Diagnosing cardiac diseases relies heavily on electrocardiogram (ECG) analysis, but detecting myocardial infarction-related arrhythmias remains challenging due to irregular heartbeats and signal variations. Despite advancements in machine learning, achieving both high accuracy and low computational cost for arrhythmia classification remains a critical issue. Computer-aided diagnosis systems can play a key role in early detection, reducing mortality rates associated with cardiac disorders. This study proposes a fully automated approach for ECG arrhythmia classification using deep learning and machine learning techniques to improve diagnostic accuracy while minimizing processing time. The methodology consists of three stages: 1) preprocessing, where ECG signals undergo noise reduction and feature extraction; 2) feature identification, where deep convolutional neural network (CNN) blocks, combined with data augmentation and transfer learning, extract key parameters; 3) classification, where a hybrid CNN-SVM model is employed for arrhythmia recognition. CNN-extracted features were fed into a binary support vector machine (SVM) classifier, and model performance was assessed using five-fold cross-validation. Experimental findings demonstrated that the CNN2 model achieved 85.52% accuracy, while the hybrid CNN2-SVM approach significantly improved accuracy to 97.33%, outperforming conventional methods. This model enhances classification efficiency while reducing computational complexity. The proposed approach bridges the gap between accuracy and processing speed in ECG arrhythmia classification, offering a promising solution for real-time clinical applications. Its superior performance compared to nonlinear classifiers highlights its potential for improving automated cardiac diagnosis.

Compact and robust wavelength-tunable mid-infrared fiber lasers are urgently needed in the fields of spectroscopic sensing, polymer processing, and free-space communications. In this work, we experimentally reported a high-power wavelength-tunable Er³⁺/Dy³⁺ codoped fluoride fiber laser by diode clad pumping at 974 ​nm. Adopting a ruled diffraction grating, the laser wavelength could be continuously tuned in the region of 2854 ​nm–3510 nm (656 ​nm) based on the ⁶H_13/2 ​→ ​⁶H_15/2 transition of Dy³⁺, where 3510 ​nm represented the longest wavelength achieved from a Dy³⁺-doped fluoride fiber laser. Within the wide range of 3018 ​nm–3331 nm (312 ​nm), the output power was always kept at >1 ​W, with the maximum power of 1.75 ​W obtained at 3181 ​nm. To the best of our knowledge, this is the first watt-level wavelength-tunable fiber laser in the region of >3 ​μm. Further scaling the power and expanding the tuning range are expected by increasing the pump power while protecting the pumped fiber end.

Arrhythmias stand out for having irregular cardiac rhythms, and the fast diagnosis of arrhythmias holds significant clinical importance due to its potential to mitigate adverse health outcomes. Despite the progress in this field, existing research efforts have encountered limitations, necessitating innovative approaches to address diagnostic challenges effectively. The primary objective of this research is to propose an innovative classification methodology for distinguishing five distinct arrhythmia classes: Atrial premature beat (A), normal (N), ventricular premature beat (V), right bundle branch block (R), and left bundle branch block (L). The proposed methodology involves constructing a hybrid model that incorporates an attention mechanism, utilizing electrocardiogram (ECG) data from an open-source repository. Additionally, we have incorporated an explainability feature into the model, allowing for the interpretation and explanation of its predictions. This model is designed to capitalize on the unique features of arrhythmic patterns and enhance classification metrics. Innovative techniques employed within the methodology are detailed to elucidate the rationale behind their selection and their anticipated contributions to improved model performance. Findings from this study underscore the superiority of the proposed classification model over existing methodologies. Quantitative analysis demonstrates its outstanding performance. The approach, outperforming existing methods, achieves high levels of accuracy (99.16%), specificity (99.79%), recall (99.2 ​%), precision (99.20%), F1-measure (99.16 %), and AUC (99.92%). This research advances medical diagnostics by integrating advanced machine-learning techniques to enhance arrhythmia detection.

Optical wireless (OW) communication systems face significant challenges, such as signal attenuation due to atmospheric absorption, scattering, and noise from hardware components, which degrade detection sensitivity. To address these challenges, we propose a digital processing algorithm that combines finite-impulse response filtering with dynamic synchronization based on pulse addition and subtraction. Unlike conventional methods, which typically rely solely on hardware optimization or basic thresholding techniques, the proposed approach integrates filtering and synchronization to improve weak-signal detection and reduce noise-induced errors. The proposed algorithm was implemented and verified using a field-programmable gate array. Experiments conducted in an indoor OW communication environment demonstrate that the proposed algorithm significantly improves the detection sensitivity by approximately 6 ​dB and 5 ​dB at communication rates of 3.5 Mbps and 5.0 Mbps, respectively. Specifically, under darkroom conditions and a bit error rate of 1 ​× ​10⁻⁷, the detection sensitivity was improved from −38.56 dBm to −44.77 dBm at 3.5 Mbps and from −37.12 dBm to −42.29 dBm at 5 Mbps. The proposed algorithm is crucial for the future capture and tracking of signals at large dispersion angles and in underwater and long-distance communication scenarios.

With the advancement of electronic countermeasures, airborne synthetic aperture radar (SAR) systems are facing increasing challenges in maintaining effective performance in hostile environments. In particular, high-power interference can severely degrade SAR imaging and signal processing, often rendering target detection impossible. This highlights the urgent need for robust anti-interference solutions in both the signal processing and image processing domains. While current methods address interference across various domains, techniques such as waveform modification and spatial filtering typically increase the system costs and complexity. To overcome these limitations, we propose a novel approach that leverages the multi-domain characteristics of interference to efficiently suppress narrowband interference and repeater modulation interference. Specifically, narrowband interference is mitigated using notch filtering, a signal processing technique that effectively filters out unwanted frequencies, while repeater modulation interference is addressed through strong signal amplitude normalization, which enhances both the signal and image processing quality. These methods were validated through tests on real SAR data, demonstrating significant improvements in the imaging performance and system robustness. Our approach offers valuable insights for advancing anti-interference technologies in SAR systems and provides a cost-effective solution to enhance their resilience in complex electronic warfare environments.

Existing image manipulation localization (IML) techniques require large, densely annotated sets of forged images. This requirement greatly increases labeling costs and limits a model’s ability to handle manipulation types that are novel or absent from the training data. To address these issues, we present CLIP-IML, an IML framework that leverages contrastive language-image pre-training (CLIP). A lightweight feature-reconstruction module transforms CLIP token sequences into spatial tensors, after which a compact feature-pyramid network and a multi-scale fusion decoder work together to capture information from fine to coarse levels. We evaluated CLIP-IML on ten public datasets that cover copy-move, splicing, removal, and artificial intelligence (AI)-generated forgeries. The framework raises the average F1-score by 7.85% relative to the strongest recent baselines and secures either first- or second-place performance on every dataset. Ablation studies show that CLIP pre-training, higher resolution inputs, and the multi-scale decoder each make complementary contributions. Under six common post-processing perturbations, as well as the compression pipelines used by Facebook, Weibo, and WeChat, the performance decline never exceeds 2.2%, confirming strong practical robustness. Moreover, CLIP-IML requires only a few thousand annotated images for training, which markedly reduces data-collection and labeling effort compared with previous methods. All of these results indicate that CLIP-IML is highly generalizable for image tampering localization across a wide range of tampering scenarios.

Deep neural networks (DNNs) have found extensive applications in safety-critical artificial intelligence systems, such as autonomous driving and facial recognition systems. However, recent research has revealed their susceptibility to backdoors maliciously injected by adversaries. This vulnerability arises due to the intricate architecture and opacity of DNNs, resulting in numerous redundant neurons embedded within the models. Adversaries exploit these vulnerabilities to conceal malicious backdoor information within DNNs, thereby causing erroneous outputs and posing substantial threats to the efficacy of DNN-based applications. This article presents a comprehensive survey of backdoor attacks against DNNs and the countermeasure methods employed to mitigate them. Initially, we trace the evolution of the concept from traditional backdoor attacks to backdoor attacks against DNNs, highlighting the feasibility and practicality of generating backdoor attacks against DNNs. Subsequently, we provide an overview of notable works encompassing various attack and defense strategies, facilitating a comparative analysis of their approaches. Through these discussions, we offer constructive insights aimed at refining these techniques. Finally, we extend our research perspective to the domain of large language models (LLMs) and synthesize the characteristics and developmental trends of backdoor attacks and defense methods targeting LLMs. Through a systematic review of existing studies on backdoor vulnerabilities in LLMs, we identify critical open challenges in this field and propose actionable directions for future research.