a b s t r a c tLight Detection and Ranging (LiDAR) is a remote sensor able to extract three-dimensional information. Environmental models in forest areas have been benefited by the use of LiDAR-derived information in the last years. A multiple linear regression (MLR) with previous stepwise feature selection is the most common method in the literature to develop those models. MLR defines the relation between the set of field measurements and the statistics extracted from a LiDAR flight. Machine learning has emerged as a suitable tool to improve classic stepwise MLR results on LiDAR. Unfortunately, few studies have been proposed to compare the quality of the multiple machine learning approaches. This paper presents a comparison between the classic MLR-based methodology and regression techniques in machine learning (neural networks, support vector machines, nearest neighbour, ensembles such as random forests) with special emphasis on regression trees. The selected techniques are applied to real LiDAR data from two areas in the province of Lugo (Galizia, Spain). The results confirm that classic MLR is outperformed by machine learning techniques and concretely, our experiments suggest that Support Vector Regression with Gaussian kernels statistically outperforms the rest of the techniques.

Analyzing land use and land cover (LULC) using remote sensing (RS) imagery is essential for many environmental and social applications. The increase in availability of RS data has led to the development of new techniques for digital pattern classification. Very recently, deep learning (DL) models have emerged as a powerful solution to approach many machine learning (ML) problems. In particular, convolutional neural networks (CNNs) are currently the state of the art for many image classification tasks. While there exist several promising proposals on the application of CNNs to LULC classification, the validation framework proposed for the comparison of different methods could be improved with the use of a standard validation procedure for ML based on cross-validation and its subsequent statistical analysis. In this paper, we propose a general CNN, with a fixed architecture and parametrization, to achieve high accuracy on LULC classification over RS data from different sources such as radar and hyperspectral. We also present a methodology to perform a rigorous experimental comparison between our proposed DL method and other ML algorithms such as support vector machines, random forests, and k-nearest-neighbors. The analysis carried out demonstrates that the CNN outperforms the rest of techniques, achieving a high level of performance for all the datasets studied, regardless of their different characteristics.

Machine learning combines inductive and automated techniques for recognizing patterns. These techniques can be used with remote sensing datasets to map aboveground biomass (AGB) with an acceptable degree of accuracy for evaluation and management of forest ecosystems. Unfortunately, statistically rigorous comparisons of machine learning algorithms are scarce. The aim of this study was to compare the performance of the 3 most common nonparametric machine learning techniques reported in the literature, vis., Support Vector Machine (SVM), k-nearest neighbor (kNN) and Random Forest (RF), with that of the parametric multiple linear regression (MLR) for estimating AGB from Landsat-5 Thematic Mapper (TM) spectral reflectance data, texture features derived from the Normalized Difference Vegetation Index (NDVI), and topographical features derived from a digital elevation model (DEM). The results obtained for 99 permanent sites (for calibration/validation of the models) established during the winter of 2011 by systematic sampling in the state of Durango (Mexico), showed that SVM performed best once the parameterization had been optimized. Otherwise, SVM could be outperformed by RF. However, the kNN yielded the best overall results in relation to the goodness-of-fit measures. The findings confirm that nonparametric machine learning algorithms are powerful tools for estimating AGB with datasets derived from sensors with medium spatial resolution. Résumé. L'apprentissage automatique combine des techniques inductives et automatisées pour la reconnaissance des formes. Ces techniques peuvent être utilisées avec des ensembles de données de télédétection pour cartographier la biomasse aérienne « aboveground biomass » (AGB) avec un degré de précision acceptable pour l'évaluation et la gestion des écosystèmes forestiers. Malheureusement, des comparaisons statistiquement rigoureuses des algorithmes d'apprentissage automatique sont rares. Le but de cette étude était de comparer les performances des 3 méthodes d'apprentissage automatique non paramétriques les plus fréquemment rapportées dans la littérature, vis., les machines à vecteurs de support « Support Vector Machine » (SVM), les k plus proches voisins « k-nearest neighbor » (kNN) et les forêts aléatoires « Random Forest » (RF), avec celle de la régression linéaire multiple paramétrique (MLR) pour l'estimation de l'AGB provenant des données de réflectance spectrale de Landsat-5 Thematic Mapper (TM), des caractéristiques de texture dérivées de l'indice de végétation par différence normalisée « Normalized Difference Vegetation Index » (NDVI) et des caractéristiques topographiques dérivées d'un modèle numérique de terrain « digital elevation model » (DEM).Les résultats obtenus pour 99 sites permanents (pour la calibration/validation des modèles) établis au cours de l'hiver 2011 par l'échantillonnage systématique dans l' État de Durango (Mexique), ont montré que les SVM montrent leurs meilleures performances une fois que le paramétrage a été optimisé. Par ailleurs, les SVM pourraient ê...

Object detection using remote sensing data is a key task of the perception systems of self-driving vehicles. While many generic deep learning architectures have been proposed for this problem, there is little guidance on their suitability when using them in a particular scenario such as autonomous driving. In this work, we aim to assess the performance of existing 2D detection systems on a multi-class problem (vehicles, pedestrians, and cyclists) with images obtained from the on-board camera sensors of a car. We evaluate several one-stage (RetinaNet, FCOS, and YOLOv3) and two-stage (Faster R-CNN) deep learning meta-architectures under different image resolutions and feature extractors (ResNet, ResNeXt, Res2Net, DarkNet, and MobileNet). These models are trained using transfer learning and compared in terms of both precision and efficiency, with special attention to the real-time requirements of this context. For the experimental study, we use the Waymo Open Dataset, which is the largest existing benchmark. Despite the rising popularity of one-stage detectors, our findings show that two-stage detectors still provide the most robust performance. Faster R-CNN models outperform one-stage detectors in accuracy, being also more reliable in the detection of minority classes. Faster R-CNN Res2Net-101 achieves the best speed/accuracy tradeoff but needs lower resolution images to reach real-time speed. Furthermore, the anchor-free FCOS detector is a slightly faster alternative to RetinaNet, with similar precision and lower memory usage.

Quantifying biomass consumption and carbon release is critical to understanding the role of fires in the carbon cycle and air quality. We present a methodology to estimate the biomass consumed and the carbon released by the California Rim fire by integrating postfire airborne LiDAR and multitemporal Landsat Operational Land Imager (OLI) imagery. First, a support vector regression (SVR) model was trained to estimate the aboveground biomass (AGB) from LiDAR-derived metrics over the unburned area. The selected model estimated AGB with an R 2 of 0.82 and RMSE of 59.98 Mg/ha. Second, LiDAR-based biomass estimates were extrapolated to the entire area before and after the fire, using Landsat OLI reflectance bands, Normalized Difference Infrared Index, and the elevation derived from LiDAR data. The extrapolation was performed using SVR models that resulted in R 2 of 0.73 and 0.79and RMSE of 87.18 (Mg/ha) and 75.43 (Mg/ha) for the postfire and prefire images, respectively. After removing bias from the AGB extrapolations using a linear relationship between estimated and observed values, we estimated the biomass consumption from postfire LiDAR and prefire Landsat maps to be 6.58 ± 0.03 Tg (10 12 g), which translate into 12.06 ± 0.06 Tg CO2 e released to the atmosphere, equivalent to the annual emissions of 2.57 million cars.

No abstract

a b s t r a c tLight detection and ranging (LiDAR) has become an important tool in forestry. LiDAR-derived models are mostly developed by means of multiple linear regression (MLR) after stepwise selection of predictors. An increasing interest in machine learning and evolutionary computation has recently arisen to improve regression use in LiDAR data processing. Although evolutionary machine learning has already proven to be suitable for regression, evolutionary computation may also be applied to improve parametric models such as MLR. This paper provides a hybrid approach based on joint use of MLR and a novel genetic algorithm for the estimation of the main forest stand variables. We show a comparison between our genetic approach and other common methods of selecting predictors. The results obtained from several LiDAR datasets with different pulse densities in two areas of the Iberian Peninsula indicate that genetic algorithms perform better than the other methods statistically. Preliminary studies suggest that a lack of parametric conditions in field data and possible misuse of parametric tests may be the main reasons for the better performance of the genetic algorithm. This research confirms the findings of previous studies that outline the importance of evolutionary computation in the context of LiDAR analisys of forest data, especially when the size of fieldwork datatasets is reduced.

Data streaming classification has become an essential task in many fields where real-time decisions have to be made based on incoming information. Neural networks are a particularly suitable technique for the streaming scenario due to their incremental learning nature. However, the high computation cost of deep architectures limits their applicability to high-velocity streams, hence they have not yet been fully explored in the literature. Therefore, in this work, we aim to evaluate the effectiveness of complex deep neural networks for supervised classification in the streaming context. We propose an asynchronous deep learning framework in which training and testing are performed simultaneously in two different processes. The data stream entering the system is dual fed into both layers in order to concurrently provide quick predictions and update the deep learning model. This separation reduces processing time while obtaining high accuracy on classification. Several time-series datasets from the UCR repository have been simulated as streams to evaluate our proposal, which has been compared to other methods such as Hoeffding trees, drift detectors, and ensemble models. The statistical analysis carried out verifies the improvement in performance achieved with our dual-pipeline deep learning framework, that is also competitive in terms of computation time.