Search

The use of high spatial resolution Satellite Image Time Series (SITS) provides an opportunity for a wide spectrum of Earth surface monitoring applications such as Land Use/Land Cover (LULC) mapping. Whereas the majority of Time Series (TS) classification literature concentrates on the analysis of raw 1D signals, here, we investigate a framework for LULC mapping based on 2D encoded multivariate SITS data to enhance their classification performances. In this novel approach, multivariate SITS data are transformed from 1D signals to 2D images using several encoding techniques namely Gramian Angular Summation field (GASF), Gramian angular difference field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP). Successively, a new multi-band image is derived and it is used as input to a state-of-the-art convolutional neural network (CNN) classification model. The possibility to effectively encode multivariate TS data into 2D representations paves the way to reuse the huge amount of research findings coming from the general field of computer vision and build on reliable and robust methods that have been demonstrated their quality in a multitude of downstream applications. Experiments carried out on three real-world benchmarks covering large spatial areas with contrasted land cover features, namely: Dordogne department in France, Reunion Island an oversee French territory and Koumbia municipality in Burkina Faso, underline the quality of the proposed framework when compared to standard approaches for land cover mapping from SITS and recent methods for multivariate TS classification. Matter of fact, our new framework outperforms the classification performances of standard land cover classification strategies based on the raw TS information achieving an average F1-score of 89.34%, 90.26% and 78.94% for the Reunion Island, Dordogne and Koumbia study site, respectively with an increasing of at least 2.5 points w.r.t. the best competing approach. • A novel approach for Land use land cover mapping is proposed. • 2D encoded multivariate satellite image time series are used. • A set of complementary transforming techniques was introduced and combined. • An in depth experimental evaluation is provided compared to the commonly used methods. [ABSTRACT FROM AUTHOR]

The main use of satellite imagery concerns the process of the spectral and spatial dimensions of the data. However, to extract useful information, the temporal dimension also has to be accounted for which increases the complexity of the problem. For this reason, there is a need for suitable data mining techniques for this source of data. In this work, we developed a data mining methodology to extract multidimensional sequential patterns to characterize temporal behaviors. We then used the extracted multidimensional sequences to build a classifier, and show how the patterns help to distinguish between the classes. We evaluated our technique using a real-world dataset containing information about land use in Mali (West Africa) to automatically recognize if an area is cultivated or not. [ABSTRACT FROM AUTHOR]

• Novel approach to estimate forest characteristics from GEDI with neural networks • Accuracy up to 1.53 m for forest heights and 27.35 m 3. ha −1 for trunk volume • Metric-free method has similar accuracy to metric-based method for both variables • Neural networks estimates showed almost zero bias for all forest height ranges Full waveform (FW) LiDAR systems have proven their effectiveness to map forest biophysical variables in the last two decades, owing to their ability of measuring, with high accuracy, forest vertical structures. The Global Ecosystem Dynamics Investigation (GEDI) system on board the International Space Station (ISS) is the latest FW spaceborne LiDAR instrument for the continuous observation of Earth's forests. FW systems rely on very sophisticated pre-processing steps to generate a priori metrics in order to leverage their capabilities for the accurate estimation of the aforementioned forest characteristics. The ever-expanding volume of acquired GEDI data, which to date comprises more than 25 billion acquired unfiltered shots, and along with the pre-processed data, amounting to more than 90 TB of data, raises new challenges in terms of adapted preprocessing methods for the suitable exploitation of such a huge and complex amount of LiDAR data. To overcome the issues related to the generation of relevant metrics from GEDI data, we propose a new metric-free approach to estimate canopy dominant heights (H dom) and wood volume (V) of Eucalyptus plantations over five different regions in Brazil. To avoid metric computation, we leverage deep learning techniques and, more in detail, convolutional neural networks with the aim to analyze the GEDI Level 1B geolocated waveforms. Performance comparisons were conducted between four convolutional neural network (CNN) variants using GEDI waveform data (either untouched, or subsetted) and a metric based Random Forest regressor (RF). Additionally, we tested if our framework can improve the generalization of the models to different distant regions. First, the models were trained using data from all the study regions. Cross validated results showed that the CNN based models compared well against their RF counterpart for both H dom and V. The RMSE on the estimation of H dom from the CNN based models varied between 1.54 and 1.94 m with a coefficient of determination (R 2) between 0.86 and 0.91, while the RF model produced an accuracy on H dom estimates of 1.45 m (R 2 = 0.92). For V, CNN based estimations ranged from 27.76 to 33.33 m 3. ha −1 (R 2 between 0.82 and 0.88), while for RF, the RMSE was 27.61 m 3. ha −1 (R 2 = 0.88). Next, model generalization was assessed by means of a spatial transfer experiment. For H dom , both the CNN and RF approaches showed similar performances to a global model, however, the CNN based approach showed higher variability on the estimation accuracy, and the variability was related to the forest structure between the trained and tested data (similar tree heights yield better accuracies). For the estimation of V, considering both approaches, the accuracy was dependent on the allometric relationship between H dom and V in the training and testing regions while lower accuracies on V were obtained when the testing and training regions exhibited a different allometric relationship. [ABSTRACT FROM AUTHOR]

Change Detection (CD) aims to distinguish surface changes based on bi-temporal remote sensing images. In recent years, deep neural models have made a breakthrough in CD processes. However, training a deep neural model requires a large volume of labelled training samples that are time-consuming and labour-intensive to acquire. With the aim of learning an accurate CD model with limited labelled data, we propose SENECA : a method based on a CD Siamese network, which takes advantage of both Transfer Learning (TL) and Active Learning (AL) to handle the constraint of limited supervision. More precisely, we jointly use AL and TL to adapt a CD model trained on a labelled source domain to a (related) target domain featured by restricted access to labelled data. We report results from an experimental evaluation involving five pairs of images acquired via Sentinel-2 satellites between 2015 and 2018 in various locations picked all over Asia and USA. The results show the beneficial effects of the proposed AL and TL strategies on the accuracy of the decisions made by the CD Siamese network and depict the merit of the proposed approach over competing CD baselines. • A Siamese network to detect changes in bi-temporal optical remote sensing images. • Transfer learning to adapt a model trained in a source domain to a target domain. • Segmentation-based active learning for transfer learning with weak supervision. • Experiments with Sentinel-2 imagery show the effectiveness of the method. [ABSTRACT FROM AUTHOR]

• Five Sentinel-2 images were tested to classify 3 soil surface characteristics. • Classifications were based on Random Forest models following two approaches. • Both approaches provided correct spatiotemporal classifications of SSCs. • Classification uncertainties were estimated and provided at field scale. • In future, such SSCs classifications might allow soil infiltration parameters mapping. Soil surface characteristics (SSCs) are of high importance for water infiltration processes in crop fields. As SSCs present strong spatiotemporal variability influenced by climatic conditions and agricultural practices, their monitor has already been explored by using UAV images and multispectral remote sensing. However, each technique has encountered difficulties characterizing this spatiotemporal variability. The objective of this work was to explore the potential of Sentinel-2 images to assess three SSCs – the green vegetation fraction, dry vegetation fraction and physical soil surface structure – at several dates. This work explored two approaches for classifying these three SSCs from five Sentinel-2 images acquired from August to November 2016. In the "single-date" approach, a Random Forest Classifier (RFC) model was trained to classify one SSC j from a dataset extracted from one Sentinel-2 image i (model noted RF_sd i,SSCj). In the "multi-date" approach, a RFC model was trained to classify one SSC j from a dataset extracted from the five Sentinel-2 images (noted RF_md SSCj). The classification analysis showed that i) the RF_sd i,SSCj and RF_md SSCj models provided accurate performances (overall accuracy > 0.79) regardless of the studied SSC j and the tested Sentinel-2 image, ii) the RF_sd i,SSCj model did not allow the classification of SSC classes that were not observed on the studied date, and iii) the RF_md SSCj model allowed the classification of all SSC classes observed in the five Sentinel-2 images. This indicated that several Sentinel-2 images can favourably be used to increase knowledge of spatiotemporal representation of SSCs by extending results of infrequent, localized and cumbersome field work. [ABSTRACT FROM AUTHOR]

Operational canopy height mapping at high resolution remains a challenging task at country-level. Most of the existing state-of-the-art inversion methods propose physically-based schemes which are specifically tuned for local scales. Only few approaches in the literature have attempted to produce country or global scale estimates, mostly by means of data-driven approaches and multi-spectral data sources. In this paper, we propose a robust deep learning approach that exploits single-pass interferometric TanDEM-X data to generate accurate forest height estimates from a single interferometric bistatic acquisition. The model development is driven by considerations on both the final performance and the trustworthiness of the model for large-scale deployment in the context of tropical forests. We train and test our model over the five tropical sites of the AfriSAR 2016 campaign, situated in the West Central state of Gabon, performing spatial cross-validation experiments to test its generalization capability. We define a specific training dataset and input predictors to develop a robust model for country-scale inference, by finding an optimal trade-off between the model performance and the large-scale reliability. The proposed model achieves an overall estimation bias of 0.12 m, a mean absolute error of 3.90 m, a root mean squared error of 5.08 m and a coefficient of determination of 0.77. Finally, we generate a time-tagged country-scale canopy height map of Gabon at 25 m resolution, discussing the potential and challenges of these kinds of products for their application in different scenarios and for the monitoring of forest changes. • Country-scale mapping of canopy height over Gabon, Africa. • Bistatic TanDEM-X InSAR features essential for bio-physical parameter estimation. • State-of-the-art canopy height estimation performance with deep learning. • Generation of calibrated estimates independent of SAR and InSAR geometries. • Novel reliability estimation approach of the country-scale product. [ABSTRACT FROM AUTHOR]

Abstract: Current work on assembling a set of local patterns such as rules and class association rules into a global model for the prediction of a target usually focuses on the identification of the minimal set of patterns that cover the training data. In this paper we present a different point of view: the model of a class has been built with the purpose to emphasize the typical features of the examples of the class. Typical features are modeled by frequent itemsets extracted from the examples and constitute a new representation space of the examples of the class. Prediction of the target class of test examples occurs by computation of the distance between the vector representing the example in the space of the itemsets of each class and the vectors representing the classes. It is interesting to observe that in the distance computation the critical contribution to the discrimination between classes is given not only by the itemsets of the class model that match the example but also by itemsets that do not match the example. These absent features constitute some pieces of information on the examples that can be considered for the prediction and should not be disregarded. Second, absent features are more abundant in the wrong classes than in the correct ones and their number increases the distance between the example vector and the negative class vectors. Furthermore, since absent features are frequent features in their respective classes, they make the prediction more robust against over-fitting and noise. The usage of features absent in the test example is a novel issue in classification: existing learners usually tend to select the best local pattern that matches the example and do not consider the abundance of other patterns that do not match it. We demonstrate the validity of our observations and the effectiveness of LODE, our learner, by means of extensive empirical experiments in which we compare the prediction accuracy of LODE with a consistent set of classifiers of the state of the art. In this paper we also report the methodology that we adopted in order to determine automatically the setting of the learner and of its parameters. [Copyright &y& Elsevier]