Deep convolutional neural networks (DCNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains. However, acquiring a suitably large data set for training DCNN is ...often a significant challenge. This is a major issue in the remote sensing domain, where we have extremely large collections of satellite and aerial imagery, but lack the rich label information that is often readily available for other image modalities. In this letter, we investigate the use of DCNN for land-cover classification in high-resolution remote sensing imagery. To overcome the lack of massive labeled remote-sensing image data sets, we employ two techniques in conjunction with DCNN: transfer learning (TL) with fine-tuning and data augmentation tailored specifically for remote sensing imagery. TL allows one to bootstrap a DCNN while preserving the deep visual feature extraction learned over an image corpus from a different image domain. Data augmentation exploits various aspects of remote sensing imagery to dramatically expand small training image data sets and improve DCNN robustness for remote sensing image data. Here, we apply these techniques to the well-known UC Merced data set to achieve the land-cover classification accuracies of 97.8 ± 2.3%, 97.6 ± 2.6%, and 98.5 ± 1.4% with CaffeNet, GoogLeNet, and ResNet, respectively.
Earth observation (EO) sensors deliver data at daily or weekly intervals. Most land use and land cover classification (LULC) approaches, however, are designed for cloud-free and mono-temporal ...observations. The increasing temporal capabilities of today’s sensors enable the use of temporal, along with spectral and spatial features.Domains such as speech recognition or neural machine translation, work with inherently temporal data and, today, achieve impressive results by using sequential encoder-decoder structures. Inspired by these sequence-to-sequence models, we adapt an encoder structure with convolutional recurrent layers in order to approximate a phenological model for vegetation classes based on a temporal sequence of Sentinel 2 (S2) images. In our experiments, we visualize internal activations over a sequence of cloudy and non-cloudy images and find several recurrent cells that reduce the input activity for cloudy observations. Hence, we assume that our network has learned cloud-filtering schemes solely from input data, which could alleviate the need for tedious cloud-filtering as a preprocessing step for many EO approaches. Moreover, using unfiltered temporal series of top-of-atmosphere (TOA) reflectance data, our experiments achieved state-of-the-art classification accuracies on a large number of crop classes with minimal preprocessing, compared to other classification approaches.
•A set of pre-processing techniques for LUCAS in-situ data for high resolution Sentinel-2 imagery are tested systematically.•Overall, LUCAS in-situ samples are a valuable source for ground truth data ...for land cover image classifications using Sentinel-2 imagery.•Positional correction increases the overall accuracy by 3.7%.•A newly proposed pre-processing scheme with 7 semantic classes is applied to derive a nationwide land cover classification for Germany with 93.1% accuracy
In this study, we test the use of Land Use and Coverage Area frame Survey (LUCAS) in-situ reference data for classifying high-resolution Sentinel-2 imagery at a large scale. We compare several pre-processing schemes (PS) for LUCAS data and propose a new PS for a fully automated classification of satellite imagery on the national level. The image data utilizes a high-dimensional Sentinel-2-based image feature space. Key elements of LUCAS data pre-processing include two positioning approaches and three semantic selection approaches. The latter approaches differ in the applied quality measures for identifying valid reference points and by the number of LU/LC classes (7–12). In an iterative training process, the impact of the chosen PS on a Random Forest image classifier is evaluated. The results are compared to LUCAS reference points that are not pre-processed, which act as a benchmark, and the classification quality is evaluated by independent sets of validation points. The classification results show that the positional correction of LUCAS points has an especially positive effect on the overall classification accuracy. On average, this improves the accuracy by 3.7%. This improvement is lowest for the most rigid sample selection approach, PS2, and highest for the benchmark data set, PS0. The highest overall accuracy is 93.1% which is achieved by using the newly developed PS3; all PS achieve overall accuracies of 80% and higher on average. While the difference in overall accuracy between the PS is likely to be influenced by the respective number of LU/LC classes, we conclude that, overall, LUCAS in-situ data is a suitable source for reference information for large scale high resolution LC mapping using Sentinel-2 imagery. Existing sample selection approaches developed for Landsat imagery can be transferred to Sentinel-2 imagery, achieving comparable semantic accuracies while increasing the spatial resolution. The resulting LC classification product that uses the newly developed PS is available for Germany via DOI: https://doi.org/10.15489/1ccmlap3mn39.
In recent years, large amount of high spatial-resolution remote sensing (HRRS) images are available for land-cover mapping. However, due to the complex information brought by the increased spatial ...resolution and the data disturbances caused by different conditions of image acquisition, it is often difficult to find an efficient method for achieving accurate land-cover classification with high-resolution and heterogeneous remote sensing images. In this paper, we propose a scheme to apply deep model obtained from labeled land-cover dataset to classify unlabeled HRRS images. The main idea is to rely on deep neural networks for presenting the contextual information contained in different types of land-covers and propose a pseudo-labeling and sample selection scheme for improving the transferability of deep models. More precisely, a deep Convolutional Neural Networks (CNNs) is first pre-trained with a well-annotated land-cover dataset, referred to as the source data. Then, given a target image with no labels, the pre-trained CNN model is utilized to classify the image in a patch-wise manner. The patches with high confidence are assigned with pseudo-labels and employed as the queries to retrieve related samples from the source data. The pseudo-labels confirmed with the retrieved results are regarded as supervised information for fine-tuning the pre-trained deep model. To obtain a pixel-wise land-cover classification with the target image, we rely on the fine-tuned CNN and develop a hybrid classification by combining patch-wise classification and hierarchical segmentation. In addition, we create a large-scale land-cover dataset containing 150 Gaofen-2 satellite images for CNN pre-training. Experiments on multi-source HRRS images, including Gaofen-2, Gaofen-1, Jilin-1, Ziyuan-3, Sentinel-2A, and Google Earth platform data, show encouraging results and demonstrate the applicability of the proposed scheme to land-cover classification with multi-source HRRS images.
•A method to learn transferable deep model for 5-class land-cover (LC) classification.•A labeled dataset consisting of 150 Gaofen-2 images for LC classification.•It improves LC classification performance about 20% using multi-source RS images.•The method shows good transferability on different sensors and geolocations.
Machine learning offers the potential for effective and efficient classification of remotely sensed imagery. The strengths of machine learning include the capacity to handle data of high ...dimensionality and to map classes with very complex characteristics. Nevertheless, implementing a machine-learning classification is not straightforward, and the literature provides conflicting advice regarding many key issues. This article therefore provides an overview of machine learning from an applied perspective. We focus on the relatively mature methods of support vector machines, single decision trees (DTs), Random Forests, boosted DTs, artificial neural networks, and k-nearest neighbours (k-NN). Issues considered include the choice of algorithm, training data requirements, user-defined parameter selection and optimization, feature space impacts and reduction, and computational costs. We illustrate these issues through applying machine-learning classification to two publically available remotely sensed data sets.
The huge amount of data currently produced by modern Earth Observation (EO) missions has allowed for the design of advanced machine learning techniques able to support complex Land Use/Land Cover ...(LULC) mapping tasks. The Copernicus programme developed by the European Space Agency provides, with missions such as Sentinel-1 (S1) and Sentinel-2 (S2), radar and optical (multi-spectral) imagery, respectively, at 10 m spatial resolution with revisit time around 5 days. Such high temporal resolution allows to collect Satellite Image Time Series (SITS) that support a plethora of Earth surface monitoring tasks. How to effectively combine the complementary information provided by such sensors remains an open problem in the remote sensing field. In this work, we propose a deep learning architecture to combine information coming from S1 and S2 time series, namely TWINNS (TWIn Neural Networks for Sentinel data), able to discover spatial and temporal dependencies in both types of SITS. The proposed architecture is devised to boost the land cover classification task by leveraging two levels of complementarity, i.e., the interplay between radar and optical SITS as well as the synergy between spatial and temporal dependencies. Experiments carried out on two study sites characterized by different land cover characteristics (i.e., the Koumbia site in Burkina Faso and Reunion Island, a overseas department of France in the Indian Ocean), demonstrate the significance of our proposal.
Accurate information about the location, extent, and type of Land Cover (LC) is essential for various applications. The only recent available country-wide LC map of Iran was generated in 2016 by the ...Iranian Space Agency (ISA) using Moderate Resolution Imaging Spectroradiometer (MODIS) images with a considerably low accuracy. Therefore, the production of an up-to-date and accurate Iran-wide LC map using the most recent remote sensing, machine learning, and big data processing algorithms is required. Moreover, it is important to develop an efficient method for automatic LC generation for various time periods without the need to collect additional ground truth data from this immense country. Therefore, this study was conducted to fulfill two objectives. First, an improved Iranian LC map with 13 LC classes and a spatial resolution of 10 m was produced using multi-temporal synergy of Sentinel-1 and Sentinel-2 satellite datasets applied to an object-based Random forest (RF) algorithm. For this purpose, 2,869 Sentinel-1 and 11,994 Sentinel-2 scenes acquired in 2017 were processed and classified within the Google Earth Engine (GEE) cloud computing platform allowing big geospatial data analysis. The Overall Accuracy (OA) and Kappa Coefficient (KC) of the final Iran-wide LC map for 2017 was 95.6% and 0.95, respectively, indicating the considerable potential of the proposed big data processing method. Second, an efficient automatic method was developed based on Sentinel-2 images to migrate ground truth samples from a reference year to automatically generate an LC map for any target year. The OA and KC for the LC map produced for the target year 2019 were 91.35% and 0.91, respectively, demonstrating the efficiency of the proposed method for automatic LC mapping. Based on the obtained accuracies, this method can potentially be applied to other regions of interest for LC mapping without the need for ground truth data from the target year.
•We present an exhaustive evaluation of Guided Regularized Random Forest (GRRF), a feature selection method based on Random Forest.•GRRF does not require fixing a priori the number of features to be ...selected or setting a threshold of the feature importance.•GRRF features provide similar (or slightly better) results than when using all the features.•Comparing GRRF and RF features, the mean overall accuracy increases by almost 6% in classification and, the RMSE decreases by almost 2% in regression.
New Earth observation missions and technologies are delivering large amounts of data. Processing this data requires developing and evaluating novel dimensionality reduction approaches to identify the most informative features for classification and regression tasks. Here we present an exhaustive evaluation of Guided Regularized Random Forest (GRRF), a feature selection method based on Random Forest. GRRF does not require fixing a priori the number of features to be selected or setting a threshold of the feature importance. Moreover, the use of regularization ensures that features selected by GRRF are non-redundant and representative. Our experiments based on various kinds of remote sensing images, show that GRRF selected features provides similar results to those obtained when using all the available features. However, the comparison between GRRF and standard random forest features shows substantial differences: in classification, the mean overall accuracy increases by almost 6% and, in regression, the decrease in RMSE almost reaches 2%. These results demonstrate the potential of GRRF for remote sensing image classification and regression. Especially in the context of increasingly large geodatabases that challenge the application of traditional methods.
Samples play a crucial role in the supervised classification of remote sensing images. However, labeling large samples for training a classifier or deep learning network is not only time-consuming ...but also labor-intensive. In this paper, a novel land cover classification with nonparametric sample augmentation is proposed to improve the performance of hyperspectral remote sensing images (HRSIs) classification. First, initial samples with limited quantity are selected randomly from the ground truth map. Second, based on the gray image, a nonparametric adaptive region generation (NARG) algorithm is developed for utilizing the contextual information around each sample. Then, an nonparametric sample augmentation algorithm is developed with NARG to explore reliable samples iteratively around each initial sample. Finally, the above steps are fused into an iterative progress to obtain the final classification map. Compared with some typical traditional methods and some widely used deep learning methods based on four real HRSIs, our proposed approach exhibits some advantages in improving the visual performance and quantitative accuracies of HRSIs classification, such as the improvement is about 2.0% ~ 10.34% for four real HRSIs in term of the overall accuracy.
In this paper, we present a patch-based land use and land cover classification approach using Sentinel-2 satellite images. The Sentinel-2 satellite images are openly and freely accessible, and are ...provided in the earth observation program Copernicus. We present a novel dataset, based on these images that covers 13 spectral bands and is comprised of ten classes with a total of 27 000 labeled and geo-referenced images. Benchmarks are provided for this novel dataset with its spectral bands using state-of-the-art deep convolutional neural networks. An overall classification accuracy of 98.57% was achieved with the proposed novel dataset. The resulting classification system opens a gate toward a number of earth observation applications. We demonstrate how this classification system can be used for detecting land use and land cover changes, and how it can assist in improving geographical maps. The geo-referenced dataset EuroSAT is made publicly available at https://github.com/phelber/eurosat.