning ◽  
High Accuracy ◽  
Detection Accuracy ◽  
Remote Sensing Images ◽  
Large Area ◽  
Detection Model ◽  
Aircraft Detection

Aircraft detection from high-resolution remote sensing images is important for civil and military applications. Recently, detection methods based on deep learning have rapidly advanced. However, they require numerous samples to train the detection model and cannot be directly used to efficiently handle large-area remote sensing images. A weakly supervised learning method (WSLM) can detect a target with few samples. However, it cannot extract an adequate number of features, and the detection accuracy requires improvement. We propose a cascade convolutional neural network (CCNN) framework based on transfer-learning and geometric feature constraints (GFC) for aircraft detection. It achieves high accuracy and efficient detection with relatively few samples. A high-accuracy detection model is first obtained using transfer-learning to fine-tune pretrained models with few samples. Then, a GFC region proposal filtering method improves detection efficiency. The CCNN framework completes the aircraft detection for large-area remote sensing images. The framework first-level network is an image classifier, which filters the entire image, excluding most areas with no aircraft. The second-level network is an object detector, which rapidly detects aircraft from the first-level network output. Compared with WSLM, detection accuracy increased by 3.66%, false detection decreased by 64%, and missed detection decreased by 23.1%.

scholarly journals Unsupervised Haze Removal for High-Resolution Optical Remote-Sensing Images Based on Improved Generative Adversarial Networks

2020 ◽  
Vol 12 (24) ◽  
pp. 4162
Author(s):  
Anna Hu ◽  
Zhong Xie ◽  
Yongyang Xu ◽  
Mingyu Xie ◽  
Liang Wu ◽  
...  
Keyword(s):  
Remote Sensing ◽  
High Resolution ◽  
Training Data ◽  
Generative Adversarial Networks ◽  
Optical Remote Sensing ◽  
Remote Sensing Images ◽  
Data Set ◽  
Adversarial Networks ◽  
Ground Object ◽  
Edge Sharpening

One major limitation of remote-sensing images is bad weather conditions, such as haze. Haze significantly reduces the accuracy of satellite image interpretation. To solve this problem, this paper proposes a novel unsupervised method to remove haze from high-resolution optical remote-sensing images. The proposed method, based on cycle generative adversarial networks, is called the edge-sharpening cycle-consistent adversarial network (ES-CCGAN). Most importantly, unlike existing methods, this approach does not require prior information; the training data are unsupervised, which mitigates the pressure of preparing the training data set. To enhance the ability to extract ground-object information, the generative network replaces a residual neural network (ResNet) with a dense convolutional network (DenseNet). The edge-sharpening loss function of the deep-learning model is designed to recover clear ground-object edges and obtain more detailed information from hazy images. In the high-frequency information extraction model, this study re-trained the Visual Geometry Group (VGG) network using remote-sensing images. Experimental results reveal that the proposed method can recover different kinds of scenes from hazy images successfully and obtain excellent color consistency. Moreover, the ability of the proposed method to obtain clear edges and rich texture feature information makes it superior to the existing methods.

Crowd, Crops and Machines: How crowdsourced annotations can help towards crops classification

2021 ◽  
Author(s):  
Alexis Guillot ◽  
Shaodi You ◽  
Hans van't Woud ◽  
Matthijs Perenboom ◽  
Amanda Kruijver ◽  
...  
Keyword(s):  
Neural Network ◽  
Random Forest ◽  
Convolutional Neural Network ◽  
Addis Ababa ◽  
Training Data ◽  
Generative Adversarial Networks ◽  
Temporal Dimension ◽  
Study Region ◽  
Crop Fields ◽  
Sentinel 2

&lt;p&gt;The use of artificial intelligence and specifically deep learning (DL) approaches in the domain of remote sensing is increasing. Such methods provide excellent results and show great potential for future applications. Earth observation sensors are able to deliver data with higher spatial, spectral and temporal resolutions. In this project, we use Sentinel-2 multispectral data and couple this input with a crowd annotated very high resolution (VHR) map which is generated in the video-game Cerberus, developed by the company BlackShore. In Cerberus, players are able to map features, like buildings, forest and specific types of crop fields, that are subsequently used as input for the Machine Learning (ML) pipeline. The ML pipeline is applied to classify crop fields in a larger region.&lt;/p&gt;&lt;p&gt;The main objective of this research is to study the accuracy of a model in detecting and describing the type of crop and whether the addition of a temporal dimension increases the accuracy. We will be experimenting with different methods that take their root in DL. The study region shown to Cerberus-players is Oromia in Ethiopia, south of the capital Addis Ababa. Using Sentinel-2 data, we aim to extend the generated maps to cover Ethiopia.&lt;/p&gt;&lt;p&gt;First, we will implement two DL methods; Random Forest (RF), and a 3D Convolutional Neural Network (CNN) that do not make use of the temporal dimension in order to have a baseline of the expected accuracy from a single multi-spectral image. Next, we will investigate four models that make use of time series: 1) a hybrid convolutional neural network-random forest (CNN-RF); 2) a 3D CNN that takes as input the output of a stack of 3D CNNs; 3) a model based on Recurrent Neural Networks (RNNs) performing pixel-based classification; and 4) an innovative method that combines the strength of RNNs, CNNs and Generative Adversarial Networks.&lt;span&gt;&amp;#160;&lt;/span&gt;&lt;/p&gt;&lt;p&gt;We are now implementing the methods and shall report on results at EGU April 2021. For future research, it could be a very interesting case to study the possibility of generalizing the combined approach of crowd annotated training data with extended classification over larger regions and generalizing to other areas.&lt;/p&gt;

scholarly journals An Operational Framework for Mapping Irrigated Areas at Plot Scale Using Sentinel-1 and Sentinel-2 Data

2021 ◽  
Vol 13 (13) ◽  
pp. 2584
Author(s):  
Hassan Bazzi ◽  
Nicolas Baghdadi ◽  
Ghaith Amin ◽  
Ibrahim Fayad ◽  
Mehrez Zribi ◽  
...  
Keyword(s):  
Time Series ◽  
Selection Criteria ◽  
Vegetation Index ◽  
Data Availability ◽  
Training Data ◽  
Classification Model ◽  
In Situ Data ◽  
Ndvi Time Series ◽  
Sentinel 2

In this study, we present an operational methodology for mapping irrigated areas at plot scale, which overcomes the limitation of terrain data availability, using Sentinel-1 (S1) C-band SAR (synthetic-aperture radar) and Sentinel-2 (S2) optical time series. The method was performed over a study site located near Orléans city of north-central France for four years (2017 until 2020). First, training data of irrigated and non-irrigated plots were selected using predefined selection criteria to obtain sufficient samples of irrigated and non-irrigated plots each year. The training data selection criteria is based on two irrigation metrics; the first one is a SAR-based metric derived from the S1 time series and the second is an optical-based metric derived from the NDVI (normalized difference vegetation index) time series of the S2 data. Using the newly developed irrigation event detection model (IEDM) applied for all S1 time series in VV (Vertical-Vertical) and VH (Vertical-Horizontal) polarizations, an irrigation weight metric was calculated for each plot. Using the NDVI time series, the maximum NDVI value achieved in the crop cycle was considered as a second selection metric. By fixing threshold values for both metrics, a dataset of irrigated and non-irrigated samples was constructed each year. Later, a random forest classifier (RF) was built for each year in order to map the summer agricultural plots into irrigated/non-irrigated. The irrigation classification model uses the S1 and NDVI time series calculated over the selected training plots. Finally, the proposed irrigation classifier was validated using real in situ data collected each year. The results show that, using the proposed classification procedure, the overall accuracy for the irrigation classification reaches 84.3%, 93.0%, 81.8%, and 72.8% for the years 2020, 2019, 2018, and 2017, respectively. The comparison between our proposed classification approach and the RF classifier built directly from in situ data showed that our approach reaches an accuracy nearly similar to that obtained using in situ RF classifiers with a difference in overall accuracy not exceeding 6.2%. The analysis of the obtained classification accuracies of the proposed method with precipitation data revealed that years with higher rainfall amounts during the summer crop-growing season (irrigation period) had lower overall accuracy (72.8% for 2017) whereas years encountering a drier summer had very good accuracy (93.0% for 2019).

scholarly journals Deep Neural Networks with Transfer Learning for Forest Variable Estimation Using Sentinel-2 Imagery in Boreal Forest

2021 ◽  
Vol 13 (12) ◽  
pp. 2392
Author(s):  
Heikki Astola ◽  
Lauri Seitsonen ◽  
Eelis Halme ◽  
Matthieu Molinier ◽  
Anne Lönnqvist
Keyword(s):  
Neural Networks ◽  
Transfer Learning ◽  
Forest Inventory ◽  
Deep Neural Networks ◽  
Training Data ◽  
Optimal Size ◽  
Large Area ◽  
Sampling Window ◽  
Hidden Layer ◽  
Sentinel 2

Estimation of forest structural variables is essential to provide relevant insights for public and private stakeholders in forestry and environmental sectors. Airborne light detection and ranging (LiDAR) enables accurate forest inventory, but it is expensive for large area analyses. Continuously increasing volume of open Earth Observation (EO) imagery from high-resolution (<30 m) satellites together with modern machine learning algorithms provide new prospects for spaceborne large area forest inventory. In this study, we investigated the capability of Sentinel-2 (S2) image and metadata, topography data, and canopy height model (CHM), as well as their combinations, to predict growing stock volume with deep neural networks (DNN) in four forestry districts in Central Finland. We focused on investigating the relevance of different input features, the effect of DNN depth, the amount of training data, and the size of image data sampling window to model prediction performance. We also studied model transfer between different silvicultural districts in Finland, with the objective to minimize the amount of new field data needed. We used forest inventory data provided by the Finnish Forest Centre for model training and performance evaluation. Leaving out CHM features, the model using RGB and NIR bands, the imaging and sun angles, and topography features as additional predictive variables obtained the best plot level accuracy (RMSE% = 42.6%, |BIAS%| = 0.8%). We found 3×3 pixels to be the optimal size for the sampling window, and two to three hidden layer DNNs to produce the best results with relatively small improvement to single hidden layer networks. Including CHM features with S2 data and additional features led to reduced relative RMSE (RMSE% = 28.6–30.7%) but increased the absolute value of relative bias (|BIAS%| = 0.9–4.0%). Transfer learning was found to be beneficial mainly with training data sets containing less than 250 field plots. The performance differences of DNN and random forest models were marginal. Our results contribute to improved structural variable estimation performance in boreal forests with the proposed image sampling and input feature concept.

Sign in / Sign up

Export Citation Format

Share Document