Search

Abstract Scene classification is a crucial research problem in remote sensing (RS) that has attracted many researchers recently. It has many challenges due to multiple issues, such as: the complexity of remote sensing scenes, the classes overlapping (as a scene may contain objects that belong to foreign classes), and the difficulty of gaining sufficient labeled scenes. Deep learning (DL) solutions and in particular convolutional neural networks (CNN) are now state-of-the-art solution in RS scene classification; however, CNN models need huge amounts of annotated data, which can be costly and time-consuming. On the other hand, it is relatively easy to acquire large amounts of unlabeled images. Recently, Self-Supervised Learning (SSL) is proposed as a method that can learn from unlabeled images, potentially reducing the need for labeling. In this work, we propose a deep SSL method, called RS-FewShotSSL, for RS scene classification under the few shot scenario when we only have a few (less than 20) labeled scenes per class. Under this scenario, typical DL solutions that fine-tune CNN models, pre-trained on the ImageNet dataset, fail dramatically. In the SSL paradigm, a DL model is pre-trained from scratch during the pretext task using the large amounts of unlabeled scenes. Then, during the main or the so-called downstream task, the model is fine-tuned on the labeled scenes. Our proposed RS-FewShotSSL solution is composed of an online network and a target network both using the EfficientNet-B3 CNN model as a feature encoder backbone. During the pretext task, RS-FewShotSSL learns discriminative features from the unlabeled images using cross-view contrastive learning. Different views are generated from each image using geometric transformations and passed to the online and target networks. Then, the whole model is optimized by minimizing the cross-view distance between the online and target networks. To address the problem of limited computation resources available to us, our proposed method uses a novel DL architecture that can be trained using both high-resolution and low-resolution images. During the pretext task, RS-FewShotSSL is trained using low-resolution images, thereby, allowing for larger batch sizes which significantly boosts the performance of the proposed pipeline on the task of RS classification. In the downstream task, the target network is discarded, and the online network is fine-tuned using the few labeled shots or scenes. Here, we use smaller batches of both high-resolution and low-resolution images. This architecture allows RS-FewshotSSL to benefit from both large batch sizes and full image sizes, thereby learning from the large amounts of unlabeled data in an effective way. We tested RS-FewShotSSL on three RS public datasets, and it demonstrated a significant improvement compared to other state-of-the-art methods such as: SimCLR, MoCo, BYOL and IDSSL.

Cross-modal text-image retrieval in remote sensing (RS) provides a flexible retrieval experience for mining useful information from RS repositories. However, existing methods are designed to accept queries formulated in the English language only, which may restrict accessibility to useful information for non-English speakers. Allowing multilanguage queries can enhance the communication with the retrieval system and broaden access to the RS information. To address this limitation, this article proposes a multilanguage framework based on transformers. Specifically, our framework is composed of two transformer encoders for learning modality-specific representations, the first is a language encoder for generating language representation features from the textual description, while the second is a vision encoder for extracting visual features from the corresponding image. The two encoders are trained jointly on image and text pairs by minimizing a bidirectional contrastive loss. To enable the model to understand queries in multiple languages, we trained it on descriptions from four different languages, namely, English, Arabic, French, and Italian. The experimental results on three benchmark datasets (i.e., RSITMD, RSICD, and UCM) demonstrate that the proposed model improves significantly the retrieval performances in terms of recall compared to the existing state-of-the-art RS retrieval methods.

Unimodal biometric systems rely on a single source or unique individual biological trait for measurement and examination. Fingerprint-based biometric systems are the most common, but they are vulnerable to presentation attacks or spoofing when a fake fingerprint is presented to the sensor. To address this issue, we propose an enhanced biometric system based on a multimodal approach using two types of biological traits. We propose to combine fingerprint and Electrocardiogram (ECG) signals to mitigate spoofing attacks. Specifically, we design a multimodal deep learning architecture that accepts fingerprints and ECG as inputs and fuses the feature vectors using stacking and channel-wise approaches. The feature extraction backbone of the architecture is based on data-efficient transformers. The experimental results demonstrate the promising capabilities of the proposed approach in enhancing the robustness of the system to presentation attacks.

The electrocardiogram (ECG) signal is shown to be promising as a biometric. To this end, it has been demonstrated that the analysis of ECG signals can be considered as a good solution for increasing the biometric security levels. This can be mainly due to its inherent robustness against presentation attacks. In this work, we present a deep contrastive learning-based system for ECG biometric identification. The proposed system consists of three blocks: a feature extraction backbone based on short time Fourier transform (STFT), a contrastive learning network, and a classification network. We evaluated the proposed system on the Heartprint dataset, a new ECG biometrics multi-session dataset. The experimental analysis shows promising capabilities of the proposed method. In particular, it yields an average top1 accuracy of 98.02% on a new dataset built by gathering 1539 ECG records from 199 subjects collected in multiple sessions with an average interval between sessions of 47 days.

The electrocardiogram (ECG) signal produced by the human heart is an emerging biometric modality that can play an important role in the future generation’s identity recognition with the support of machine learning techniques. One of the major obstacles in the progress of this modality is the lack of public datasets with a long interval between sessions of data acquisition to verify the uniqueness and permanence of the biometric signature of the heart of a subject. To address this issue, we put forward Heartprint, a large biometric database of multisession ECG signals comprising 1539 records captured from the fingers of 199 healthy subjects. The capturing time for each record was 15 s, and recordings were made in resting and reading conditions. They were collected in multiple sessions over ten years, and the average interval between first session (S1) and third session (S3L) was 1572.2 days. The dataset also covers several demographic classes such as genders, ethnicities, and age groups. The combination of raw ECG signals and demographic information turns the Heartprint dataset, which is made publicly available online, into a valuable resource for the development and evaluation of biometric recognition algorithms.

In this paper, we propose a novel end-to-end learnable architecture based on Dense Convolutional Networks (DCN) for the classification of electrocardiogram (ECG) signals. This architecture is based on two main modules: the first is a generative module and the second is a discriminative one. The task of the generative module is to convert the one dimensional ECG signal into an image by means of fully connected, up-sampling, and convolution layers. The discriminative module takes as input the generated image and carries out feature learning and classification. To handle the data imbalance problem characterizing the ECG data, we propose to use the focal loss (FL) that is based on the idea of reshaping the standard cross-entropy loss such that it reduces the loss assigned to well-classified ECG beats. In the experiments, we validate the method using the well-known MIT-BIH arrhythmia database in four different scenarios, using four classes in the first scenario, five in the second and 12 in the third. Finally, supraventricular versus the other three and ventricular versus the other three from the scenario with four classes are used as the fourth scenario. The results obtained show that the method proposed here achieves a significant accuracy improvement over all previous state-of-the-art methods.

The multi-label classification problem in Unmanned Aerial Vehicle (UAV) images is particularly challenging compared to single-label classification due to its combinatorial nature. To tackle this issue, we propose in this paper a deep learning approach based on encoder-decoder neural network architecture with channel and spatial attention mechanisms. Specifically, the encoder module which is based on a pre-trained convolutional neural network (CNN) has the task to transform the input image to a set of feature maps using an opportune feature combination. To improve the feature representation further, this module incorporates a squeeze excitation (SE) layer for modelling the interdependencies between the channels of the feature maps. The decoder module which is based on a long short terms memory (LSTM) network has the task of generating, in a sequential way, the classes present in the image. At each time step, it predicts the next class-label by aligning its hidden state to the corresponding region in the image by means of an adaptive spatial attention mechanism. The experiments carried out on two UAV datasets with a spatial resolution of 2-cm show that our method is promising in predicting the labels present in the image while attending the relevant objects in the image. Additionally, it is able to provide better classification results compared to state-of-the-art methods.

We present a new method for multi-source semi-supervised domain adaptation in remote sensing scene classification. The method consists of a pre-trained convolutional neural network (CNN) model, namely EfficientNet-B3, for the extraction of highly discriminative features, followed by a classification module that learns feature prototypes for each class. Then, the classification module computes a cosine distance between feature vectors of target data samples and the feature prototypes. Finally, the proposed method ends with a Softmax activation function that converts the distances into class probabilities. The feature prototypes are also divided by a temperature parameter to normalize and control the classification module. The whole model is trained on both the unlabeled and labeled target samples. It is trained to predict the correct classes utilizing the standard cross-entropy loss computed over the labeled source and target samples. At the same time, the model is trained to learn domain invariant features using another loss function based on entropy computed over the unlabeled target samples. Unlike the standard cross-entropy loss, the new entropy loss function is computed on the model’s predicted probabilities and does not need the true labels. This entropy loss, called minimax loss, needs to be maximized with respect to the classification module to learn features that are domain-invariant (hence removing the data shift), and at the same time, it should be minimized with respect to the CNN feature extractor to learn discriminative features that are clustered around the class prototypes (in other words reducing intra-class variance). To accomplish these maximization and minimization processes at the same time, we use an adversarial training approach, where we alternate between the two processes. The model combines the standard cross-entropy loss and the new minimax entropy loss and optimizes them jointly. The proposed method is tested on four RS scene datasets, namely UC Merced, AID, RESISC45, and PatternNet, using two-source and three-source domain adaptation scenarios. The experimental results demonstrate the strong capability of the proposed method to achieve impressive performance despite using only a few (six in our case) labeled target samples per class. Its performance is already better than several state-of-the-art methods, including RevGrad, ADDA, Siamese-GAN, and MSCN.

With the rapid growth of fingerprint-based biometric systems, it is essential to ensure the security and reliability of the deployed algorithms. Indeed, the security vulnerability of these systems has been widely recognized. Thus, it is critical to enhance the generalization ability of fingerprint presentation attack detection (PAD) cross-sensor and cross-material settings. In this work, we propose a novel solution for addressing the case of a single source domain (sensor) with large labeled real/fake fingerprint images and multiple target domains (sensors) with only few real images obtained from different sensors. Our aim is to build a model that leverages the limited sample issues in all target domains by transferring knowledge from the source domain. To this end, we train a unified generative adversarial network (UGAN) for multidomain conversion to learn several mappings between all domains. This allows us to generate additional synthetic images for the target domains from the source domain to reduce the distribution shift between fingerprint representations. Then, we train a scale compound network (EfficientNetV2) coupled with multiple head classifiers (one classifier for each domain) using the source domain and the translated images. The outputs of these classifiers are then aggregated using an additional fusion layer with learnable weights. In the experiments, we validate the proposed methodology on the public LivDet2015 dataset. The experimental results show that the proposed method improves the average classification accuracy over twelve classification scenarios from 67.80 to 80.44% after adaptation.

The efficient performance of metal-ion batteries strongly depends on electrode materials characteristics. Two-dimensional (2D) materials are among promising electrode materials for metal-ion battery cells, owing to their excellent structural and electronic properties. Two-dimensional graphdiyne has been recently fabricated and revealed unique storage capacities and fast charging rates. The current study explores the performance of the novel phosphorated-triphenylene graphdiyne (P-TpG) monolayer as an anode material for Li-, Na-, K-, Mg-, and Ca-ions storage via extensive density functional theory (DFT) simulations. Our results reveal that the stable structure of P-TpG monolayers delivers ultra-high storage capacities of ~2148, ~1696, ~1017, and ~2035 mA·h·g−1 for Li-, Na-, K-, and Ca- ions, respectively. Notably, the metallic electronic behavior is illustrated by adsorbing metal-ions on the P-TpG nanosheets, suggesting a good electronic conductivity. The NEB results demonstrate that P-TpG can serve as an outstanding candidate for the optimal charging/discharging process. This theoretical study suggests P-TpG nanosheets as a highly promising candidate for the design of advanced metal-ion batteries with remarkable charge capacities and optimal charging/discharging rates.

The heart is potentially a highly secured biometric modality. Although many templates have been proposed to be extracted from heart-signal for biometric authentication, they have yet to reach a single digit equal error rate (EER) of false matches and false non-matches when applied on large across-session data sets, where gallery and probe data are taken from different sessions. However, since different templates possess different strengths, the fusion of them has a great potential to improve the authentication performance. We propose an efficient template selection algorithm to select a suitable subset of templates from a given set to obtain a minimal EER. The fusion of the subset of templates selected by this algorithm from a set of seven state-of-the-art templates has obtained a significant 5% reduction of EER in authentication in our experiments on a large database of finger-based ECG signals captured in two different sessions.

Fingerprint-based biometric systems have grown rapidly as they are used for various applications including mobile payments, international border security, and financial transactions. The widespread nature of these systems renders them vulnerable to presentation attacks. Hence, improving the generalization ability of fingerprint presentation attack detection (PAD) in cross-sensor and cross-material setting is of primary importance. In this work, we propose a solution based on a transformers and generative adversarial networks (GANs). Our aim is to reduce the distribution shift between fingerprint representations coming from multiple target sensors. In the experiments, we validate the proposed methodology on the public LivDet2015 dataset provided by the liveness detection competition. The experimental results show that the proposed architecture yields an increase in average classification accuracy from 68.52% up to 83.12% after adaptation.

Most of the existing domain adaptation (DA) methods proposed in the context of remote sensing imagery assume the presence of the same land-cover classes in the source and target domains. Yet, this assumption is not always realistic in practice as the target domain may contain additional classes unknown to the source leading to the so-called open set DA. Under this challenging setting, the problem turns to reducing the distribution discrepancy between the shared classes in both domains besides the detection of the unknown class samples in the target domain. To deal with the openset problem, we propose an approach based on adversarial learning and pareto-based ranking. In particular, the method leverages the distribution discrepancy between the source and target domains using min-max entropy optimization. During the alignment process, it identifies candidate samples of the unknown class from the target domain through a pareto-based ranking scheme that uses ambiguity criteria based on entropy and the distance to source class prototype. Promising results using two cross-domain datasets that consist of very high resolution and extremely high resolution images, show the effectiveness of the proposed method.

In this paper, we propose a method for localizing the optic nerve head and segmenting the optic disc/cup in retinal fundus images. The approach is based on a simple two-stage Mask-RCNN compared to sophisticated methods that represent the state-of-the-art in the literature. In the first stage, we detect and crop around the optic nerve head then feed the cropped image as input for the second stage. The second stage network is trained using a weighted loss to produce the final segmentation. To further improve the detection in the first stage, we propose a new fine-tuning strategy by combining the cropping output of the first stage with the original training image to train a new detection network using different scales for the region proposal network anchors. We evaluate the method on Retinal Fundus Images for Glaucoma Analysis (REFUGE), Magrabi, and MESSIDOR datasets. We used the REFUGE training subset to train the models in the proposed method. Our method achieved 0.0430 mean absolute error in the vertical cup-to-disc ratio (MAE vCDR) on the REFUGE test set compared to 0.0414 obtained using complex and multiple ensemble networks methods. The models trained with the proposed method transfer well to datasets outside REFUGE, achieving a MAE vCDR of 0.0785 and 0.077 on MESSIDOR and Magrabi datasets, respectively, without being retrained. In terms of detection accuracy, the proposed new fine-tuning strategy improved the detection rate from 96.7% to 98.04% on MESSIDOR and from 93.6% to 100% on Magrabi datasets compared to the reported detection rates in the literature.

This manuscript introduces a computational approach to micro-damage problems using deep learning for the prediction of loading deflection curves. The location of applied forces, dimensions of the specimen and material parameters are used as inputs of the process. The micro-damage is modelled with a gradient-enhanced damage model which ensures the well-posedness of the boundary value and yields mesh-independent results in computational methods such as FEM. We employ the Adam optimizer and Rectified linear unit activation function for training processes and research into the deep neural network architecture. The performance of our approach is demonstrated through some numerical examples including the three-point bending specimen, shear bending on L-shaped specimen and different failure mechanisms.

Electromechanical coupling devices have been playing an indispensable role in modern engineering. Particularly, flexoelectricity, an electromechanical coupling effect that involves strain gradients, has shown promising potential for future miniaturized electromechanical coupling devices. Therefore, simulation of flexoelectricity is necessary and inevitable. In this paper, we provide an overview of numerical procedures on modeling flexoelectricity. Specifically, we summarize a generalized formulation including the electrostatic stress tensor, which can be simplified to retrieve other formulations from the literature. We further show the weak and discretization forms of the boundary value problem for different numerical methods, including isogeometric analysis and mixed FEM. Several benchmark problems are presented to demonstrate the numerical implementation. The source code for the implementation can be utilized to analyze and develop more complex flexoelectric nano-devices.

This work presents a deep learning method for scene description. (1) Background: This method is part of a larger system, called BlindSys, that assists the visually impaired in an indoor environment. The method detects the presence of certain objects, regardless of their position in the scene. This problem is also known as image multi-labeling. (2) Methods: Our proposed deep learning solution is based on a light-weight pre-trained CNN called SqueezeNet. We improved the SqueezeNet architecture by resetting the last convolutional layer to free weights, replacing its activation function from a rectified linear unit (ReLU) to a LeakyReLU, and adding a BatchNormalization layer thereafter. We also replaced the activation functions at the output layer from softmax to linear functions. These adjustments make up the main contributions in this work. (3) Results: The proposed solution is tested on four image multi-labeling datasets representing different indoor environments. It has achieved results better than state-of-the-art solutions both in terms of accuracy and processing time. (4) Conclusions: The proposed deep CNN is an effective solution for predicting the presence of objects in a scene and can be successfully used as a module within BlindSys.

In this paper, we present a new algorithm for cross-domain classification in aerial vehicle images based on generative adversarial networks (GANs). The proposed method, called Siamese-GAN, learns invariant feature representations for both labeled and unlabeled images coming from two different domains. To this end, we train in an adversarial manner a Siamese encoder–decoder architecture coupled with a discriminator network. The encoder–decoder network has the task of matching the distributions of both domains in a shared space regularized by the reconstruction ability, while the discriminator seeks to distinguish between them. After this phase, we feed the resulting encoded labeled and unlabeled features to another network composed of two fully-connected layers for training and classification, respectively. Experiments on several cross-domain datasets composed of extremely high resolution (EHR) images acquired by manned/unmanned aerial vehicles (MAV/UAV) over the cities of Vaihingen, Toronto, Potsdam, and Trento are reported and discussed.

This paper presents an automatic solution to the problem of detecting and counting cars in unmanned aerial vehicle (UAV) images. This is a challenging task given the very high spatial resolution of UAV images (on the order of a few centimetres) and the extremely high level of detail, which require suitable automatic analysis methods. Our proposed method begins by segmenting the input image into small homogeneous regions, which can be used as candidate locations for car detection. Next, a window is extracted around each region, and deep learning is used to mine highly descriptive features from these windows. We use a deep convolutional neural network (CNN) system that is already pre-trained on huge auxiliary data as a feature extraction tool, combined with a linear support vector machine (SVM) classifier to classify regions into “car” and “no-car” classes. The final step is devoted to a fine-tuning procedure which performs morphological dilation to smooth the detected regions and fill any holes. In addition, small isolated regions are analysed further using a few sliding rectangular windows to locate cars more accurately and remove false positives. To evaluate our method, experiments were conducted on a challenging set of real UAV images acquired over an urban area. The experimental results have proven that the proposed method outperforms the state-of-the-art methods, both in terms of accuracy and computational time.