End-to-End System for Estimating Crystallization Kinetics Using a Deep Learning-Based Approach

Remote sensing is a key tool to derive glacier surface velocities but existing mapping methods, such as cross-correlation techniques, can fail where surface properties change temporally or where large velocity variations occur spatially. High-resolution datasets, such as UAV imagery, offer a promising solution to tackle these issues and to study small-scale glacier dynamics, but new workflows are required to handle such data. Therefore, we tested the potential of new deep learning-based image-matching algorithms for deriving glacier surface velocities across the ablation area of a glacier with strong spatial variability in surface velocities (<5 m/yr to >100 m/yr) and substantial changes in surface properties between image acquisitions. For a thorough comparison of state-of-the-art methods and sensors, we applied three different techniques (cross-correlation using geoCosiCorr3D, feature tracking with ORB using SeaIceDrift and the new deep learning-based method using ICEpy4D) and three different platforms (Sentinel-2, PlanetScope, UAVs) to estimate glacier surface velocities. Results showed lowest errors for velocities derived with the deep learning-based approach applied to UAV imagery (RMSE = 2.17 m/yr, R2 = 0.99), followed by cross-correlation using Sentinel-2 images (RMSE = 21.0 m/yr, R2 = 0.59) and the deep learning-based approach with PlanetScope data (RMSE = 21.28 m/yr, R2 = 0.36). Cross-correlation with geoCosiCorr3D resulted in comparably high errors with the UAV dataset (RMSE = 36.22 m/yr, R2 = 0.24), whereas ORB-based feature tacking showed lowest performance with all sensors. Spatial patterns of computed velocities indicate that applying existing cross-correlation methods for areas with regular displacements or low glacier velocities yields suitable results on UAV data, but innovative deep learning-based approaches are required for resolving rapid changes in velocities or in surface properties. This novel method benefits from improved keypoint detection and matching through training using neural networks and data characterized by challenging geometries, outlier minimization and more robust descriptors by applying cross-attention layers. We conclude that continued development of deep learning-based feature tracking approaches for glacier velocity computations may substantially improve UAV-based velocity derivations applied to challenging situations. This method is able to deliver reliable displacement data in situations where traditional methods fail, which implies a new level of detail in understanding and interpreting glacier dynamics.

One of the critical issues resulting from the increasing penetration of social media is the spread of fake news. This can damage public information and influence mass opinion, leading to conflict. To overcome this problem, machine learning and deep learning-based approaches have been continuously developed to detect fake news on various social media platforms automatically. This article aims to compare the effectiveness of these two approaches in detecting fake news. The methods used include the implementation of traditional machine learning algorithms, such as Support Vector Machines (SVM) and Random Forest, as well as deep learning-based approaches, including Long Short-Term Memory and Self-Organizing Maps. Datasets containing real and fake news from various social media sources are used to train and evaluate these models. Model performance is measured based on accuracy, precision, recall, and F1-score. This study aims to determine which approach is more effective and identify challenges in implementing these algorithms in a dynamic social media environment. The results obtained show that the Random Forest algorithm achieves an accuracy level of 100%, surpassing other algorithms, including Long Short-Term Memory with an F-1 Score of 97%, Self-Organizing Map with an F-1 Score of 96%, and Support Vector Machine with an F-1 Score of 92%.

Osteoarthritis is a condition that poses a risk to the knee joint, resulting in pain and impaired function. However, traditional knee X-ray evaluations using the Kellgren-Lawrence grading system have proven to be inefficient. These evaluations are subjective, time-consuming, and labor-intensive, particularly in busy hospital settings. The objective of this research was to present a deep learning-based approach that can detect knee joint regions in medical images. By addressing the limitations of traditional methods, the aim was to develop a more efficient and automated approach for knee joint analysis. The proposed method utilizes the Faster R-CNN model, which consists of a region proposal network (RPN) and Fast R-CNN. The RPN generates region proposals that potentially contain knee joint regions, while the Fast R-CNN network categorizes and extracts features from these proposals. To train the model, a dataset of knee joint images was employed. The performance of the model was evaluated using metrics, such as accuracy, precision, recall, F1-score, and mean IoU (Intersection Over Union). The results demonstrated the high accuracy of the proposed method in detecting knee joint regions. The model achieved a mean IoU of 94.5, indicating a strong overlap between the predicted and ground truth regions. These findings highlight the potential of deep learning-based approaches in automating medical image analysis, specifically in the diagnosis and management of knee joint disorders. This study emphasizes the significance of leveraging advanced technologies, such as deep learning, in medical imaging. By developing more efficient and accurate methods for identifying knee joint regions in medical images, it becomes feasible to enhance patient outcomes and healthcare delivery. The proposed deep learning-based approach showcases promising results, paving the way for further advancements in the field of medical image analysis and contributing to improved diagnostic capabilities for knee joint disorders.

Optimal bidding strategy of a virtual power plant in day-ahead energy and frequency regulation markets: A deep learning-based approach

Object detection and recognition is an essential task in computer vision with numerous real-world applications such as surveillance, self-driving cars, and robotics. In recent years, deep learning-based approaches have significantly improved the accuracy and speed of object detection and recognition. The You Only Look Once version 3 (YOLOv3) algorithm is a popular deep learning-based approach that can detect and recognize objects in real-time. The Common Objects in Context (COCO) dataset is a large-scale dataset with over 330,000 labeled images and more than 2.5 million object instances, making it a popular choice for object detection and recognition tasks. In this paper, we propose a deep learning-based approach for real-time object detection and recognition using the YOLOv3 architecture and COCO dataset. We evaluate our approach based on several performance metrics, including mean average precision (mAP), frames per second (FPS), total object detection time, object detection accuracy, false positive rate, number of detected objects, and mean intersection over union (mIoU). Our results show that our approach achieves a mean average precision of 0.76 on the COCO dataset and a real-time performance of 40 frames per second on a single GPU. Additionally, our approach achieves an object detection accuracy of 93.5%, a false positive rate of 6.5%, and a mean intersection over union of 0.65. Our proposed approach shows promising results for real-time object detection and recognition and can be applied to various real-world applications.

In the OFDM-based digital terrestrial broadcasting systems, impulsive noise is a significant factor affecting communication quality. A prominent method to suppress impulsive noise is to incorporate a memoryless nonlinearity at the receiver front-end of the OFDM demodulator, in which parameter estimation of memoryless nonlinearity directly impact the effectiveness of impulsive noise suppression. In this paper, we proposes a deep learning-based memoryless nonlinearity approach for impulsive noise suppression. The proposed method can adaptively estimate the parameters of the memoryless nonlinearity in dynamic impulsive noise environments and achieve totically-optimal parameter estimation. To specific, we design a High-Amplitude Priority Downsampling method to extract the key amplitude characteristics from the input signal, which effectively resolves the issue of extracting amplitude features of impulsive noise. Besides, to address the issue of performance degradation due to insufficient training samples, we propose a novel training method that integrates progressive fine-tuning to complete the training only using few samples. Furthermore, we conduct experiments on signal-to-noise ratio (SNR) and bit error rate (BER) of the signal after impulsive noise suppression. The results validate that the parameters estimated by the proposed method can approximate the theoretical optimal values and the proposed method can effectively suppress impulsive noise and outperform the traditional methods in terms of SNR and BER.

Predicting the Remaining Useful Life (RUL) of turbofan engines can prevent air disasters caused by component degradation. It is an important procedure in prognostics and health management (PHM). Therefore, a deep learning-based RUL prediction approach is proposed. The CMAPSS benchmark dataset is used to determine the RUL of aviation engines, focusing specifically on the FD001 and FD003 sub-datasets.In this study, we propose a CAELSTM (Convolutional Autoencoder and Attention-based LSTM) hybrid model for RUL prediction. First, the sub-datasets are preprocessed, and a piecewise linear degradation model is applied. The proposed model utilizes an autoencoder followed by an LSTM layer with an attention mechanism, which focuses on the most relevant components of the sequences. A fully connected layer of the convolutional neural network is used to further process the important features. Finally, the proposed model is evaluated and compared with other approaches. The results show that the model surpasses state-of-the-art methods, achieving RMSE values of 14.44 and 13.40 for FD001 and FD003, respectively. Other evaluation criteria, such as MAE and scoring, were also used, with MAE achieving values of 10.49 and 10.68 for FD001 and FD003, respectively. The scoring achieved values of 282.38 and 264.47 for the same sub-datasets. These results highlight the model's promise for improving prognostics and health management (PHM) systems, offering a dependable tool for predictive maintenance in aerospace and related fields. They also demonstrate the effectiveness and superiority of the model in enhancing aviation safety.

This paper proposes a fuzzy-label weighted deep learning-based image classification approach for assessing computed tomography (CT) image quality. More specifically, we want to determine if a captured CT image passes Quality Assessment (QA) with a certain radiation dose. Our contributions here include proposing a fuzzy-label weighting method and introduces the concept of a "fuzzy-label" (to reflect the confidence of the ground-truth annotation by annotator) to aid in the training of deep learning-based image classification. We proposed an ensemble/assimilation method to determine the image quality at entire CT image-level using CT windowing (i.e. clipping of the CT image to 8-bit grayscale with respect to the various window-width (WW) and window-length (WL)), similar to what a human would do manually to assess the CT image quality in the factory setting. Experimental results showed that our proposed fuzzy-label weighted deep learning-based image classification approach (trained using annotations provided by one single annotator) significantly outperforms that of its traditional baseline image classification approach.

Monitoring the CO2 plume evolution is essential for ensuring geologic storage security and integrity. Traditional numerical simulation-based data assimilation workflow is computationally expensive, and this is further complicated by the fact that geologic uncertainty must be incorporated for robust performance prediction. Therefore, reservoir simulation and model calibration accounting for geologic uncertainty are not amenable to real time monitoring of the CO2 plume evolution for large-scale applications. We propose a deep learning-based approach which enables near real time CO2 plume visualization and rapid data assimilation incorporating multiple geological realizations for predicting future CO2 plume evolution and area of review (AOR) determination. The proposed deep learning-based data assimilation framework considers geological uncertainty utilizing multiple plausible models for training data generation. Rather than utilizing all available geologic realizations, a representative subset is sampled using dissimilarity measures in flow patterns computed via multidimensional scaling (MDS) and streamline time-of-flight. The approach substantially reduces training data generation cost while preserving the uncertainty inherent in the original ensemble of geomodels. The CO2 plume evolution is represented using ‘onset time’ images, depicting the calendar time when the CO2 saturation exceeds a prespecified threshold value at a given location. The use of a single CO2 onset time image instead of multiple CO2 saturation snapshots across different timesteps significantly reduces the dimensionality of the problem, making the deep learning model robust and scalable for large-scale field applications. A variational autoencoder encodes the onset time images into latent variables, which are predicted by another neural networks using the available monitoring data. The power and efficacy of the proposed method are demonstrated through application to a large-scale field case, the Illinois Basin-Decatur Carbon Sequestration Project. The available monitoring data consists of bottom-hole pressure at the injector, distributed pressure data and CO2 saturation log data at the monitoring well. Out of 200 geostatistical realizations, 10 representative models are selected by the MDS while preserving diversity of the geologic model. Additional calibration parameters including transmissibility and pore volume multipliers are applied to the selected realizations for generating a comprehensive training dataset. The trained ML model is then employed for reservoir model calibration, significantly accelerating the calibration process and enabling real time CO2 plume imaging from the monitoring data. The trained deep learning model achieves history matching of both pressure and saturation responses in seconds. The calibrated models are then used for forecasting future CO2 plume migration and the AOR assessment. The deep learning-based data assimilation approach enables near real time monitoring and verification of field-scale CO2 sequestration projects while accounting for geologic uncertainty. Utilizing the trained deep learning model, reservoir model calibration and prediction of CO2 plume evolution is performed within seconds, orders of magnitude faster compared to traditional history matching.

BGNN4VD: Constructing Bidirectional Graph Neural-Network for Vulnerability Detection

Foot placement prediction can be important for exoskeleton and prosthesis controllers, human-robot interaction, or body-worn systems to prevent slips or trips. Previous studies investigating foot placement prediction have been limited to predicting foot placement during the swing phase, and do not fully consider contextual information such as the preceding step or the stance phase before push-off. In this study, we propose a deep learning-based foot placement prediction approach, sequentially processing data from three IMU sensors mounted on the pelvis and feet. The raw sensor data are pre-processed to generate multi-variable time-series data for training two deep learning models, where the first model estimates the gait progression and the second model subsequently predicts the next foot placement. The ground truth gait phase data and foot placement data are acquired from a motion capture system. Ten healthy subjects were invited to walk naturally at different speeds on a treadmill. In cross-subject learning, the trained models had a mean distance error of 5.93 cm for foot placement prediction. In single-subject learning, the prediction accuracy improved with additional training data, and a mean distance error of 2.60 cm was achieved by fine-tuning the cross-subject validated models with the target subject data. Even from 25–81% in the gait cycle, mean distance errors were only 6.99 cm and 3.22 cm for cross-subject learning and single-subject learning, respectively.

Knowing which residues of a protein are important for its function is of paramount importance for understanding the molecular basis of this function and devising ways of modifying it for medical or biotechnological applications. Due to the difficulty in detecting these residues experimentally, prediction methods are essential to cope with the sequence deluge that is filling databases with uncharacterized protein sequences. Deep learning approaches are especially well suited for this task due to the large amounts of protein sequences for training them, the trivial codification of this sequence data to feed into these systems, and the intrinsic sequential nature of the data that makes them suitable for language models. As a consequence, deep learning-based approaches are being applied to the prediction of different types of functional sites and regions in proteins. This review aims to give an overview of the current landscape of methodologies so that interested users can have an idea of which kind of approaches are available for their proteins of interest. We also try to give an idea of how these systems work, as well as explain their limitations and high dependence on the training set so that users are aware of the quality of expected results.

Several studies have shown that the phone number and call behavior generated by a phone call reveal the type of phone call. By analyzing the phone number rules and call behavior patterns, we can recognize the fraudulent phone call. The success of this recognition heavily depends on the particular set of features that are used to construct the classifier. Since these features are human-labor engineered, any change introduced to the telephone fraud can render these carefully constructed features ineffective. In this paper, we show that we can automate the feature engineering process and, thus, automatically recognize the fraudulent phone call by applying our proposed novel approach based on deep learning. We design and construct a new classifier based on Call Detail Records (CDR) for fraudulent phone call recognition and find that the performance achieved by our deep learning-based approach outperforms competing methods. Experimental results demonstrate the effectiveness of the proposed approach. Specifically, in our accuracy evaluation, the obtained accuracy exceeds 99%, and the most performant deep learning model is 4.7% more accurate than the state-of-the-art recognition model on average. Furthermore, we show that our deep learning approach is very stable in real-world environments, and the implicit features automatically learned by our approach are far more resilient to dynamic changes of a fraudulent phone number and its call behavior over time. We conclude that the ability to automatically construct the most relevant phone number features and call behavior features and perform accurate fraudulent phone call recognition makes our deep learning-based approach a precise, efficient, and robust technique for fraudulent phone call recognition.

Deep learning-based approaches have achieved state-of-the-art performances in numerous application domains and require training data to be centrally available. However, training data may not be centrally available in domains like medical data imaging; further, the hosting institutions/clients may be reluctant to share the data owing to security and privacy issues. Federated learning trains several local models at the institution level collaboratively and then aggregates the knowledge obtained to get a global model that generalizes to data from these participating institutions without the need for data sharing among them. This work proposes a novel federated learning-based model for malaria diagnosis using stem cells. The novelty lies in fine-tuning pre-trained models, including Xception, Resnet50 and VGG19, for the malaria diagnosis and using the most generalizable model in the federated learning setup. Additionally, we propose two novel parameter aggregation strategies, namely average weights aggregation and weighted averaged weights aggregation, to get the global model. Experiments conducted on the Malaria Cells Images Dataset confirm the usefulness of the proposed federated learning-based approach; specifically, the performance of the Xception backbone-based global model is 100% with weighted averaged weights aggregation scheme. The performance of the ResNet50-based federated model reports an accuracy of 96.0% and remains competitive with the state-of-the-art models even when the latter models are trained centrally. These results show that the proposed methodology is helpful in medical data imaging applications where institutions have limited training data and are reluctant to share their data due to inherent security and privacy issues.

Multi-robot exploration—i.e., the problem of mapping unknown features of an environment—is fundamental in many tasks, including search and rescue, planetary exploration, and environmental monitoring. This article surveys recent literature with the aim at identifying current research trends, open challenges, and future directions. Since the first formalization of the exploration problem, current research has extended systems and algorithms to map 3D environments and more complex phenomena and considered real-world constraints. Deep learning-based approaches have seen some preliminary applications in decision-making. While current research has been progressing towards systems that can work in the real world, long-term exploration in large unstructured environments with complex phenomena to map is still an open problem. This introduces opportunities for exciting research, including but not limited to generalized frameworks that bridge the gap between theory and practice, learning-based approaches, and robustness to failure and attacks, so that robots can be deployed safely in real-world environments.