Automated time series vegetation data classification has received much attention in recent years. It has important benefits in remote sensing (RS) field such as land cover vegetation quality and productivity estimation. Recently, deep learning techniques have achieved increasing success in RS data classification including vegetation mapping. However, these approaches do not consider the vegetation's characteristics such as effective rate of crop utilization that can be useful in the process of vegetation classification. Motivated by this issue and for easy multi-temporal vegetation analysis, mapping and monitoring, in this paper; a neural network model is proposed to extract the features automatically and classify the vegetation's sentiment and purposes. We conduct experiments using multi-temporal public available Sentinel-2A datasets and the obtained experimental results are evaluated with the established criteria.

Certainly, unsupervised learning plays a critical role in the vast majority of the remote sensing (RS) applications due to the labeling costs. This research study focuses on a self-supervised representation learning approach for RS change detection (CD) based on increasing mutual information between features gathered from various views of a common context (same patch) in a non-linear way, inspired from a recently introduced unsupervised Deep Learning (DL) approach named InfoMax (DIM). The investigated model can learn image representations in unsupervised fashion that significantly produce promising results on CD. Outstandingly, using self-supervised learning, DIM model can learn representations on multispectral, hyperspectral or radar data. Our research reveals promising evidence of the significant potential utility of deep unsupervised features learning in other RS tasks.

Mapping vegetation quality is a vital and challenging task in remote sensing field. Due to changes of reflective features periods over time, many researchers are experiencing difficulties in retrieving automatically the type of the vegetation that meet their research needs. Recently, deep learning techniques has become the fastest-growing trend in remote sensing data classification including vegetation mapping data. To overcome the challenge of learning deep models and for easily multi-temporal vegetation mapping and monitoring, in this paper, we propose a Bidirectional Gated Recurrent Unit Network (BGRU) model, which is based on history information and able to deal with long-term sequential data using only few parameters to extract useful features using forward and backward gates for their automatically classification. Multi-temporal publicly available Sentinel-2A datasets with vegetation as the main theme are used to validate the proposed model and the obtained experimental results are evaluated with established criteria.

In this work, a deep learning approach has been developed to carry out optical remote sensing pansharpening by the fusion of high spectral and spatial information from two different sources. In the proposed approach, the combination of multimodal information is achieved at multiple levels. The cross fusion deep network (CNet) is designed to directly integrate information from training dataset; this is accomplished by using trainable cross connections between the Multispectral (MS) and the Panchromatic (PAN) images processing branches. To further highlight the benefits of using multiples cross fusion levels for pansharpening, comparison with baselines networks was carried out in this work using three fusion strategies: early, late, and the newly proposed cross fusion. The proposed fusion strategy was evaluated on images from Quickbird sensor and achieved good performance.

Pan-sharpening is a challenging ill-posed problem, which aims to restore a high-resolution multispectral image (MSHR) from its low-resolution (MSLR) image combined with a high-resolution panchromatic image (PAN). Powerful deep learning based techniques have achieved state-of-the-art performance in pan-sharpening. However, they can underperform when handling images with unknown land structures and areas. In this paper, a new designed UNet like network, able to learn the relationship between a set of randomly degraded MSLR images and their corresponding original MSHR images is developed. We propose to employ a degradation module on the training images in addition to learn a down- /up-sampling model by the deep network, allowing the construction of UNet like architecture for pansharpening. Experimental results show that the proposed model improves the visual quality of the obtained MSHR images while keeping a low reconstruction error.

With the development of deep learning (DL) techniques, recent research on pansharpening has improved the reconstruction accuracy in both spectral and spatial domains. Pansharpening is defined as the task of restoring high-frequencies details of high-resolution multispectral (MSHR) images from its low-resolution counterpart (MSLR) by exploiting panchromatic (PAN) high spatial resolution information. The proposed method exploits band independent deep neural network (BIN) for pansharpening, which uses a single band as input at once in combination with its PAN counterpart to minimize the residual error; this last has obtained significant improvement in accuracy with more stability and robustness against state-of-the-art methods. This perfection is due to the ability to train a very deep network through indirect and free data augmentation technique by considering each band as a separate training input, which makes it possible to duplicate each MSLR image into$s$inputs ($s$is number of bands).

Land cover classification is one of the most important research fields in the remote sensing community. With the rapid growth of remote sensing sensors, it has received much attention in the last years, and rich temporal, spectral and spatial information were generated. Existing land cover classification approaches can be categorized into single time and multi-temporal observations processing techniques. Due to changes of reflective features over time, multi-temporal land cover classification approaches may be hard to be implemented. In these investigations, a new hybrid neural networks model is proposed to tackle such issues, and in particular, for easily multi-temporal vegetation mapping and monitoring. The proposed model combines convolutional and recurrent neural networks to capture local features and long-term dependencies of multi-temporal remote sensing images. This combination is expected to extract useful features for their automatically classification. The proposed model is applied to multi-temporal publicly available Sentinel-2A datasets with vegetation as the main theme, and obtained experimental results are evaluated with established criteria.

Convolutional Neural Network (CNN) has boosted the performance of Very High Resolution (VHR) remote sensing data classification. Moreover, the continuous development of CNN techniques for image scenes description has entered a new challenge. The deep neural network models require a huge number of training samples, which is the main limitation of processing remote sensing data. To overcome this issue, a new method, based on the Capsule neural network for VHR image scenes recognition is proposed in this work. Experiments on the public Aerial Image Dataset (AID) benchmark, containing different areal categories with sub-meter spatial resolution are conducted. The obtained results demonstrate the effectiveness of the proposed method, as compared with the classical CNN model.

Change detection in land and urban environments has been an important task in remote sensing field. Deep learning has recently received an increasing attention from researchers and has been successfully applied for many domains. In remote sensing, it is very difficult to construct a large-scale well-annotated dataset due to the expense of data acquisition and the costly annotation, which limits its development. In this work, we introduce deep transfer learning as a way of overcoming the application of deep learning techniques for optical remote sensing change detection. Through several experiments, we investigate various uses of pre-trained Convolutional Neural Networks (CNNs) models. Our experiments on bi-temporal dataset show that VGG16 and ResNet networks consistently yield the best performances across considered strategies. It also appears that fine-tuning pre-trained CNN models is the best performing strategies.

Automated classification of citation function in scientific text is a new emerging research topic inspired by traditional citation analysis in applied linguistic and scientometric fields.The aim is to classify citations in scholarly publication in order to identify author's purpose or motivation for quoting or citing a particular paper.Several citation schemes have been proposed to classify the citations into different functions.However, it is extremely challenging to find standard scheme to classify citations, and some of the proposed schemes have similar functions.Moreover, most of previous studies mainly used classical machine learning methods such as support vector machine and neural networks with a number of manually created features.These features are incomplete and suffer from time-consuming and error prone weakness.To address these problems, we present a new citation scheme with less functions and propose a deep learning model for classification.The citation sentences and author's information were fed to convolutional neural networks to build citation and author representations.A corpus was built using the proposed scheme and a number of experiments were carried out to assess the model.Experimental results have shown that the proposed approach outperforms the existing methods in term of accuracy, precision and recall.