Search Results (1,628)

Predicting foreign exchange prices is a challenging yet important task due to the complex, volatile, and fluctuating nature of the data. Although deep learning models are efficient, accurate predictions of closing prices and future price directions remain difficult. This study proposes a dual-input deep-learning long short-term memory (LSTM) model for forecasting the EUR/USD closing price and predicting price direction using both fundamental and technical indicators. The model outperforms the second-best model, achieving a 29% reduction in mean absolute error (MAE) and root mean squared error (RMSE) in the training set and reductions of 24% and 23% in MAE and RMSE, respectively, in the test set. These results are confirmed through forecasting simulations, where performance metrics are consistent with those from the training phase. Finally, the model generates reliable three-day price forecasts, providing valuable insights into price direction. Full article

Whole-slide imaging (WSI) in cytopathology poses challenges related to segmentation accuracy, computational efficiency, and image acquisition artifacts. This study aims to evaluate the performance of deep-learning models for instance segmentation in cervical cytology, benchmarking them against state-of-the-art methods on both public and institutional datasets. We tested three architectures—U-Net, vision transformer (ViT), and Detectron2—and evaluated their performance on the ISBI 2014 and CNseg datasets using panoptic quality (PQ), dice similarity coefficient (DSC), and intersection over union (IoU). All models were trained on CNseg and tested on an independent institutional dataset. Data preprocessing involved manual annotation using QuPath, patch extraction guided by GeoJSON files, and exclusion of regions containing less than 60% cytologic material. Our models achieved superior segmentation performance on public datasets, reaching up to 98% PQ. Performance decreased on the institutional dataset, likely due to differences in image acquisition and the presence of blurred nuclei. Nevertheless, the models were able to detect blurred nuclei, highlighting their robustness in suboptimal imaging conditions. In conclusion, the proposed models offer an accurate and efficient solution for instance segmentation in cytology WSI. These results support the development of reliable AI-powered tools for digital cytology, with potential applications in automated screening and diagnostic workflows. Full article

Magnetic flux leakage (MFL) inspection systems are widely used for detecting pipeline defects in industrial sites. However, the acquired MFL signals are affected by field noise, such as electromagnetic interference and mechanical vibrations, which degrade the performance of the developed models. In addition, the noise type or intensity is unknown or changes dynamically during the test phase in contrast to the training phase. To address the above challenges, this paper introduces a novel noise-robust deep-learning framework to remove the noise component in the original signal and learn its noise-invariant feature representation. This can handle the unseen noise pattern and mitigate the impact of dynamic noises on MFL inspection systems. Specifically, we propose a transformer-based architecture for denoising, which encodes noisy input signals into a latent space and reconstructs them into clean signals. We also devise an up–down sampling denoising block to better filter the noise component and generate a noise-invariant representation for weld-defect detection. Finally, extensive experiments demonstrate that the proposed approach effectively improves detection accuracy under both static and dynamic noise conditions, highlighting its value in real-world industrial applications. Full article

Predicting stock prices is a popular area of study within the realms of data mining and machine learning. Precise forecasting can assist investors in mitigating the risks associated with their investments. Given the unpredictable nature of the stock market, influenced by policy changes, stock data often display high levels of fluctuation and randomness, aligning closely with the prevailing market sentiment. Moreover, diverse datasets related to stocks are rich in historical data that can be leveraged to forecast future trends. However, traditional forecasting models struggle to harness this information effectively, which restricts their predictive capabilities and accuracy. To improve the existing issues, this research introduces a novel stock prediction model based on a deep-learning neural network, named after EL-MTSA, which leverages the multifaceted characteristics of stock data along with ensemble learning optimization. In addition, a new evaluation index via market-wide sentiment analysis is designed to enhance the forecasting performance of the stock prediction model by adeptly identifying the latent relationship between the target stock index and dynamic market sentiment factors. Subsequently, many demonstration experiments were conducted on three practical stock datasets, the CSI 300, SSE 50, and CSI A50 indices, respectively. Experiential results show that the proposed EL-MTSA model has achieved a superior predictive performance, surpassing various comparison models. In addition, the EL-MTSA can analyze the impact of market sentiment and media reports on the stock market, which is more consistent with the real trading situation in the stock market, and indicates good predictive robustness and credibility. Full article

To address the challenges in existing carbon fiber prepreg surface defect detection processes, specifically difficulties in small target detection and inaccuracies in elongated defects with large aspect ratios, this study proposes an enhanced YOLOv8s-based models for defect detection on the surface of carbon fiber unidirectional band prepreg. The proposed model respectively integrates two attention mechanisms, a Global Attention Mechanism (GAM) and a Deformable Large Kernel Attention (DLKA) Mechanism, into the architecture of YOLOv8s, which is a lightweight version of YOLO optimized for speed. The attention mechanisms are inserted between the backbone network and the detection head, respectively, to enhance feature extraction before target localization. The YOLOv8s-GAM model achieves a mean average precision (mAP@0.5) of 84.4%, precision of 79.3%, and recall of 78.3%, while the YOLOv8s-DLKA model shows improved performance with mAP@0.5 of 86.4%, precision of 82.4%, and recall of 80.2%. Compared with the original YOLOv8s model, these two modified models demonstrate improvements in mAP@0.5 of 1.6% and 3.6%, precision gains of 0.9% and 4.1%, and recall enhancements of 1.4% and 3.6%, respectively. These models provide technical solutions for precise defect identification on the surface of carbon fiber unidirectional band prepreg. Full article

In crop growth, irrigation has to be adjusted according to developmental stages. Smart agriculture requires the accurate prediction of growth status and timely intervention to improve the quality of agricultural products, but this task faces significant challenges due to variable environmental factors. To address this issue, this study proposes a data-driven irrigation method to enhance crop yield. Our approach harvests extensive datasets to train and optimize an integrated deep-learning architecture combining Informer, Long Short-Term Memory (LSTM) networks, and Exponential Weighted Moving Average (EWMA) models. Controlled greenhouse experiments validated the reliability and practicality of the proposed prediction and intervention strategy. The results showed that the model accurately issued irrigation warnings 3–5 days in advance. Compared to traditional fixed irrigation, the model significantly reduced irrigation frequency while maintaining the same or even better growth conditions. In terms of plant quantity, the experimental group increased by 410.0%, while the control group grew by 50.0%. Additionally, the experimental group’s average plant height was 21.8% higher than that of the control group. These results demonstrate the efficacy of the proposed irrigation prediction method in enhancing crop growth and yield, providing a novel strategy for future agricultural planning and management. Full article

Knitting, a cornerstone of textile manufacturing, is uniquely challenging to automate, particularly in terms of converting fabric designs into precise, machine-readable instructions. This research bridges the gap between textile production and robotic automation by proposing a novel deep learning-based pipeline for reverse knitting to integrate vision-based robotic systems into textile manufacturing. The pipeline employs a two-stage architecture, enabling robots to first identify front labels before inferring complete labels, ensuring accurate, scalable pattern generation. By incorporating diverse yarn structures, including single-yarn (sj) and multi-yarn (mj) patterns, this study demonstrates how our system can adapt to varying material complexities. Critical challenges in robotic textile manipulation, such as label imbalance, underrepresented stitch types, and the need for fine-grained control, are addressed by leveraging specialized deep-learning architectures. This work establishes a foundation for fully automated robotic knitting systems, enabling customizable, flexible production processes that integrate perception, planning, and actuation, thereby advancing textile manufacturing through intelligent robotic automation. Full article

The Internet of Things (IoT) has revolutionized modern communication systems by enabling seamless connectivity among low-power devices. However, the increasing demand for high-performance wireless networks necessitates advanced frameworks that optimize both energy efficiency (EE) and security. Cell-free massive multiple-input multiple-output (CF m-MIMO) has emerged as a promising solution for IoT networks, offering enhanced spectral efficiency, low-latency communication, and robust connectivity. Nevertheless, balancing EE and security in such systems remains a significant challenge due to the stringent power and computational constraints of IoT devices. This study employs secrecy energy efficiency (SEE) as a key performance metric to evaluate the trade-off between power consumption and secure communication efficiency. By jointly considering energy consumption and secrecy rate, our analysis provides a comprehensive assessment of security-aware energy efficiency in CF m-MIMO-based IoT networks. To enhance SEE, we introduce a hybrid deep-learning (DL) framework that integrates convolutional neural networks (CNN) and long short-term memory (LSTM) networks for joint EE and security optimization. The CNN extracts spatial features, while the LSTM captures temporal dependencies, enabling a more robust and adaptive modeling of dynamic IoT communication patterns. Additionally, a multi-objective improved biogeography-based optimization (MOIBBO) algorithm is utilized to optimize hyperparameters, ensuring an improved balance between convergence speed and model performance. Extensive simulation results demonstrate that the proposed MOIBBO-CNN–LSTM framework achieves superior SEE performance compared to benchmark schemes. Specifically, MOIBBO-CNN–LSTM attains an SEE gain of up to 38% compared to LSTM and 22% over CNN while converging significantly faster at early training epochs. Furthermore, our results reveal that SEE improves with increasing AP transmit power up to a saturation point (approximately 9.5 Mb/J at $P_{\mathrm{AP}}^{max}=500$ mW), beyond which excessive power consumption limits efficiency gains. Additionally, SEE decreases as the number of APs increases, underscoring the need for adaptive AP selection strategies to mitigate static power consumption in backhaul links. These findings confirm that MOIBBO-CNN–LSTM offers an effective solution for optimizing SEE in CF m-MIMO-based IoT networks, paving the way for more energy-efficient and secure IoT communications. Full article

Skin cancer, particularly melanoma, is one of the leading causes of cancer-related deaths. It is essential to detect and start the treatment in the early stages for it to be effective and to improve survival rates. This study developed and evaluated a deep learning-based classification model to classify the skin lesion images as benign (non-cancerous) and malignant (cancerous). In this study, we used the ISIC 2016 dataset to train the segmentation model and the Kaggle dataset of 10,000 images to train the classification model. We applied different data pre-processing techniques to enhance the robustness of our model. We used the segmentation model to generate a binary segmentation mask and used it with the corresponding pre-processed image by overlaying its edges to highlight the lesion region, before feeding it to the classification model. We used transfer learning, using ResNet-50 as a backbone model for a feedforward network. We achieved an accuracy of 92.80%, a precision of 98.64%, and a recall of 86.80%. From our study, we have found that integrating deep learning techniques with proper data pre-processing improves the model’s performance. Future work will focus on expanding the datasets and testing more architectures to improve the performance metrics of the model. Full article

In the clinical setting, the efficacy of single-agent immune checkpoint inhibitors (ICIs) in triple-negative breast cancer (TNBC) remains suboptimal. Therefore, there is a pressing need to develop predictive biomarkers to identify non-responders. Considering that cancer-associated fibroblasts (CAFs) represent an integral component of the tumor microenvironment that affects the stiffness of solid tumors on shear-wave elastography (SWE) imaging, wound healing CAFs (WH CAFs) were identified in highly heterogeneous TNBC. This subtype highly expressed vitronectin (VTN) and constituted the majority of CAFs. Moreover, WH CAFs were negatively correlated with CD8⁺ T cell infiltration levels and influenced tumor proliferation in the Eo771 mouse model. Furthermore, multi-omics analysis validated its role in immunosuppression. In order to non-invasively classify patients as responders or non-responders to ICI monotherapy, a deep learning model was constructed to classify the level of WH CAFs based on SWE imaging. As anticipated, this model effectively distinguished the level of WH CAFs in tumors. Based on the classification of the level of WH CAFs, while tumors with a high level of WH CAFs were found to exhibit a poor response to anti programmed cell death protein 1 (PD-1) monotherapy, they were responsive to the combination of anti-PD-1 and erdafitinib, a selective fibroblast growth factor receptor (FGFR) inhibitor. Overall, these findings establish a reference for a novel non-invasive method for predicting ICI efficacy to guide the selection of TNBC patients for precision treatment in clinical settings. Full article

Sleep apnea, characterized by its high prevalence and serious health consequences, faces a critical bottleneck in diagnosis. Polysomnography (PSG), the gold standard, is costly and cumbersome, while wearable devices struggle with quality control and patient compliance, rendering them as unsuitable for both large-scale screening and continuous monitoring. To address these challenges, this research introduces a contactless, low-cost, and accurate event-level sleep apnea detection method leveraging frequency-modulated continuous-wave (FMCW) radar technology. The core of our approach is a novel deep-learning model, built upon the U-Net architecture and augmented with self-attention mechanisms and squeeze-and-excitation (SE) modules, meticulously designed for the precise event-level segmentation of sleep apnea from FMCW radar signals. Crucially, we integrate blood oxygen saturation (SpO₂) prediction as an auxiliary task within a multitask-learning framework to enhance the model’s feature extraction capabilities and clinical utility by capturing physiological correlations between apnea events and oxygen levels. Rigorous evaluation in a clinical dataset, comprising data from 35 participants, with synchronized PSG and radar data demonstrated a performance exceeding that of the baseline methods (Base U-Net and CNN–MHA), achieving a high level of accuracy in event-level segmentation (with an F1-score of 0.8019) and OSA severity grading (91.43%). These findings underscore the significant potential of our radar-based event-level detection system as a non-contact, low-cost, and accurate solution for OSA assessment. This technology offers a promising avenue for transforming sleep apnea diagnosis, making large-scale screening and continuous home monitoring a practical reality and ultimately leading to improved patient outcomes and public health impacts. Full article

Worldwide, heart disease is the leading cause of mortality. Cardiac auscultation, when conducted by a trained professional, is a non-invasive, cost-effective, and readily available method for the initial assessment of cardiac health. Automated heart sound analysis offers a promising and accessible approach to supporting cardiac diagnosis. This work introduces a novel method for classifying heart sounds as normal or abnormal by leveraging time-frequency representations. Our approach combines three distinct time-frequency representations—short-time Fourier transform (STFT), mel-scale spectrogram, and wavelet synchrosqueezed transform (WSST)—to create images that enhance classification performance. These images are used to train five convolutional neural networks (CNNs): AlexNet, VGG-16, ResNet50, a CNN specialized in STFT images, and our proposed CNN model. The method was trained and tested using three public heart sound datasets: PhysioNet/CinC Challenge 2016, CirCor DigiScope Phonocardiogram Dataset 2022, and another open database. While individual representations achieve maximum accuracy of ≈85.9%, combining STFT, mel, and WSST boosts accuracy to ≈99%. By integrating complementary time-frequency features, our approach demonstrates robust heart sound analysis, achieving consistent classification performance across diverse CNN architectures, thus ensuring reliability and generalizability. Full article

Land use and land cover (LULC) classification through remote sensing imagery serves as a cornerstone for environmental monitoring, resource management, and evidence-based urban planning. While Synthetic Aperture Radar (SAR) and optical sensors individually capture distinct aspects of Earth’s surface, their complementary nature SAR excelling in structural and all-weather observation and optical sensors providing rich spectral information—offers untapped potential for improving classification robustness. However, the intrinsic differences in their imaging mechanisms (e.g., SAR’s coherent scattering versus optical’s reflectance properties) pose significant challenges in achieving effective multimodal fusion for LULC analysis. To address this gap, we propose a multimodal deep-learning framework that systematically integrates SAR and optical imagery. Our approach employs a dual-branch neural network, with two fusion paradigms being rigorously compared: the Early Fusion strategy and the Late Fusion strategy. Experiments on the SEN12MS dataset—a benchmark containing globally diverse land cover categories—demonstrate the framework’s efficacy. Our Early Fusion strategy achieved 88% accuracy (F1 score: 87%), outperforming the Late Fusion approach (84% accuracy, F1 score: 82%). The results indicate that optical data provide detailed spectral signatures useful for identifying vegetation, water bodies, and urban areas, whereas SAR data contribute valuable texture and structural details. Early Fusion’s superiority stems from synergistic low-level feature extraction, capturing cross-modal correlations lost in late-stage fusion. Compared to state-of-the-art baselines, our proposed methods show a significant improvement in classification accuracy, demonstrating that multimodal fusion mitigates single-sensor limitations (e.g., optical cloud obstruction and SAR speckle noise). This study advances remote sensing technology by providing a precise and effective method for LULC classification. Full article

This study explores the application of Temporal Fusion Transformers (TFTs) to improve the predictability of hourly potential hydropower production for a small run–of–the–river hydropower plant in Portugal. Accurate hourly power forecasts are essential for optimizing participation in the spot electricity market, where deviations incur penalties. This research introduces the novel application of the TFT, a deep–learning model tailored for time series forecasting and uncovering complex patterns, to predict hydropower production based on meteorological data, historical production records, and plant capacity. Key challenges such as filtering observed hydropower outputs (to remove strong, and unpredictable human influence) and adapting the historical series to installed capacity increases are discussed. An analysis of meteorological information from several sources, including ground information, reanalysis, and forecasting models, was also undertaken. Regarding the latter, precipitation forecasts from the European Centre for Medium–Range Weather Forecasts (ECMWF) proved to be more accurate than those of the Global Forecast System (GFS). When combined with ECMWF data, the TFT model achieved significantly higher accuracy in potential hydropower production predictions. This work provides a framework for integrating advanced machine learning models into operational hydropower scheduling, aiming to reduce classical modeling efforts while maximizing energy production efficiency, reliability, and market performance. Full article

Structural damage detection is essential for civil infrastructure safety. The challenges in noise sensitivity, multi-scale feature extraction, and handling bidirectional temporal dependencies are often encountered by traditional methods such as vibration analysis and computer vision. Although potential solutions are offered by recent deep-learning advancements, limitations are frequently imposed by low interpretability and the incapability to adaptively prioritize crucial features within complex time-series data. To address these, a novel hybrid deep-learning framework is proposed. It is integrated with multi-scale convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) networks, and attention mechanisms. Localized time-frequency features are captured from vibration signals by the CNN using multi-scale kernels. Bidirectional temporal dependencies are skillfully captured by the BiLSTM. The interpretability is improved by the attention mechanism through dynamic feature weighting. Experiments on a simulated steel frame demonstrate that detection accuracy and robustness can be enhanced by this framework. This work promotes structural health monitoring, providing a practical tool for engineering applications. Full article