Search

International audience; Hyperspectral imagery classification is a challenging task due to the large number of spectral bands, and low number of labeled samples. To overcome these issues, we propose a novel approach for hyperspectral image classification based on feature selection and semi-supervised hypergraph convolutional network working with small number of labeled samples. Firstly, we propose a new unsupervised feature selection method based on an information theoretic criterion. Relevant spectral features are automatically selected while preserving the physical properties of hyperspectral data. Secondly, we construct a spectro-spatial hypergraph in order to represent the complex relationships between pixels. Finally, we propose a semi-supervised hypergraph convolutional network which integrates local vertex features and hypergraph topology in the convolutional layers. The aim of this step is to preserve the spectro-spatial features and to cope with the high correlation between hypernodes during classification. The main advantage of the proposed approach is to allow the automatic selection of relevant spectral bands while preserving the spatial and spectral features. In addition, by accounting for the relationship between pixels leads to improved classification results even when the number of labeled samples is low. Experiments are conducted on two real hyperspectral images show that the proposed approach reaches competitive good performances, and achieves better classification performances compared to state-of-the-art methods.

International audience; The obsession with how the brain and behavior are related is a challenge for cognitive neuroscience research, for which functional magnetic resonance imaging (fMRI) has significantly improved our understanding of brain functions and dysfunctions. In this paper, we propose a novel multi-modal spatial cerebral graph based on an attention mechanism called MSCGATE that combines both fMRI modalities: task-, and rest-fMRI based on spatial and cerebral features to preserve the rich complex structure between brain voxels. Moreover, it attempts to project the structural-functional brain connections into a new multi-modal latent representation space, which will subsequently be inputted to our trace regression predictive model to output each subject’s behavioral score. Experiments on the InterTVA dataset reveal that our proposed approach outperforms other graph representation learning-based models, in terms of effectiveness and performance.

International audience; The classification of hyperspectral image is a challenging task due to the high dimensional space, with large number of spectral bands, and low number of labeled training samples. To overcome these challenges, we propose a novel methodology for hyperspectral image classification based on multi-view deep neural networks which fuses both spectral and spatial features by using only a small number of labeled samples. Firstly, we process the initial hyperspectral image in order to extract a set of spectral and spatial features. Each spectral vector is the spectral signature of each pixel of the image. The spatial features are extracted using a simple deep autoencoder, which seeks to reduce the high dimensionality of data taking into account the neighborhood region for each pixel. Secondly, we propose a multi-view deep autoencoder model which allows fusing the spectral and spatial features extracted from the hyperspectral image into a joint latent representation space. Finally, a semi-supervised graph convolutional network is trained based on thee fused latent representation space to perform the hyperspectral image classification. The main advantage of the proposed approach is to allow the automatic extraction of relevant information while preserving the spatial and spectral features of data, and improve the classification of hyperspectral images even when the number of labeled samples is low. Experiments are conducted on three real hyperspectral images respectively Indian Pines, Salinas, and Pavia University datasets. Results show that the proposed approach is competitive in classification performances compared to state-of-the-art.

International audience; In this research study, we deal with remote sensing data analysis over high dimensional space formed by hyperspectral images. This task is generally complex due to the large spectral, spatial richness, and mixed pixels. Thus, several spectral un-mixing methods have been proposed to discriminate mixing spectra by estimating the classes and their presence rates. However, information related to mixed pixel composition is very interesting for some applications, but it is insufficient for many others. Thus, it is necessary to have much more data about the spatial localization of the classes detected during the spectral un-mixing process. To solve the above-mentioned problem and specify the spatial location of the different land cover classes in the mixed pixel, sub-pixel mapping techniques were introduced. This manuscript presents a novel sub-pixel mapping process relying on K-SVD (K-singular value decomposition) learning and total variation as a spatial regularization parameter (SMKSVD-TV: Sub-pixel Mapping based on K-SVD dictionary learning and Total Variation). The proposed approach adopts total variation as a spatial regularization parameter, to make edges smooth, and a pre-constructed spatial dictionary with the K-SVD dictionary training algorithm to have more spatial configurations at the sub-pixel level. It was tested and validated with three real hyperspectral data. The experimental results reveal that the attributes obtained by utilizing a learned spatial dictionary with isotropic total variation allowed improving the classes sub-pixel spatial localization, while taking into account pre-learned spatial patterns. It is also clear that the K-SVD dictionary learning algorithm can be applied to construct a spatial dictionary, particularly for each data set.

International audience; This paper proposes a novel approach based on adaptive dimensionality reduction (ADR) and a semi-supervised 3-D convolutional neural network (3-D CNN) for the spectro-spatial classification of hyperspectral images (HSIs). It tackles the problem of curse of dimensionality and the limited number of training samples by selecting the most relevant spectral bands. The selected bands should be informative, discriminative and distinctive. They are fed into a semi-supervised 3-D CNN feature extractor, then a linear regression classifier to produce the classification map. In fact, the proposed semi-supervised 3-D CNN model seeks to extract the deep spectral and spatial features based on convolutional encoder-decoder to enhance the HSI classification. It uses several 3-D convolution and max-pooling layers to extract these features from the selected relevant bands. The main advantage of the proposed approach is to reduce the high dimensionality of HSI, preserve the relevant spectro-spatial information and enhance the classification using few labeled training samples. Experimental studies are carried out on three real HSI data sets: Indian Pines, Pavia University, and Salinas. The obtained results show that the proposed approach performs better than other deep learning-based methods including CNN-based methods, and significantly improves the classification accuracy of HSIs.

In this paper, we propose a deep learning model based on deep multi-view representation learning, to address the video object segmentation task. The proposed model emphasizes the importance of the inherent correlation between video frames and incorporates a multi-view representation learning based on deep canonically correlated autoencoders. The multi-view representation learning in our model provides an efficient mechanism for capturing inherent correlations by jointly extracting useful features and learning better representation into a joint feature space, i.e., shared representation. To increase the training data and the learning capacity, we train the proposed model with pairs of video frames, i.e., Fa and Fb. During the segmentation phase, the deep canonically correlated auto encoders model encodes useful features by processing multiple reference frames together, which is used to detect the frequently reappearing. Our model enhances the state-of-the-art deep learning-based methods that mainly focus on learning discriminative foreground representations over appearance and motion. Experimental results over two large benchmarks demonstrate the ability of the proposed method to outperform competitive approaches and to reach good performances, in terms of semantic segmentation.

Recently, classification and dimensionality reduction (DR) have become important issues of hyperspectral image (HSI) analysis. Especially, HSI classification is a challenging task due to the high-dimensional feature space, with a large number of spectral bands, and a low number of labeled samples. In this paper, we propose a new HSI classification approach, which is called fused 3-D spectral-spatial deep neural networks for hyperspectral image classification. We propose an unsupervised band selection method to avoid the problem of redundancy between spectral bands and automatically find a set of groups Ck each one containing similar spectral bands. Moreover, the model uses the different groups of selected bands to extract spectral-spatial features in order to improve the classification rate. Each group is associated with a 3-D CNN model, which are then fused to improve the precision of classification. The main advantage of the proposed method is to keep the initial spectral-spatial features by automatically selecting relevant spectral bands, which improves the classification of HSI using a low number of labeled samples. Experiments on two real HSIs, Indian Pines and Salinas datasets, are performed to demonstrate the effectiveness of the proposed method. Results show that the proposed method reaches competitive good performances, and achieves better classification rates compared to various state-of-the-art techniques.

International audience; In this paper, we propose a novel adaptive band selection approach for hyperspectral image semantic interpretation. This approach is based on constrained band selection (CBS) method and extracted knowledge coming from tensor locality preserving projection. The extracted knowledge is presented as a set of rules which are used to evaluate the importance of spectral bands for classes discrimination. Based on these extracted rules and the CBS approach, relevant bands are selected to enhance the hyperspectral image semantic interpretation. The main advantage of the proposed adaptive band selection approach is to allow the automatic selection of discriminant, distinctive and informative spectral bands, and improve the semantic interpretation of hyperspectral images. Experimental results on real images show that the proposed band selection approach reaches competitive good performances, in terms of classification accuracy. Hyperspectral Imagery Semantic Interpretation Based on Adaptive Constrained Band... | Request PDF. Available from: https://www.researchgate.net/publication/323194459_Hyperspectral_Imagery_Semantic_Interpretation_Based_on_Adaptive_Constrained_Band_Selection_and_Knowledge_Extraction_Techniques [accessed Feb 16 2018].

The classification of hyperspectral images (HSI) is a challenging task due to the imbalance between the high dimensionality of spectral features, i.e., the large number of spectral bands and the scarcity of labeled training samples. Moreover, the curse of dimensionality problem deteriorates the classification rate and increases computational complexity. To alleviate these issues, we propose in this paper a novel approach based on deep Restricted Boltzmann Machine (RBM), which improves the spectro-spatial classification of HSI by extracting meaningful features, i.e., finding a better representation of hyperspectral samples. The proposed approach can be divided into three phases; i) spectro-spatial graph construction, ii) deep feature extraction, and iii) spectro-spatial classification. To fully exploit the inherent spatial distribution of the HSI and preserve the spectro-spatial features, the joint similarity measurement encoding both the spectral and spatial features is designed for graph construction. By using the spectro-spatial graph, the proposed RBM can ultimately learn and discriminative representation of hyperspectral samples in the hidden layer. Finally, the extracted deep features vectors are feed as input to Deep Belief Network (DBN) and logistic regression (LR) for the classification. The main advantage of the proposed approach is to learn a better representation of HSI, preserve the deep spectro-spatial features and improve the classification accuracies. Experiments are conducted on two real HSI, i.e., Indian Pines, and Pavia University. The obtained results show that the proposed approach achieves better classification performances compared to other state-of-the-art approaches.

Various Internet solutions take their power processing and analysis from cloud computing services. Internet of Things (IoT) applications started discovering the benefits of computing, processing, and analysis on the device itself aiming to reduce latency for time-critical applications. However, on-device processing is not suitable for resource-constraints IoT devices. Edge computing (EC) came as an alternative solution that tends to move services and computation more closer to consumers, at the edge. In this letter, we study and discuss the applicability of merging deep learning (DL) models, i.e., convolutional neural network (CNN), recurrent neural network (RNN), and reinforcement learning (RL), with IoT and information-centric networking which is a promising future Internet architecture, combined all together with the EC concept. Therefore, a CNN model can be used in the IoT area to exploit reliably data from a complex environment. Moreover, RL and RNN have been recently integrated into IoT, which can be used to take the multi-modality of data in real-time applications into account. 2019 IEEE. Manuscript received August 20, 2018; revised September 29, 2018; accepted October 29, 2018. Date of publication October 15, 2018; date of current version January 8, 2019. The work of S. Luo was supported by the National 242 Project under Grant No. 2017A149. The associate editor coordinating the review of this paper and approving it for publication was O. Popescu. (Corresponding author: Senlin Luo.) H. Khelifi, S. Luo, and B. Nour are with the Beijing Institute of Technology, Beijing 100081, China (e-mail: hakima@bit.edu.cn; luosenlin@bit.edu.cn; n.boubakr@bit.edu.cn). Scopus 2-s2.0-85055018213

International audience; Vehicular Ad Hoc Network has attracted both research and industrial community due to its benefits in facilitating human life and enhancing the security and comfort. However, various issues have been faced in such networks such as information security, routing reliability, dynamic high mobility of vehicles, that influence the stability of communication. To overcome this issue, it is necessary to increase the routing protocols performances, by keeping only the stable path during the communication. The effective solutions that have been investigated in the literature are based on the link prediction to avoid broken links. In this paper, we propose a new solution based on machine learning concept for link prediction, using LR and Support Vector Regression (SVR) which is a variant of the Support Vector Machine (SVM) algorithm. SVR allows predicting the movements of the vehicles in the network which gives us a decision for the link state at a future time. We study the performance of SVR by comparing the generated prediction values against real movement traces of different vehicles in various mobility scenarios, and to show the effectiveness of the proposed method, we calculate the error rate. Finally, we compare this new SVR method with Lagrange interpolation solution

International audience; Information-centric networking (ICN), a new networking paradigm in which the focal point is a named data, has been proposed recently as an evolving concept to the actual host-centric model of the Internet that relies mainly on host addresses. In vehicular networks, where vehicles are generally moving network elements and follow a content-oriented fashion, it will be fitting to use the ICN paradigm to improve the content dissemination and reduce the content retrieval latency. By applying this concept to such networks, we focus in this paper on the content delivery issue and propose an optimized caching scheme that proactively predicts the moving direction of a vehicle and brings into the next encountered RSU cache only the required content of interest to that vehicle. According to the obtained results from different measured metrics, the proposed solution outperforms in many ways other proposed schemes in the literature. For instance, our scheme improves drastically the cache utilization, enhances the network delay, and boosts the content diversity and distribution

International audience; Spectro-spatial dimensionality reduction in HyperSpectral Images (HSI) classification is a challenging task due to the problem of curse dimensionality, i.e. the high number of spectral bands and the heterogeneity of data. In this context, many dimensionality reduction methods have been developed to overcome the high correlation between bands and the redundancy of information in order to improve the classification accuracy. Most of these methods represent the original HSI as a set of vectors. Therefore, they only exploit spectral properties, neglecting the spatial information, i.e. the spatial rearrangement is lost. To jointly take advantage of spatial and spectral information, HSI has been recently represented as a tensor. In order to preserve the spatial and spectral information, we develop a hybrid method using both the Tensor Locality Preserving Projections method (TLPP) projecting the original data into a lower subspace and the Constrained Band Selection method (CBS) to select the relevant bands. These two methods will be jointly used to get high-level quality classification. Moreover, since the two obtained classifications are uncertain and imprecise, we propose to fuse them using the Dempster-Shafer's Theory (DST) to obtain an accurate classification preserving the spectro-spatial information. The proposed approach has been applied on real HSI showing its efficiency compared with conventional dimensionality reduction methods.

International audience; This paper presents a methodology for integrating a new parameter measuring spatial relationships in the hidden Markov models (HMM) in order to detect, interpret and predict changes in urban areas from satellite images. This approach is divided into three phases: the detection of different spatial relationships in the urban area; the training of a hidden Markov model using Baum-Welch learning algorithm, integrating the changing spatial relationships obtained through the Allen's temporal algebra; the interpretation of changes in urban area and the prediction of future changes. Simulated spatiotemporal changes on synthetic data show the interest of this method for the analysis of spatiotemporal changes of relations between objects. Results allows detection and prediction to be performed from the various time series of images for the observations of spatiotemporal events such as urban expansion. It is therefore reasonable to use this approach to interpret and estimate the movement of the urban area.