Search

International audience; These last decades, Earth Observation brought a number of new perspectives from geosciences to human activity monitoring. As more data became available, Artificial Intelligence (AI) techniques led to very successful results for understanding remote sensing data. Moreover, various acquisition techniques such as Synthetic Aperture Radar (SAR) can also be used for problems that could not be tackled only through optical images. This is the case for weather-related disasters such as floods or hurricanes, which are generally associated with large clouds cover. Yet, machine learning on SAR data is still considered challenging due to the lack of available labeled data. To help the community go forward, we introduce a new dataset composed of co-registered optical and SAR images time series for the detection of flood events and new neural network approaches to leverage these two modalities. Figure 1. The SEN12-FLOOD dataset is composed of SAR and optical time series in which a flood event may occur.

Good quality reconstruction and comprehension of a scene rely on 3D estimation methods. The 3D information was usually obtained from images by stereo-photogrammetry, but deep learning has recently provided us with excellent results for monocular depth estimation. Building up a sufficiently large and rich training dataset to achieve these results requires onerous processing. In this paper, we address the problem of learning outdoor 3D point cloud from monocular data using a sparse ground-truth dataset. We propose Pix2Point, a deep learning-based approach for monocular 3D point cloud prediction, able to deal with complete and challenging outdoor scenes. Our method relies on a 2D-3D hybrid neural network architecture, and a supervised end-to-end minimisation of an optimal transport divergence between point clouds. We show that, when trained on sparse point clouds, our simple promising approach achieves a better coverage of 3D outdoor scenes than efficient monocular depth methods., 5 pages, 2 figures, to be published in 2021 International Conference on Machine Vision Applications

In this work, we present the complete workflow used to acquire a large hyperspectral dataset on a western Africa historical hydrocarbon production site, and its processing. Our goal is to study how state-of-the-art hyperspectral processing techniques can help detect hydrocarbon bearing soil either of natural origin or accidental by monitoring the health of the vegetation for exploration or environmental monitoring purposes. We present our complete workflow, from acquisition, atmospheric correction, image annotation and classification using modern machine learning techniques, and show how state-of-the-art research can be applied to real-world use cases.

International audience; This paper explores different aspects of semantic segmentation of remote sensing data using deep neural networks. Learning with deep neural networks was revolutionized by the creation of ImageNet. Remote sensing benefited of these new techniques, however Earth Observation (EO) datasets remain small in comparison. In this work, we investigate how we can progress towards the ImageNet of remote sensing. In particular, two questions are addressed in this paper. First, how robust are existing supervised learning strategies with respect to data volume? Second, which properties are expected from a large-scale EO dataset? The main contributions of this work are: (i) a strong robustness analysis of existing supervised learning strategies with respect to remote sensing data, (ii) the introduction of a new, large-scale dataset named MiniFrance.

International audience; This paper deals with classification of remote sensing data to extract objects for industrial mapping. While land-cover or urban mapping have been extensively studied, industrial cartography remains a field yet to explore, in spite of tremendous needs. We present and compare here four approaches for railway detection in very high resolution images. They use various kind of filtering approaches, including the trained filters of fully convolutional networks. Moreover, they benefit from different a-priori and post-processing techniques to make them more robust. We evaluate all approaches on a challenging dataset captured on an operating station site with complex objects.

Remote sensing technology is a remarkable tool to explore and to measure Earth’s surface features. Total and ONERA set up a collaborative partnership named New Advanced Observation Method Integration (NAOMI) that aims at adapting and developing new remote sensing techniques specifically targeted for hydrocarbons exploration and environmental protection. In this context, we integrate deep learning for classification of hyperspectral data. To detect different land uses and materials in aerial hyperspectral images, neural networks prove themselves to be very efficient tools, as they are able to learn discriminant features that help classification performance.

In this paper, the scientific outcomes of the 2016 Data Fusion Contest organized by the Image Analysis and Data Fusion Technical Committee of the IEEE Geoscience and Remote Sensing Society are discussed. The 2016 Contest was an open topic competition based on a multitemporal and multimodal dataset, which included a temporal pair of very high resolution panchromatic and multispectral Deimos-2 images and a video captured by the Iris camera on-board the International Space Station. The problems addressed and the techniques proposed by the participants to the Contest spanned across a rather broad range of topics, and mixed ideas and methodologies from the remote sensing, video processing, and computer vision. In particular, the winning team developed a deep learning method to jointly address spatial scene labeling and temporal activity modeling using the available image and video data. The second place team proposed a random field model to simultaneously perform coregistration of multitemporal data, semantic segmentation, and change detection. The methodological key ideas of both these approaches and the main results of the corresponding experimental validation are discussed in this paper.

Recently, micro-rotation confocal microscopy has enabled the acquisition of a sequence of slices of non adherent living cells obtained during a partially controlled rotation movement of the cell through the focal plane. Although we are now able to estimate the 3D position of every slice with respect to the frame, the reconstruction of the cell from the positioned slices remains a problem that this paper address. In our context, 3D spatially-varying PSF and missing data are the two main particularities of this problem. Experiments illustrate the ability of the classical EM algorithm to deconvolve efficiently cell volume and also to deal with missing data.

This paper investigates a new time domain channel estimation for MIMO OFDMA downlink systems. Compared to classical frequency domain channel estimation, time approach allows improvements in a multi-antenna system. Nevertheless, the presence of null carriers in the spectrum leads to performance degradation because of ”border effect” phenomenon. An improved time domain channel estimation for MIMO-OFDMA system is proposed for downlink transmission and compared to classical channel estimation methods. Besides, this channel estimation process can be applied in any SISO and/or MIMO multicarrier transmission.

Multiple-input multiple-output OFDM systems receive a great deal of attention due to their great potential in achieving high data rates in wireless communications. Nevertheless, performance of MIMO-OFDM systems can severely degrade due to inaccurate channel estimation especially over frequency-selective fast-varying channels. In this paper, a new efficient iterative channel estimation process based on linear regression algorithm is proposed in order to iteratively improve calculation of channel coefficients during data stage by taking into account multi-antenna interference. The efficiency of the proposed solution is demonstrated for non-orthogonal MIMO schemes over frequency selective fast-time varying channels thanks to comparison with other turbo-channel estimation techniques.

This paper presents a comparison between coherent and non-coherent multi-antenna systems for frequency and time selective channel. Performance of a coherent system mainly depends on channel estimation at the receiver. This estimation generally involves insertion of pilot symbols in a frame and thus a loss of spectral efficiency. Differential space time modulation (DSTM), which is an extension of SISO differential techniques to MIMO case, seems to be an alternative solution to attain high spectral efficiency and good performance in fast time-varying channels, where estimation of channels may be difficult. When comparisons between DSTM and coherent system are provided, perfect channel knowledge at the receiver is generally assumed, In this paper, we propose to include realistic channel estimation for coherent system as well as degradation of spectral efficiency used due to necessary pilot insertion. Simulation results show that in very fast time-varying channel non-coherent systems outperform systems with channel estimation. Besides, performance of several non-coherent receivers are provided for different channels and with channel coding

To allow efficient browsing of large image collection, we have to provide a summary of its visual content. We present in this paper a new robust approach to categorize image databases: Adaptive Robust Competition with Proximity-Based Merging (ARC-M). This algorithm relies on a non-supervised database categorization, coupled with a selection of prototypes in each resulting category. Each image is represented by a high-dimensional vector in the feature space. A principal component analysis is performed for every feature to reduce dimensionality. Then, clustering is performed in challenging conditions by minimizing a Competitive Agglomeration objective function with an extra noise cluster to collect outliers. Agglomeration is improved by a merging process based on cluster proximity verification.

Content-based image retrieval can be dramatically improved by providing a good initial database overview to the user. To address this issue, we present in this paper an adaptive robust competition. This algorithm relies on a non-supervised database categorization, coupled with a selection of prototypes in each resulting category. In our approach, each image is represented by a high-dimensional signature in the feature space, and a principal component analysis is performed for every feature to reduce dimensionality. Image database overview is computed in challenging conditions since clusters are overlapping with outliers and the number of clusters is unknown.