Search Results (130)

Surface soil moisture (SSM) plays a critical role in climate change, hydrological processes, and agricultural production. Decision trees and deep learning are widely applied to SSM retrieval. The former excels in interpretability while the latter outperforms in generalization, neither, however, integrates both. To address this issue, an attention decision forest (ADF) was developed, comprising feature extractor, soft decision tree, and tree-attention modules. The feature extractor projects raw inputs into a high-dimensional space to reveal nonlinear relationships. The soft decision tree preserves the advantages of tree models in nonlinear partitioning and local feature interaction. The tree-attention module integrates outputs from the soft tree’s subtrees to enhance overall fitting and generalization. Experiments on conterminous United States (CONUS) watershed dataset demonstrate that, upon sample-based validation, ADF outperforms traditional models with an R² of 0.868 and a ubRMSE of 0.041 m³/m³. Further spatiotemporal independent testing demonstrated the robust performance of this method, with R² of 0.643 and0.673, and ubRMSE of 0.062 and 0.065 m³/m³. Furthermore, an evaluation of the interpretability of the ADF using the Shapley Additive Interpretative Model (SHAP) revealed that the ADF was more stable than deep learning methods (e.g., DNN) and comparable to tree-based ensemble learning methods (e.g., RF and XGBoost). Both the ADF and ensemble learning methods demonstrated that, at large scales, spatiotemporal variation had the greatest impact on the SSM, followed by environmental conditions and soil properties. Moreover, the superior spatial SSM maps produced by ADF, compared with GSSM, SMAP L4 and ERA5-Land, further demonstrate ADF’s capability for large-scale mapping. ADF thus offers a novel architecture capable of integrating prediction accuracy, generalization, and interpretability. Full article

Cardiovascular diseases are the leading cause of death worldwide, with arrhythmia being a prevalent and potentially fatal condition. The multi-lead electrocardiogram (ECG) is the primary tool for detecting arrhythmias. However, existing detection methods have shortcomings: they cannot dynamically integrate inter-lead correlations with multi-scale temporal changes in cardiac electrical activity. They also lack mechanisms to simultaneously focus on key leads and time segments, and thus fail to address multi-lead redundancy or capture comprehensive spatial-temporal relationships. To solve these problems, we propose a Multi-Scale Lead-Temporal Co-Attention Framework (MS-LTCAF). Our framework incorporates two key components: a Lead-Temporal Co-Attention Residual (LTCAR) module that dynamically weights the importance of leads and time segments, and a multi-scale branch structure that integrates features of cardiac electrical activity across different time periods. Together, these components enable the framework to automatically extract and integrate features within a single lead, between different leads, and across multiple time scales from ECG signals. Experimental results demonstrate that MS-LTCAF outperforms existing methods. On the PTB-XL dataset, it achieves an AUC of 0.927, approximately 1% higher than the current optimal baseline model (DNN_zhu’s 0.918). On the LUDB dataset, it ranks first in terms of AUC (0.942), accuracy (0.920), and F1-score (0.745). Furthermore, the framework can focus on key leads and time segments through the co-attention mechanism, while the multi-scale branches help capture both the details of local waveforms (such as QRS complexes) and the overall rhythm patterns (such as RR intervals). Full article

The terahertz (THz) frequency range holds critical importance for next-generation, wireless communications and biomedical sensing applications. However, conventional metamaterial design approaches suffer from computationally intensive simulations and optimization processes that can extend over several months. This work presents an intelligent inverse design framework integrating deep neural network (DNN) surrogate modeling with success-history-based adaptive differential evolution (SHADE) for tunable graphene-based THz metasurfaces. Our DNN surrogate model achieves an exceptional coefficient of determination (R² = 0.9984) while providing a four-order-of-magnitude acceleration compared with conventional electromagnetic solvers. The SHADE-integrated framework demonstrates 96.7% accuracy in inverse design tasks with an average convergence time of 10.2 s. The optimized configurations exhibit significant tunability through graphene Fermi level modulation, as validated by comprehensive electromagnetic field analysis. This framework represents a significant advancement in automated electromagnetic design and establishes a robust foundation for intelligent photonic systems across diverse frequency regimes. Full article

With the recent 5G communication technology deployment, Cellular Vehicle-to-Everything (C-V2X) significantly enhances road safety by enabling real-time exchange of critical traffic information among vehicles, pedestrians, infrastructure, and networks. However, further research is required to address real-time application latency and communication reliability challenges. This paper explores integrating cutting-edge C-V2X technology with environmental perception systems to enhance safety at intersections and crosswalks. We propose a multi-module architecture combining C-V2X with state-of-the-art perception technologies, GPS mapping methods, and the client–server module to develop a co-operative perception system for collision avoidance. The proposed system includes the following: (1) a hardware setup for C-V2X communication; (2) an advanced object detection module leveraging Deep Neural Networks (DNNs); (3) a client–server-based co-operative object detection framework to overcome computational limitations of edge computing devices; and (4) a module for mapping GPS coordinates of detected objects, enabling accurate and actionable GPS data for collision avoidance—even for detected objects not equipped with C-V2X devices. The proposed system was evaluated through real-time experiments at the GMMRC testing track at Kettering University. Results demonstrate that the proposed system enhances safety by broadcasting critical obstacle information with an average latency of 9.24 milliseconds, allowing for rapid situational awareness. Furthermore, the proposed system accurately provides GPS coordinates for detected obstacles, which is essential for effective collision avoidance. The technology integration in the proposed system offers high data rates, low latency, and reliable communication, which are key features that make it highly suitable for C-V2X-based applications. Full article

Background: Tumor-educated platelets (TEPs) represent a promising biosource for non-invasive multi-cancer early detection (MCED). While machine learning (ML) has been applied to TEP data, the integration of explainability to reveal gene-level contributions and regulatory associations remains underutilized. This study aims to develop an interpretable ML framework for cancer detection using platelet RNA-sequencing data, combining predictive performance with biological insight. Methods: This study analyzed 2018 TEP RNA samples from 18 tumor types using seven machine learning classifiers. SHAP (Shapley Additive Explanations) was applied for model interpretability, including global feature ranking, local explanation, and gene-level dependence patterns. A weighted SHAP consensus was built by combining model-specific contributions scaled by Area Under the Receiver Operating Characteristic Curve (AUC). Regulatory insights were supported through network analysis using GeneMANIA. Results: Neural models, including shallow Neural Network (NN) and Deep Neural Network (DNN) achieved the best performance (AUC ~0.93), with Extreme Gradient Boosting (XGB) and Support Vector Machine (SVM) also performing well. Early-stage cancers were predicted with high accuracy. SHAP analysis revealed consistent top features (e.g., SLC38A2, DHCR7, IFITM3), while dependence plots uncovered conditional gene interactions involving USF3 (KIAA2018), ARL2, and DSTN. Multi-hop pathway tracing identified NFYC as a shared transcriptional hub across multiple modulators. Conclusions: The integration of interpretable ML with platelet RNA data revealed robust biomarkers and context-dependent regulatory patterns relevant to early cancer detection. The proposed framework supports the potential of TEPs as a non-invasive, information-rich medium for early cancer screening. Full article

Despite the remarkable success of Deep Neural Networks (DNNs) in Remote Sensing Image (RSI) object detection, they remain vulnerable to adversarial attacks. Numerous adversarial attack methods have been proposed for RSI; however, adding a single large-scale adversarial patch to certain high-value targets, which are typically large in physical scale and irregular in shape, is both costly and inflexible. To address this issue, we propose a strategy of using multiple compact patches. This approach introduces two fundamental challenges: (1) how to optimize patch placement for a synergistic attack effect, and (2) how to retain strong adversarial potency within size-constrained mini-patches. To overcome these challenges, we introduce the Spatially Adaptive and Distillation-Enhanced Mini-Patch Attack (SDMPA) framework, which consists of two key modules: (1) an Adaptive Sensitivity-Aware Positioning (ASAP) module, which resolves the placement challenge by fusing the model’s attention maps from both an explainable and an adversarial perspective to identify optimal patch locations, and (2) a Distillation-based Mini-Patch Generation (DMPG) module, which tackles the potency challenge by leveraging knowledge distillation to transfer adversarial information from large teacher patches to small student patches. Extensive experiments on the RSOD and MAR20 datasets demonstrate that SDMPA significantly outperforms existing patch-based attack methods. For example, against YOLOv5n on the RSOD dataset, SDMPA achieves an Attack Success Rate (ASR) of 88.3% using only three small patches, surpassing other patch attack methods. Full article

Rice, wheat, and maize are important food grains consumed by most of the population in Asian countries (like India, Japan, Singapore, Malaysia, China, and Thailand). These crops’ production is affected by biotic and abiotic factors that cause diseases in several parts of the crops (including leaves, stems, roots, nodes, and panicles). A severe infection affects the growth of the plant, thereby undermining the economy of a country, if not detected at an early stage. This may cause extensive damage to crops, resulting in decreased yield and productivity. Early safeguarding methods are overlooked because of farmers’ lack of awareness and the variety of crop diseases. This causes significant crop damage and can consequently lower productivity. In this manuscript, a lightweight vision transformer (MaxViT) with 814.7 K learnable parameters and 85 layers is designed for classifying crop diseases in paddy and wheat. The MaxViT DNN architecture consists of a convolutional block attention module (CBAM), squeeze and excitation (SE), and depth-wise (DW) convolution, followed by a ConvNeXt module. This network architecture enhances feature representation by eliminating redundant information (using CBAM) and aggregating spatial information (using SE), and spatial filtering by the DW layer cumulatively enhances the overall classification performance. The proposed model was tested using a paddy dataset (with 7857 images and eight classes, obtained from local paddy farms in Lalgudi district, Tiruchirappalli) and a wheat dataset (with 5000 images and five classes, downloaded from the Kaggle platform). The model’s classification performance for various diseases has been evaluated based on accuracy, sensitivity, specificity, mean accuracy, precision, F1-score, and MCC. During training and testing, the model’s overall accuracy on the paddy dataset was 99.43% and 98.47%, respectively. Training and testing accuracies were 94% and 92.8%, respectively, for the wheat dataset. Ablation analysis was carried out to study the significant contribution of each module to improving the performance. It was found that the model’s performance was immune to the presence of noise. Additionally, there are a minimal number of parameters involved in the proposed model as compared to pre-trained networks, which ensures that the model trains faster. Full article

With the rapid advancement of the Internet of Things (IoT), artificial intelligence (AI), and blockchain technologies, educational research has increasingly explored smart and personalized learning systems. However, current approaches often suffer from fragmented integration of health monitoring and instructional adaptation, insufficient prediction accuracy of physiological states, and unresolved concerns regarding data privacy and security. To address these challenges, this study introduces SHARP, a novel blockchain-enhanced wireless sensor networks (WSNs) framework designed for real-time student health monitoring and personalized learning in smart educational environments. Wearable sensors enable continuous collection of physiological data, including heart rate variability, body temperature, and stress indicators. A deep neural network (DNN) processes these inputs to detect students’ physical and affective states, while a reinforcement learning (RL) algorithm dynamically generates individualised educational recommendations. A Proof-of-Authority (PoA) blockchain ensures secure, immutable, and transparent data management. Preliminary evaluations in simulated smart classrooms demonstrate significant improvements: the DNN achieves a 94.2% F1-score in state recognition, the RL module reduces critical event response latency, and energy efficiency improves by 23.5% compared to conventional baselines. Notably, intervention groups exhibit a 156% improvement in quiz scores over control groups. Compared to existing solutions, SHARP uniquely integrates multi-sensor physiological monitoring, real-time AI-based personalization, and blockchain-secured data governance in a unified framework. This results in superior accuracy, higher energy efficiency, and enhanced data integrity compared to prior IoT-based educational platforms. By combining intelligent sensing, adaptive analytics, and secure storage, SHARP offers a scalable and privacy-preserving solution for next-generation smart education. Full article

Agriculture in India supports millions of livelihoods and is a major force behind economic expansion. Challenges in modern agriculture depend on environmental factors (such as soil quality and climate variability) and biotic factors (such as pests and diseases). These challenges can be addressed by advancements in technology (such as sensors, internet of things, communication, etc.) and data-driven approaches (such as machine learning (ML) and deep learning (DL)), which can help with crop yield and sustainability in agriculture. This study introduces an innovative deep neural network (DNN) approach for identifying leaf diseases in paddy crops at an early stage. The proposed neural network is a hybrid DL model comprising feature extraction, channel attention, inception with residual, and classification blocks. Channel attention and inception with residual help extract comprehensive information about the crops and potential diseases. The classification module uses softmax to obtain the score for different classes. The importance of each block is analyzed via an ablation study. To understand the feature extraction ability of the modules, extracted features at different stages are fed to the SVM classifier to obtain the classification accuracy. This technique was experimented on eight classes with 7857 paddy crop images, which were obtained from local paddy fields and freely available open sources. The classification performance of the proposed technique is evaluated according to accuracy, sensitivity, specificity, F1 score, MCC, area under curve (AUC), and receiver operating characteristic (ROC). The model was fine-tuned by setting the hyperparameters (such as batch size, learning rate, optimizer, epoch, and train and test ratio). Training, validation, and testing accuracies of 99.91%, 99.87%, and 99.49%, respectively, were obtained for 20 epochs with a learning rate of 0.001 and sgdm optimizer. The proposed network robustness was studied via an ablation study and with noisy data. The model’s classification performance was evaluated for other agricultural data (such as mango, maize, and wheat diseases). These research outcomes can empower farmers with smarter agricultural practices and contribute to economic growth. Full article

Currently, samples are a critical driving force in the application of deep learning. However, the use of samples encounters problems, such as an inconsistent annotation quality, mismatches between images and labels, and a lack of fine-grained labels. Refining sample labels is essential for training a sophisticated model. Refining sample labels through manual verification is labor-intensive, especially for training large models. Additionally, existing label refinement methods based on deep neural networks (DNNs) typically rely on image features to directly predict segmentation results, often overlooking the potential information embedded in existing noisy labels. To address these challenges and shortcomings, this study proposes a novel remote sensing sample label refinement (LR) network, named the identify–update–refine network (IUR-Net). IUR-Net leverages newly acquired remote sensing images and their corresponding noisy labels to automatically identify erroneous regions, update them with more accurate information, and refine the results to improve label quality. A multi-scale, error-aware localization module (Ms-EALM) is designed to capture label–image inconsistencies, enabling the more accurate localization of erroneous label regions. To evaluate the proposed framework, we first constructed and publicly released two benchmark datasets for the label refinement task: WHU-LR and EVLAB-LR. The experimental results on these datasets demonstrate that the labels refined by IUR-Net not only outperform the baseline model in both IoU and F1 scores, but also effectively identify errors in noisy annotations. Full article

The accurate estimation of the remaining useful life (RUL) of lithium-ion batteries (LIBs) is essential for ensuring safety and enabling effective battery health management systems. To address this challenge, data-driven solutions leveraging advanced machine learning and deep learning techniques have been developed. This study introduces a novel framework, Deep Neural Networks with Memory Features (DNNwMF), for predicting the RUL of LIBs. The integration of memory features significantly enhances the model’s accuracy, and an autoencoder is incorporated to optimize the feature representation. The focus of this work is on feature engineering and uncovering hidden patterns in the data. The proposed model was trained and tested using lithium-ion battery cycle life datasets from NASA’s Prognostic Centre of Excellence and CALCE Lab. The optimized framework achieved an impressive RMSE of 6.61%, and with suitable modifications, the DNN model demonstrated a prediction accuracy of 92.11% for test data, which was used to estimate the RUL of Nissan Leaf Gen 01 battery modules. Full article

High demand for 6G wireless has made photonics-aided D-band (110–170 GHz) communication a research priority. Photonics-aided technology integrates optical and wireless communications to boost spectral efficiency and transmission distance. This study presents a Radio-over-Fiber (RoF) communication system utilizing photonics-aided technology for 4600 m long-distance D-band transmission. We successfully show the transmission of a 50 Gbps (25 Gbaud) QPSK signal utilizing a 128.75 GHz carrier frequency. Notwithstanding these encouraging outcomes, RoF systems encounter considerable obstacles, including pronounced nonlinear distortions and phase noise related to laser linewidth. Numerous factors can induce nonlinear impairments, including high-power amplifiers (PAs) in wireless channels, the operational mechanisms of optoelectronic devices (such as electrical amplifiers, modulators, and photodiodes), and elevated optical power levels during fiber transmission. Phase noise (PN) is generated by laser linewidth. Despite the notable advantages of classical Volterra series and deep neural network (DNN) methods in alleviating nonlinear distortion, they display considerable performance limitations in adjusting for phase noise. To address these problems, we propose a novel post-processing approach utilizing a two-dimensional convolutional neural network (2D-CNN). This methodology allows for the extraction of intricate features from data preprocessed using traditional Digital Signal Processing (DSP) techniques, enabling concurrent compensation for phase noise and nonlinear distortions. The 4600 m long-distance D-band transmission experiment demonstrated that the proposed 2D-CNN post-processing method achieved a Bit Error Rate (BER) of 5.3 × 10⁻³ at 8 dBm optical power, satisfying the soft-decision forward error correction (SD-FEC) criterion of 1.56 × 10⁻² with a 15% overhead. The 2D-CNN outperformed Volterra series and deep neural network approaches in long-haul D-band RoF systems by compensating for phase noise and nonlinear distortions via spatiotemporal feature integration, hierarchical feature extraction, and nonlinear modelling. Full article

Depth completion generates a comprehensive depth map by utilizing sparse depth data inputs, supplemented by guidance provided by an RGB image. Deep neural network models depend on annotated datasets for optimal training. However, when the quantity of training data is limited, the generalization capability of deep neural network (DNN)-based methods diminishes considerably. Moreover, acquiring a large dataset of depth maps is challenging and resource intensive. To address these challenges, we introduce a novel few-shot learning approach for depth completion. Our approach integrates noisy-student training with knowledge distillation (KD) techniques to enhance model performance and generalization. We incorporate both the noisy-student training and KD modules into a basic deep regression network using a non-local spatial propagation network (NLSPN) for depth completion. The noisy-student training framework enhances the model’s performance and generalization capabilities by introducing controlled noise and self-learning mechanisms. Within our few-shot learning framework for depth completion, the KD mechanism transfers advanced capabilities from the teacher model to the student model. Experimental evaluations demonstrate that our approach effectively addresses the challenges associated with depth completion tasks, particularly in scenarios with limited training data and a constrained number of available samples. Full article

Currently, finger vein recognition (FVR) stands as a pioneering biometric technology, with convolutional neural networks (CNNs) and Transformers, among other advanced deep neural networks (DNNs), consistently pushing the boundaries of recognition accuracy. Nevertheless, these DNNs are inherently characterized by static, continuous-valued neuron activations, necessitating intricate network architectures and extensive parameter training to enhance performance. To address these challenges, we introduce an adaptive firing threshold-based spiking neural network (ATSNN) for FVR. ATSNN leverages discrete spike encodings to transforms static finger vein images into spike trains with spatio-temporal dynamic features. Initially, Gabor and difference of Gaussian (DoG) filters are employed to convert image pixel intensities into spike latency encodings. Subsequently, these spike encodings are fed into the ATSNN, where spiking features are extracted using biologically plausible local learning rules. Our proposed ATSNN dynamically adjusts the firing thresholds of neurons based on average potential tensors, thereby enabling adaptive modulation of the neuronal input-output response and enhancing network robustness. Ultimately, the spiking features with the earliest emission times are retained and utilized for classifier training via a support vector machine (SVM). Extensive experiments conducted across three benchmark finger vein datasets reveal that our ATSNN model not only achieves remarkable recognition accuracy but also excels in terms of reduced parameter count and model complexity, surpassing several existing FVR methods. Furthermore, the sparse and event-driven nature of our ATSNN renders it more biologically plausible compared to traditional DNNs. Full article

The rapid development of deep neural networks (DNNs), such as convolutional neural networks and transformer-based large language models, has significantly advanced AI applications. However, these advances have introduced substantial computational and data demands, presenting challenges for the development of systolic array accelerators, which excel in tensor operations. Systolic array accelerators are typically developed using two approaches: scale-up, which increases the size of a single array, and scale-out, which involves multiple parallel arrays of fixed size. Scale-up achieves high performance in large-scale matrix multiplications, while scale-out offers better energy efficiency for lower-dimensional matrix multiplications. However, neither approach can simultaneously maintain both high performance and high energy efficiency across the full spectrum of DNN tasks. In this work, we propose a hybrid approach that integrates scale-up and scale-out techniques. We use mapping space exploration in a multi-tenant application environment to assign DNN operations to specific systolic array modules, thereby optimizing performance and energy efficiency. Experiments show that our proposed hybrid systolic array accelerator reduces energy consumption by up to 8% on average and improves throughput by up to 57% on average, compared to TPUv3 across various DNN models. Full article