Search Results (10,566)

In this research, a multi-channel target speech enhancement scheme is proposed that is based on deep learning (DL) architecture and assisted by multi-source tracking using a labeled random finite set (RFS) framework. A neural network based on minimum variance distortionless response (MVDR) beamformer is considered as the beamformer of choice, where a residual dense convolutional graph-U-Net is applied in a generative adversarial network (GAN) setting to model the beamformer for target speech enhancement under reverberant conditions involving multiple moving speech sources. The input dataset for this neural architecture is constructed by applying multi-source tracking using multi-sensor generalized labeled multi-Bernoulli (MS-GLMB) filtering, which belongs to the labeled RFS framework, to obtain estimations of the sources’ positions and the associated labels (corresponding to each source) at each time frame with high accuracy under the effect of undesirable factors like reverberation and background noise. The tracked sources’ positions and associated labels help to correctly discriminate the target source from the interferers across all time frames and generate time–frequency (T-F) masks corresponding to the target source from the output of a time-varying, minimum variance distortionless response (MVDR) beamformer. These T-F masks constitute the target label set used to train the proposed deep neural architecture to perform target speech enhancement. The exploitation of MS-GLMB filtering and a time-varying MVDR beamformer help in providing the spatial information of the sources, in addition to the spectral information, within the neural speech enhancement framework during the training phase. Moreover, the application of the GAN framework takes advantage of adversarial optimization as an alternative to maximum likelihood (ML)-based frameworks, which further boosts the performance of target speech enhancement under reverberant conditions. The computer simulations demonstrate that the proposed approach leads to better target speech enhancement performance compared with existing state-of-the-art DL-based methodologies which do not incorporate the labeled RFS-based approach, something which is evident from the 75% ESTOI and PESQ of 2.70 achieved by the proposed approach as compared with the 46.74% ESTOI and PESQ of 1.84 achieved by Mask-MVDR with self-attention mechanism at a reverberation time (RT60) of 550 ms. Full article

Background/Objectives: Even with today’s advancements, cancer still represents a major cause of mortality worldwide. One important aspect of cancer progression that has a big impact on diagnosis, prognosis, and treatment plans is accurate lymph node metastasis evaluation. However, regardless of the imaging method used, this process is challenging and time-consuming. This research aimed to develop and validate an automatic detection and segmentation system for superficial lymph node evaluation based on multimodal ultrasound images, such as traditional B-mode, Doppler, and elastography, using deep learning techniques. Methods: The suggested approach incorporated a Mask R-CNN architecture designed specifically for the detection and segmentation of lymph nodes. The pipeline first involved noise reduction preprocessing, after which morphological and textural feature segmentation and analysis were performed. Vascularity and stiffness parameters were further examined in Doppler and elastography pictures. Metrics, including accuracy, mean average precision (mAP), and dice coefficient, were used to assess the system’s performance during training and validation on a carefully selected dataset of annotated ultrasound pictures. Results: During testing, the Mask R-CNN model showed an accuracy of 92.56%, a COCO AP score of 60.7 and a validation score of 64. Furter on, to improve diagnostic capabilities, Doppler and elastography data were added. This allowed for improved performance across several types of ultrasound images and provided thorough insights into the morphology, vascularity, and stiffness of lymph nodes. Conclusions: This paper offers a novel use of deep learning for automated lymph node assessment in ultrasound imaging. This system offers a dependable tool for doctors to evaluate lymph node metastases efficiently by fusing sophisticated segmentation techniques with multimodal image processing. It has the potential to greatly enhance patient outcomes and diagnostic accuracy. Full article

Cardiovascular diseases (CVDs) are the leading cause of death worldwide, claiming over 17 million lives annually. Early detection of conditions like heart murmurs, often indicative of heart valve abnormalities, is critical for improving patient outcomes. Traditional diagnostic methods, including physical auscultation and advanced imaging techniques, are constrained by their reliance on specialized clinical expertise, inherent procedural invasiveness, substantial financial costs, and limited accessibility, particularly in resource-limited healthcare environments. This study presents a novel convolutional recurrent neural network (CRNN) model designed for the non-invasive classification of heart murmurs. The model processes heart sound recordings using advanced pre-processing techniques such as z-score normalization, band-pass filtering, and data augmentation (Gaussian noise, time shift, and pitch shift) to enhance robustness. By combining convolutional and recurrent layers, the CRNN captures spatial and temporal features in audio data, achieving an accuracy of 90.5%, precision of 89%, and recall of 87%. These results underscore the potential of machine-learning technologies to revolutionize cardiac diagnostics by offering scalable, accessible solutions for the early detection of cardiovascular conditions. This approach paves the way for broader applications of AI in healthcare, particularly in underserved regions where traditional resources are scarce. Full article

The deep learning architecture YOLO (You Only Look Once) has demonstrated its superior visual detection performance in various computer vision tasks and has been widely applied in the field of automatic surface defect detection. In this paper, we propose a lightweight YOLOv8-based method for the quality inspection of car body welding spots. We developed a TA-YOLOv8 network structure which has an improved Task-Aligned (TA) head detection, designed to handle a small sample size, imbalanced positive and negative samples, and high-noise characteristics of Body-in-White welding spot data. By learning with fewer parameters, the model achieves more efficient and accurate classification. Additionally, our algorithm framework can perform anomaly segmentation and classification on our open-world raw datasets obtained from actual production environments. The experimental results show that the lightweight module improves the processing speed by an average of 2.8%, with increases in detection the mAP@50-95 and recall rate of 1.35% and 0.1226, respectively. Full article

Speaker recognition is essential in smart voice applications for personal identification. Current state-of-the-art techniques primarily focus on ideal acoustic conditions. However, the traditional spectrogram struggles to differentiate between noise, reverberation, and speech. To overcome this challenge, MFCC can be replaced with the output from a self-supervised learning model. This study introduces a TDNN enhanced with a pre-trained model for robust performance in noisy and reverberant environments, referred to as PNR-TDNN. The PNR-TDNN employs HuBERT as its backbone, while the TDNN is an improved ECAPA-TDNN. The pre-trained model employs the Canopy/Mini Batch k-means++ strategy. In the TDNN architecture, several enhancements are implemented, including a cross-channel fusion mechanism based on Res2Net. Additionally, a non-average attention mechanism is applied to the pooling operation, focusing on the weight information of each channel within the Squeeze-and-Excitation Net. Furthermore, the contribution of individual channels to the pooling of time-domain frames is enhanced by substituting attentive statistics with multi-head attention statistics. Validated by zhvoice in noisy conditions, the minimized PNR-TDNN demonstrates a 5.19% improvement in EER compared to CAM++. In more challenging environments with noise and reverberation, the minimized PNR-TDNN further improves EER by 3.71% and 9.6%, respectively, and MinDCF by 3.14% and 3.77%, respectively. The proposed method has also been validated on the VoxCeleb1 and cn-celeb_v2 datasets, representing a significant breakthrough in the field of speaker recognition under challenging conditions. This advancement is particularly crucial for enhancing safety and protecting personal identification in voice-enabled microphone applications. Full article

It has been demonstrated that interactive speech and noise modeling outperforms traditional speech modeling-only methods for speech enhancement (SE). With a dual-branch topology that simultaneously predicts target speech and noise signals and employs bidirectional information communication between the two branches, the quality of the enhanced speech is significantly improved. However, the dual-branch topology greatly increases the model complexity and deployment cost, thus limiting its practicality. In this paper, we propose UniInterNet, a unidirectional information interaction-based dual-branch network to achieve noise modeling-assisted SE without any increase in complexity. Specifically, the noise branch still receives information from the speech branch to achieve more accurate noise modeling. Subsequently, the noise modeling results are utilized to assist the learning of the speech branch during backpropagation, while the speech branch no longer receives the auxiliary information from the noise branch, so only the speech branch is required during model deployment. Experimental results demonstrate that under the causal inference condition, the performance of UniInterNet only marginally decreases compared to the corresponding bidirectional information interaction scheme, while the model inference complexity is reduced by about 75%. With comparable overall performance, UniInterNet also outperforms previous interactive speech and noise modeling-based benchmarks in terms of causal inference and model complexity. Furthermore, UniInterNet surpasses other existing competitive methods. Full article

This paper examines federated learning, a decentralized machine learning paradigm, focusing on privacy challenges. We introduce differential privacy mechanisms to protect privacy and quantify their impact on global model performance. Using evolutionary game theory, we establish a framework to analyze strategy dynamics and define utilities for different strategies based on Gaussian noise powers and training iterations. A differential privacy federated learning model (DPFLM) is analyzed within this framework. A key contribution is the thorough existence and stability analysis, identifying evolutionarily stable strategies (ESSs) and confirming their stability through simulations. This research provides theoretical insights for enhancing privacy protection in federated learning systems. Full article

With the continuous development of industrial automation, diesel engines play an increasingly important role in various types of construction machinery and power generation equipment. Improving the dynamic and static performance of the speed control system of single-cylinder diesel engines can not only significantly improve the efficiency of the equipment, but also effectively reduce energy consumption and emissions. Particle swarm optimization (PSO) fuzzy PID control algorithms have been widely used in many complex engineering problems due to their powerful global optimization capability and excellent adaptability. Currently, PSO-based fuzzy PID control research mainly integrates hybrid algorithmic strategies to avoid the local optimum problem, and lacks optimization of the dynamic noise suppression of the input error and the rate of change of the error. This makes the algorithm susceptible to the coupling of the system uncertainty and measurement disturbances during the parameter optimization process, leading to performance degradation. For this reason, this study proposes a new framework based on the synergistic optimization of the untraceable Kalman filter (UKF) and PSO fuzzy PID control for the speed control system of a single-cylinder diesel engine. A PSO-optimized fuzzy PID controller is designed by obtaining accurate speed estimation data using the UKF. The PSO is capable of quickly adjusting the fuzzy PID parameters so as to effectively alleviate the nonlinearity and uncertainty problems during the operation of diesel engines. By establishing a Matlab/Simulink simulation model, the diesel engine speed step response experiments (i.e., startup experiments) and load mutation experiments were carried out, and the measurement noise and process noise were imposed. The simulation results show that the optimized diesel engine speed control system is able to reduce the overshoot by 76%, shorten the regulation time by 58%, and improve the noise reduction by 25% compared with the conventional PID control. Compared with the PSO fuzzy PID control algorithm without UKF noise reduction, the optimized scheme reduces the overshoot by 20%, shortens the regulation time by 48%, and improves the noise reduction effect by 23%. The results show that the PSO fuzzy PID control method with integrated UKF has superior control performance in terms of system stability and accuracy. The algorithm significantly improves the responsiveness and stability of diesel engine speed, achieves better control effect in the optimization of diesel engine speed control, and provides a useful reference for the optimization of other diesel engine control systems. In addition, this study establishes the GT-POWER model of a 168 F single-cylinder diesel engine, and compares the cylinder pressure and fuel consumption under four operating conditions through bench tests to ensure the physical reasonableness of the kinetic input parameters and avoid algorithmic optimization on the distorted front-end model. Full article

Synthetic aperture radar (SAR) serves as a pivotal remote sensing technology, offering critical support for ship monitoring, environmental observation, and national defense. Although optical detection methods have achieved good performance, SAR imagery still faces challenges, including speckle, complex backgrounds, and small, dense targets. Reducing false alarms and missed detections while improving detection performance remains a key objective in the field. To address these issues, we propose SMEP-DETR, a transformer-based model with multi-edge enhancement and parallel dilated convolutions. This model integrates a speckle denoising module, a multi-edge information enhancement module, and a parallel dilated convolution and attention pyramid network. Experimental results demonstrate that SMEP-DETR achieves the high $mAP$ 98.6% on SSDD, 93.2% in HRSID, and 80.0% in LS-SSDD-v1.0, surpassing several state-of-the-art algorithms. Visualization results validate the model’s capability to effectively mitigate the impact of speckle noise while preserving valuable information in both inshore and offshore scenarios. Full article

This study describes a sensor system for continuous monitoring of volatile organic compounds (VOCs) emitted from small industrial facilities in urban centers, such as automobile paint facilities and printing facilities. Previously, intermittent measurements were made using expensive flame ionization detector (FID)-type instruments that were impossible to install, resulting in a lack of continuous management. This paper develops a low-cost sensor system for full-time management and consists of multi-sensor systems to increase the spatial resolution in the pipe. To improve the accuracy and reliability of this system, a new reinforced adaptive neuro fuzzy inference system (RANFIS) model with enhanced preprocessing based on the adaptive neuro fuzzy inference system (ANFIS) model is proposed. For this purpose, a smart sensor module consisting of low-cost metal oxide semiconductors (MOSs) and photo-ionization detectors (PIDs) is fabricated, and an operating controller is configured for real-time data acquisition, analysis, and evaluation. In the front part of the RANFIS, interquartile range (IQR) is used to remove outliers, and gradient analysis is used to detect and correct data with abnormal change rates to solve nonlinearities and outliers in sensor data. In the latter stage, the complex nonlinear relationship of the data was modeled using the ANFIS to reliably handle data uncertainty and noise. For practical verification, a toluene evaporation chamber with a sensor system for monitoring was built, and the results of real-time data sensing after training based on real data were compared and evaluated. As a result of applying the RANFIS model, the RMSE of the MQ135, MQ138, and PID-A15 sensors were 3.578, 11.594, and 4.837, respectively, which improved the performance by 87.1%, 25.9%, and 35.8% compared to the existing ANFIS. Therefore, the precision within 5% of the measurement results of the two experimentally verified sensors shows that the proposed RANFIS-based sensor system can be sufficiently applied in the field. Full article

Bacterial biofilms play an important role in the pathogenesis of infectious diseases but are also very relevant in other fields such as the food industry. This fact has led to an increased focus on the early identification of these structures as prophylaxes to prevent biofilm-related contaminations or infections. One of the objectives of the present study was to assess the effectiveness of NIR (Near Infrared) spectroscopy in the detection and differentiation of biofilms from different bacterial species, namely Staphylococcus epidermidis, Staphylococcus aureus, Enterococcus faecium, Salmonella Typhymurium, Escherichia coli, Listeria monocytogenes, and Lactiplantibacillus plantarum. Additionally, we aimed to examine the capability of this technology to specifically identify S. aureus biofilms on glass surfaces commonly used as storage containers and processing equipment. We developed a detailed methodology for data acquisition and processing that takes into consideration the biochemical composition of these biofilms. To improve the quality of the spectral data, SNV (Standard Normal Variate) and Savitzky–Golay filters were applied, which correct systematic variations and eliminate random noise, followed by an exploratory analysis that revealed significant spectral differences in the NIR range. Then, we performed principal component analysis (PCA) to reduce data dimensionality and, subsequently, a Random Forest discriminant statistical analysis was used to classify biofilms accurately and reliably. The samples were organized into two groups, a control set and a test set, for the purpose of performing a comparative analysis. Model validation yielded an accuracy of 80.00% in the first analysis (detection and differentiation of biofilm) and 93.75% in the second (identification of biofilm on glass surfaces), thus demonstrating the efficacy of the proposed method. These results demonstrate that this technique is effective and reliable, indicating great potential for its application in the field of biofilm detection. Full article

To achieve accurate control of Multi-Input and Multi-Output (MIMO) physical plants, it is crucial to obtain correct model expressions. In practice, the prevalence of both outliers and colored noise can cause serious interference with the industrial process, thus reducing the accuracy of the identification algorithm. The algorithm of support vector regression (SVR) is proposed to address the problem of parameter estimation for MIMO systems under interference from outliers and colored noise. In order to further improve the speed of parameter estimation, random search and Bayesian optimization algorithms were introduced, and the support vector regression combining stochastic search and Bayesian optimization (RSBO-SVR) algorithm was proposed. It was verified by simulation and tank experiments. The results showed that the method has strong anti-interference ability and can achieve high-precision parameter identification. The maximum relative error of the RSBO-SVR algorithm did not exceed 4% in both the simulation and experiment. It had a maximum reduction of 99.38% in runtime compared to SVR. Full article

This research presents a comprehensive approach for mitigating noise pollution from Unmanned Aerial Vehicles (UAVs) in urban environment by using Reinforcement Learning (RL) for flight path planning. Focusing on the city of Turin, Italy, the study utilizes its diverse urban architecture to develop a detailed 3D occupancy grid map, and a population density map. A dynamic noise source model adjusts noise emissions based on the UAV velocity, while acoustic ray tracing simulates noise propagation in the environment. The Deep Deterministic Policy Gradient (DDPG) algorithm optimizes flight paths, minimizing the noise impact, and balancing both the path length and the population density located under the UAV path. The simulation results demonstrate significant noise reduction, suggesting scalability and adaptability for global urban environments, contributing to sustainable urban air mobility by addressing noise pollution. Full article

This study investigates consonant devoicing in Brazilian Portuguese (BP), in order to assess whether an ongoing sound change is taking place. We examine plural forms consisting of a stop consonant followed by a word-final sibilant, such as in redes [hedz] ~ [heds] ~ [hets] and sedes [sɛdz] ~ [sɛds] ~ [sɛts], focusing on the emergence of voiceless sibilants before word-initial vowels (e.g., redes amarelas, ‘yellow hammocks’). If sibilants remain voiceless despite a following vowel, this challenges the expected regressive voicing assimilation in BP and raises the question of the conditions under which this devoicing occurs. Data were collected through recordings of oral production from twenty Brazilian speakers, using reading and picture naming tasks. Sibilant voicing was quantified using harmonics-to-noise ratio (HNR). A linear mixed-effects model—including random intercepts and slopes for both speakers and words—reveals that sibilants are significantly more voiced before a vowel than before a pause, but this voicing is substantially reduced when the sibilant is preceded by voiceless consonants. These findings indicate an ongoing devoicing process at pre-vocalic word boundaries in BP, affecting clusters [pz, tz, kz] and [bz, dz, gz] alike. Spectrographic analyses indicate that not only the sibilants but also their preceding stop may exhibit devoicing. Moreover, minimal-pair considerations suggest that speakers potentially maintain sibilant voicing in certain lexical items to preserve intelligibility (e.g., gra[dz] ‘grades’ and se[dz] ‘headquarters’ vs. grá[ts] ‘free’ and se[ts] ‘sets’). Drawing on Exemplar Theory, we propose a competition between the influence of the phonological environment and word-final devoicing: sibilants are sometimes voiced due to a following vowel (e.g., botes argentinos [bɔtz ah.ʒẽ.’tʃi.nus] ‘Argentine boats’), but they often emerge as voiceless due to consonantal devoicing (e.g., [bɔts ah.ʒẽ.’tʃi.nus]), resulting in both expected and unexpected forms. We suggest that fine phonetic detail, whether associated with allophonic or emergent sound patterns, contributes to the construction of phonological representations. Full article

Modern radar systems commonly utilize monopulse angle estimation techniques for target angle estimation, with the phase comparison method being one of the most widely adopted approaches. While the phase comparison method achieves high estimation precision, it is highly susceptible to noise and exhibits a suboptimal performance under low Signal-to-Noise Ratio (SNR) conditions, leading to a high SNR threshold. Moreover, conventional monopulse angle estimation methods provide limited target information, as a single measurement cannot reveal the target’s motion direction. To address these shortcomings, a novel approach based on the phase comparison method is proposed in this study, with the variation in the wave path difference modeled as a first-order motion model. By accumulating the conjugate-multiplied signals over multiple time steps, the Wave Path Difference Rate of Change Ambiguity Function (WPD-ROC AF) is constructed. A fast algorithm employing the 2D Chirp-Z Transform (2D-CZT) is proposed, enabling multi-pulse angle estimation through the identification of frequency and phase values corresponding to spectral peaks. Simulation results validate that the proposed method outperforms traditional monopulse angle estimation techniques under low-SNR conditions and effectively suppresses static clutter interference. Furthermore, the sign of the WPD-ROC AF is shown to be correlated with the target’s motion direction, providing practical utility for determining the direction of movement in remote sensing scenarios. Full article