Search

Image time series, such as Satellite Image Time Series (SITS) or MRI functional sequences in the medical domain, carry both spatial and temporal information. In many pattern recognition applications such as image classification, taking into account such rich information may be crucial and discriminative during the decision making stage. However, the extraction of spatio-temporal features from image time series is difficult to handle due to the complex representation of the data cube. In this paper, we present a strategy based on Random Walk to build a novel segment-based representation of the data, passing from a 2D[Formula: see text] dimension to a 2D one, more easily manipulable and without losing too much spatial information. Such new representation is then used to feed a classical Convolutional Neural Network (CNN) in order to learn spatio-temporal features with only 2D convolutions and to classify image time series data for a particular classification problem. The influence of the way the 2D[Formula: see text] data are represented, as well as the impact of the network architectures on the results, are carefully studied. The interest of this approach is highlighted on a remote sensing application for the classification of complex agricultural crops.

Image time series (ITS) represent complex 3D (2D+t in practice) data that are now daily produced in various domains, from medical imaging to remote sensing. They contain rich spatio-temporal information allowing the observation of the evolution of a sensed scene over time. In this work, we focus on the classification task of ITS, as often available in remote sensing tasks. An underlying problem here is to consider jointly the spatial and the temporal dimensions of the data. We present Deep-STaR, a method to learn such features from ITS data to proceed to their classification. Instead of reasoning in the original 2D+t space, we investigate novel 2D planar data representations, containing both temporal and spatial information. Such representations are a novel way to structure the ITS, compatible with deep learning architectures. They are used to feed a convolutional neural network to learn spatio-temporal features with 2D convolutions, leading ultimately to classification decision. To enhance the explainability of the results, we also propose a post-hoc attention mechanism, enabled by this new approach, providing a semantic map giving some insights for the taken decision. Deep-STaR is evaluated on a remote sensing application, for the classification of agricultural crops from satellite ITS. The results highlight the benefice of this method, compared to the literature, and its interest to make easier the interpretation of ITS to understand spatio-temporal phenomena.

International audience

Satellite Image Time Series (SITS), MRI sequences, and more generally image time series, constitute $2D+t$ data providing spatial and temporal information about an observed scene. Given a pattern recognition task such as image classification, considering jointly such rich information is crucial during the decision process. Nevertheless, due to the complex representation of the data-cube, spatio-temporal features extraction from $2D+t$ data remains difficult to handle. We present in this article an approach to learn such features from this data, and then to proceed to their classification. Our strategy consists in enriching pixel time series with spatial information. It is based on Random Walk to build a novel segment-based representation of the data, passing from a $2D+t$ dimension to a $2D$ one, without loosing too much spatial information. Such new representation is then involved in an end-to-end learning process with a classical 2D Convolutional Neural Network (CNN) in order to learn spatiotemporal features for the classification of image time series. Our approach is evaluated on a remote sensing application for the mapping of agricultural crops. Thanks to a visual attention mechanism, the proposed $2D$ spatio-temporal representation makes also easier the interpretation of a SITS to understand spatiotemporal phenomenons related to soil management practices.

Satellite Image Time Series (SITS) provide valuable information for the study of the Earth’s surface. In particular, this information may improve the comprehension, the understanding and the mapping of many phenomenons such as earthquake monitoring, urban sprawling or agricultural practices. In this article, we propose a method to define new spatio-temporal features from SITS based on the measure of the temporal stability. The proposed method is based on a compression algorithm named Run Length Encoding leading to a novel image representation from which stability features can be measured. Such features can then be used in several applications such as SITS summarizing, to make easier the interpretation of such a data-cube, or the classification of spatio-temporal patterns. The preliminary results obtained from a series of 50 Sentinel-2 optical images highlight the interest of our approach in a remote sensing application.

Image time series, such as Satellite Image Time Series (SITS) or MRI functional sequences in the medical domain, carry both spatial and temporal information. In many applications such as image classification, taking into account such rich information may be crucial and discrimative during the decision making stage. However, the extraction of spatio-temporal features from image time series is difficult due to the complex representation of the data cube. In this article, we present a strategy based on Random Walk to build a novel segment-based representation of the data, passing from a \(2D+t\) dimension to a 2D one, more easily manipulable and without loosing too much spatial information. Such new representation is then used to feed a classical Convolutional Neural Network (CNN) in order to learn spatio-temporal features with only 2D convolutions and to classify image time series data for a particular classification problem. The interest of this approach is highlighted on a remote sensing application for the classification and the mapping of complex agricultural crops.

International audience; Image time series, such as Satellite Image Time Series (SITS) or MRI functional sequences in the medical domain, carry both spatial and temporal information of the sensed scene. In many applications such as image classification, taking into account such rich information may be crucial and discrimative during the decision making stage. However , the extraction of spatio-temporal features from image time series is difficult due to the complex representation of the data cube. In this article, we present a strategy based on Random Walk to build a novel segment-based representation of the data, passing from a 2D + t dimension to a 2D one, more easily usable and without loosing too much spatial information. Such new representation is then used to feed a classical Convolutional Neural Network (CNN) in order to learn spatio-temporal features with only 2D convolutions and to classify image time series data for a particular classification problem. The interest of this approach is highlighted on a remote sensing application for the classification and the mapping of complex agricultural crops.; Les séries temporelles d'images, telles que les séries temporelles d'images satellites (STIS) ou les séquences fonctionnelles d'IRM dans le domaine médical, fournissent des informations spatiales et temporelles importantes sur une scène observée. Dans de nombreuses applications telles que la classification d'images, la prise en compte de ces informations peut être cruciale et discriminatoire lors de la prise de décision. Cependant, l'extraction de caractéristiques spatio-temporelles à partir de séries temporelles d'images est difficile en raison de la complexité de la structure du cube de données. Dans cet article, nous présentons une stratégie basée sur les marches aléatoires (Random Walk) pour construire une nouvelle représentation des données reposant sur des segments, passant d'une dimension 2D + t à une dimension 2D, plus facilement manipulable et conservant une information spatiale partielle. Cette nouvelle représentation est ensuite utilisée pour définir des caractéristiques spatio-temporelles dans un réseau de neurones convolutionnel (CNN) classique lors d'un apprentissage bout en bout ayant pour objectif de classer des séries temporelles d'images. L'intérêt de cette approche est mis en évidence sur une application de télédétection pour la classification et la cartographie de parcelles agricoles.

Motion is a real issue in video sequence analysis since it describes an object in three dimensions whereas the images are defined by a 3D scene projection in a 2D plane. An important problem in tracking of a point between two successive images is that only one pixel can’t be tracked since the intensity pixel value can both change due to noise and can be confused with adjacent pixels. As a result, it’s often impossible to determine the pixel location in the next frame, based only on the local information.In this article, we are interested in facial motion tracking for talking faces. When talking, points on the face move and their surround intensity change in a complex way. Therefore, we propose a comparison study of three methods for facial motion tracking to estimate the trajectory of each facial point on a talking face by analyzing templates intensities levels belonging two successive images. The first method is based on block matching: Normalized Sum of Squared Differences (NSSD), the second on normalized cross correlation (NCC), whereas the third method concerns the Kanade–Lucas–Tomasi Tracking (KLT tracker). Results obtained are compared, based on tracked error on the trajectory of a video by measuring the root-mean-squared intensity difference between the current and the last template.

The purpose of this work is to explore the advantage of Mel Frequency Cepstral Coefficients and Linear Predictive Coding combined with Principal Component Analysis, for speaker identification system applied for Arabic and Algerian Berber language. For this aim, the multilayer perception classifier is applied where different training functions are used for recognition process to improve the global efficiency. Dynamic time wrapping (DTW) is also applied for speaker verification. A Comparative analysis with some methods reveals that the proposed method provide good efficiency for verification and identification.

International audience

This article proposes an implementation of a dynamic gesture recognition system using different textural descriptors such as the basic Local Binary Patterns (LBP), Rotation Invariant and Uniform LBP (LBPriu2), Center-Symmetric LBP (CS-LBP) and Edge Histogram Descriptor (EHD). The recognition task is performed using two variants of the Artificial Neural Network (ANN), which are the Multilayer Perceptron (MLP) and the Radial Basis Function neural network (RBF). Experiments were performed on a user-independent database with a simple background where 95.83% recognition rate was achieved. A comparison with previous works shows the efficiency of our system.

Satellite Image Time Series (SITS) provide valuable information for the automatic mapping of our territories. In this article we focus on the analysis of urban land covers from SITS, trying to evaluate the density of artificialized areas. We hypothesize that such areas do not evolve significantly through time (over the interval of a year) compared to other nonartificialized areas (e.g., agricultural crops, vegetation). The proposed approach is based on a spatio-temporal characteristic measuring the temporal stability of a zone, extracted using the Run Length Encoding method. Preliminary results obtained on a series of 41 SENTINEL-2 images highlight the ability of our approach to discriminate different urban land-cover classes (e.g., artificial areas, high density vs. low density housing areas).

Background: Compared with weaned calves, nursed calves benefit from access to milk as nutrient-dense feed and mentoring by their dams. Both may be decisive for growth under difficult conditions such as grazing on alpine pastures, where experience could play an essential role. These factors were investigated separately by comparing nursed calves with calves weaned at 3 months of age and grazing with or without a mentor cow (n = 8 calves).; Results: Weaned calves with mentor cows spent less time lying and walked more than nursed calves. Irrespective of whether they had a mentor cow or not, the weaned calves grew slower than the nursed calves (357, 428 and 1324 g day-1 respectively). This also adversely affected slaughter weight, dressing percentage, conformation and fat cover at slaughter. In general, mentoring had no effect. Meat quality did not differ whether weaned calves were mentored or not. Nursed calves were superior in intramuscular fat content, tenderness and water-holding capacity.; Conclusion: The availability of milk appeared to be much more important to the calves under demanding grazing conditions than did mentoring. The differences in growth rate were accompanied by unexpectedly large differences in meat quality, showing that early weaning was not suitable for this system.© 2016 Society of Chemical Industry.; © 2016 Society of Chemical Industry.

International audience; The study used 81 young bulls, half of which were disbudded at 7 weeks of age. The effects of horn status during rearing and of acute physical and emotional stress just before slaughter on the physiological status at slaughter and subsequent meat quality were investigated. Bulls were reared in groups containing i) only bulls with horns, ii) only bulls without horns, or iii) mixed (half with, and half without horns). Bulls of each rearing condition were assigned to one of two slaughter conditions: with limited (LS) or with supplementary stress (SS). LS resulted in lower heart rates, stress hormone concentrations and carcass temperature, whereas SS resulted in faster post mortem pH decline and lower juiciness. Horned bulls from unmixed rearing groups had lower early pm temperature, shorter sarcomeres, and lower tenderness compared to disbudded bulls. Correlations and regression analysis revealed relationships between physiological indicators, mainly heart rate before slaughter, and meat quality, including water-holding capacity and indicators of proteolysis.

In the present study, the effects of feeding 450 g/day of rumen-protected fish oil (FO) compared to sunflower oil (SO) to Angus heifers (60 g/kg total intake) were quantified. Animal performance was not affected whereas the physico-chemical meat quality, assessed in three muscles, was slightly affected by diet. The oxidative shelf life of the perirenal fat declined with FO compared to SO. Despite the formerly shown increased n–3 fatty acid proportions of meat due to FO supplementation, a trained sensory panel identified an only slightly more intense fishy flavour in grilled steaks and beef patties from the FO compared to the SO group. In FO compared to SO patties, flavour intensity was more pronounced. The perception of off-flavours was negligible and differences between muscles were larger than between diets. In conclusion, supplementing ruminants with FO containing nutritionally beneficial n–3 fatty acids results in few side-effects on meat quality, restricted to quite faint off-flavours and a shorter fat shelf life.

This article presents a gesture recognition system based on spatio-temporal descriptors such as Histogram of Oriented Gradients 3DHOG, Histogram of Oriented Optical Flow 3DHOOF. One-against-all SVM classifier is used to classify each gesture belonging to a predefined class using a dynamic dataset. Our method consists of implementing three architectures for our classification task: 3DHOG, 3DHOOF, 3DHOG-HOOF, and investigate the benefit of concatenating the two descriptors. Good recognition rate was obtained for the third architecture where 91.66% of efficiency was obtained.

This work aims to improve the performance of solar cells with structure SnO2/CdS/CIGS/Mo by using the solar cell capacitance simulator (SCAPS-1D). Our idea is to insert the SnS layer between the absorber layer CIGS and the Mo (molybdenum) back contact. A maximum conversion efficiency of 24.45% (VOC = 0.78 V, JSC = 38.66 mA/cm2, FF = 0.80) was achieved with 1 μm-CIGS absorber layer, 50 nm-CdS buffer layer, 200 nm of ZnO window layer and 300 nm of SnS BSF layer. This study shows that ultra-thin CIGS solar cells with a BSF layer give better results compared to the work of Gloeckler (2005), which presented an efficiency of 19.62% (VOC = 0.61 V, JSC = 39.23 mA/cm2, FF = 0.81) with 3 µm thickness of CIGS in the structure ZnO/CdS/CIGS/Mo. In addition to this, the cells’ stability with temperature was studied and analysed.