We propose a deep Convolutional Neural Network (CNN) for land cover mapping in remote sensing images, with focus on urban areas. In sensing, class imbalance represents often problem tasks like mapping, as small objects get less prioritised an effort to achieve the best overall accuracy. novel approach high accuracy, while still achieving good accuracy objects. Quantifying uncertainty pixel scale is another challenge especially when using CNNs. this paper we use recent advances measuring CNNs...

Abstract Oceans constitute over 70% of the earth's surface, and marine environment ecosystems are central to many global challenges. Not only oceans an important source food other resources, but they also play a roles in climate provide crucial ecosystem services. To monitor ensure sustainable exploitation extensive data collection analysis efforts form backbone management programmes on global, regional, or national levels. Technological advances sensor technology, autonomous platforms,...

Oil spill detection in SAR images operating a hybrid-polarimetric mode is examined. We propose and review several strategies for oil data. The retrieved measures are successfully applied to data covering experiments outside Norway the Deepwater Horizon incident Gulf of Mexico. It shown that, under assumption two-scale Bragg scattering model, coherence measure may be recovered equally well from data, as full-polarimetric that this directly measurements without need any additional assumptions....

Land cover classification of remote sensing images is a challenging task due to limited amounts annotated data, highly imbalanced classes, frequent incorrect pixel-level annotations, and an inherent complexity in the semantic segmentation task. In this article, we propose novel architecture called dense dilated convolutions' merging network (DDCM-Net) address The proposed DDCM-Net consists image convolutions merged with varying dilation rates. This effectively utilizes rich combinations that...

This article compares four new automatic methods to discriminate between spruce, pine and birch, which are the dominating tree species in Norwegian forests. Airborne laser scanning hyperspectral data were used. The was used mask pixels with low or no vegetation data. A green–blue ratio remove shadow areas from canopies, normalized difference index dead non-vegetation. best method pixel classification 160 spectral channels visible near-infrared spectrum, using a deep neural network. achieved...

We propose and investigate a method for creating large scale forest height maps at 10 m resolution from Sentinel-2 data using deep neural networks. In addition, we demonstrate how clear-cutting events can be detected in time series of the resulting maps. The network architecture is convolutional based on U-Net architecture. 13 spectral bands are resampled to spatial input U-Net, which outputs map with per-pixel estimates. trained ground truth acquired airborne lidar scanning surveys three...

Automatic urban land cover classification is a fundamental problem in remote sensing, e.g., for environmental monitoring. The highly challenging, as classes generally have high interclass and low intraclass variances. Techniques to improve performance sensing include fusion of data from different sensors with modalities. However, such techniques require all modalities be available the classifier decision-making process, i.e., at test time, well training. If modality missing current...

Abstract Acoustic target classification is the process of assigning observed acoustic backscattering intensity to an category. A deep learning strategy for using a convolutional network developed, consisting encoder and decoder, which allow use pixel information more abstract features. The can learn features directly from data, learned feature space may include both frequency response school morphology. We tested method on multifrequency data collected between 2007 2018 during Norwegian...

We propose a novel architecture called the Multi-view Self-Constructing Graph Convolutional Networks (MSCG- Net) for semantic segmentation. Building on recently proposed (SCG) module, which makes use of learnable latent variables to self-construct underlying graphs directly from input features without relying manually built prior knowledge graphs, we leverage multiple views in order explicitly exploit rotational invariance airborne images. further develop an adaptive class weighting loss...

In this paper, we propose an algorithm for automatic detection of seals in aerial remote sensing images using features extracted from a pre-trained deep convolutional neural network (CNN). The method consists three stages: (i) Detection potential objects, (ii) feature extraction and (iii) classification objects. first stage is application dependent, with the aim detecting all seal pups image, expense large amount false second extracts generic image local corresponding to each detected CNN...

When the orthogonal frequency-division multiplexing (OFDM) signal is appreciably wideband, as in underwater communications, a single Doppler-shift parameter inadequate to model Doppler effect, and rate required well. Such also accommodates nonsynchronized sampling of received signal. We establish conditions for identifiability parameters, show how they affect placement null subcarriers (NSC). derive maximum-likelihood (ML) estimator corresponding conditional Crame/spl acute/r-Rao lower...

The age determination of fish is fundamental to marine resource management. This task commonly done by analysis otoliths performed manually human experts. Otolith images from Greenland halibut acquired the Institute Marine Research (Norway) were recently used train a convolutional neural network (CNN) for automatically predicting age, opening way requiring less effort and availability expertise means deep learning (DL). In this study, we demonstrate that applying CNN model trained on one lab...

Detection of avalanches is critical for keeping avalanche inventories and management emergency situations. In this paper we propose a deep-learning based detection method SAR images. We utilize an existing proposing candidate regions, on change in images from multiple passes over the same area. Then convolutional neural network used to classify whether regions contain or not. The proposed methodology applies pre-trained that has been trained classification natural RGB represent non-standard...

We show how methods proposed in the statistical community dealing with missing data may be applied for land cover classification, where optical observations are due to clouds and snow. The method is divided into two stages: 1) cloud/snow classification 2) training classification. purpose of stage determine which pixels All each image classified classes cloud, snow, water, vegetation using a suitable classifier. as cloud or snow labeled missing, this information used subsequent stage, deals...

Age-reading of fish otoliths (ear stones) is important for the sustainable management resources. However, procedure challenging and requires experienced readers to carefully examine annual growth zones. In a recent study, convolutional neural networks (CNNs) have been demonstrated perform reasonably well on automatically predicting age from otolith images. present we investigate prediction rule learned by such provide insight into features that identify certain ranges. For this purpose,...

Multifrequency echosounder data can provide a broad understanding of the underwater environment in noninvasive manner. The analysis is, hence, topic great importance for marine ecosystem. Semantic segmentation, deep learning-based method predicting class attribute each acoustic intensity, has recently been spotlight fisheries and aquatic industry since its result be used to estimate abundance organisms. However, fundamental problem with current methods is massive reliance on availability...

Graph Neural Networks (GNNs) have received increasing attention in many fields. However, due to the lack of prior graphs, their use for semantic labeling has been limited. Here, we propose a novel architecture called Self-Constructing (SCG), which makes learnable latent variables generate embeddings and self-construct underlying graphs directly from input features without relying on manually built knowledge graphs. SCG can automatically obtain optimized non-local context complex-shaped...

Multi-modality data is becoming readily available in remote sensing (RS) and can provide complementary information about the Earth’s surface. Effective fusion of multi-modal thus important for various applications RS, but also very challenging due to large domain differences, noise, redundancies. There a lack effective scalable techniques bridging multiple modality encoders fully exploiting information. To this end, we propose new multi-modality network (MultiModNet) land cover mapping based...

In this paper, we consider hybrid-polarimetric synthetic aperture radar (SAR) data of ocean surface slicks, and hypothesize that can design a system is able to discriminate between mineral oil, plant clean sea. We focus particularly on challenges related set shift the training test data. SAR images surfaces, typically caused by variation wind level incident angles directly impact backscatter intensities. evaluate several classifiers, domain adaptation strategies, multilooking strategies....

Land-cover classification based on multi-temporal satellite images for scenarios where parts of the data are missing due to, example, clouds, snow or sensor failure has received little attention in remote-sensing literature. The goal this article is to introduce support vector machine (SVM) methods capable handling land-cover classification. novelty consists combining powerful SVM regularization framework with a recent statistical theory data, resulting new method an trained each pattern,...

Vehicle detection from very-high-resolution satellite imagery has received increasing interest during the last few years. In this article, we propose an automatic system for operational traffic monitoring using optical (0.5–0.6 m resolution) of small highways with low density and a range different illumination conditions, including cloud-shadowed, hazy, partially cloudy conditions. The proposed includes cloud shadow detection, road vehicle classification, counting. main part is which...

If population-wide improvements in nutrition and physical activity behavior are to be made, change interventions must use a variety of media. This study examines whether participation facilitator-based video version the Coronary Health Improvement Project could significantly reduce coronary risk. A total 28 classes conducted worksite, medical community settings were used teach 763 middle-aged adults, ages 30-79 years, about healthy lifestyles. Four 8 weeks after baseline, follow-up measures...