Noise Interference Research Articles

AI-driven smart agriculture using hybrid transformer-CNN for real time disease detection in sustainable farming.

Plant diseases pose a significant threat to global food security, with severe implications for agricultural productivity. Early and accurate detection of these diseases is crucial, yet it remains a challenging task, significantly impacting crop yields and food supply chains. Despite the progress in artificial intelligence, particularly deep learning, challenges persist in real-world applications due to environmental noise, varying light conditions, and other complicating factors that hinder detection accuracy. This study introduces the AttCM-Alex model, a novel deep-learning framework designed to boost the detection and classification of plant diseases under challenging environmental conditions. By integrating convolutional operations with self-attention mechanisms, AttCM-Alex effectively addresses the variability in light intensity and image noise, ensuring robust performance. To simulate practical agricultural scenarios, the study employs bilinear interpolation for image dimension adjustment and introduces Salt-and-Pepper noise. Additionally, the model's robustness was evaluated by varying image brightness levels by ±10%, ±20%, and ±30%. Experimental results demonstrate that AttCM-Alex significantly outperforms traditional models, particularly in scenarios involving fluctuating light conditions and noise interference. The model achieved a peak detection accuracy of 0.97 with a 30% increase in image brightness and maintained an accuracy of 0.93 even with a 30% decrease in brightness, highlighting its robustness and reliability. The findings affirm the AttCM-Alex model as a powerful tool for real-world agricultural applications, capable of enhancing disease detection systems' accuracy and efficiency. This advancement not only supports better crop management practices but also contributes to sustainable agriculture and global food security.

Contrastive learning enhanced pseudo-labeling for unsupervised domain adaptation in person re-identification

Person re-identification (ReID) technology has many applications in intelligent surveillance and public safety. However, the domain difference between the source and target domains makes the generalization ability of the model extremely challenging. To reduce the dependence on labeled data, Unsupervised Domain Adaptation (UDA) methods have become an effective way to solve this problem. However, the influence of pseudo-label generated noise on model training in existing UDA methods is still significant, resulting in limited model performance on the target domain. For this reason, this paper proposes a contrast learning-based pseudo-label refinement with probabilistic uncertainty in the unsupervised domain, adapted to Person re-identification, aiming to improve the effectiveness of the unsupervised domain adapted to Person re-identification. We first enhance the feature representation of the target domain samples based on the contrast learning technique to improve their discrimination in the feature space, thereby enhancing the cross-domain migration performance of the model. Subsequently, an innovative loss function is proposed to effectively reduce the interference of label noise on the training process by refining the generation process of pseudo-labels, which solves the negative impact of inaccurate pseudo-labels on model training. Through a series of experimental validation, the method experiments on two large-scale public datasets, Market1501 and DukeMTMC, and the Rank-1 accuracy of the proposed method reaches 91.4% and 81.4%, with the mean average precision (mAP) of 79.0% and 67.9%, respectively, which proves that the research in this paper provides a good solution for the Person re-identification task with effective technical support for label noise processing and model generalization capability improvement.

A fault diagnosis model for inter-shaft bearing based on attention mechanism fusion and deep domain adaptation

Fault diagnosis of inter-shaft bearing is crucial for enhancing the reliability and safety of aero-engines. However, complex and variable working conditions, along with noise interference, make it challenging to ensure effective data collection and accurate fault diagnosis. To address the issue of inconsistent data distribution caused by varying working conditions, this paper proposed a deep transfer learning-based inter-shaft bearing fault diagnosis model called EPBSP. The model first used Continuous Wavelet Transform (CWT) to convert vibration signals from source and target conditions into time-frequency images. Then, a ResNet50-based feature extractor was constructed, incorporating a fusion attention mechanism residual block to effectively capture local and global information of fault features. Simultaneously, a joint loss function was constructed by combining Batch Spectral Penalization (BSP) regularization method with Domain-Adversarial Neural Network (DANN) for adversarial training, reducing the feature distribution difference between source and target domains and improving the model’s domain adaptation capability. Finally, the proposed method was analyzed using datasets from Harbin Institute of Technology and a self-built test rig, achieving fault diagnosis accuracies of 99.19% and 99%, respectively. The method outperformed existing fault diagnosis models in terms of both diagnostic accuracy and generalization ability.

Feature extraction and classification of hyperspectral images based on improved two-dimensional compact variational mode decomposition

In order to solve the problem of noise interference in hyperspectral image (HSI) feature extraction, an improved two-dimensional compact variational mode decomposition (2D-C-VMD) method is proposed to extract the spatial dimension feature information of HSI. By smoothing the low-frequency mode components obtained from the first decomposition of 2D-C-VMD, the method is called L-2D-C-VMD, which can further eliminate the adverse effects of spatial domain noise on the classification of different ground objects to a certain extent. Through simulation test analysis on the public datasets Indian Pine and Pavia University. When the Indian Pine training sample takes 4 %, the OA value of the proposed classification method can be increased by about 2 %. When the Pavia University training sample is taken at 0.5 %, the OA value of the proposed classification method can be increased by nearly 3 %, and the proposed method has better competitiveness with fewer training samples.

PAL: Boosting Skin Lesion Segmentation via Probabilistic Attribute Learning.

Skin lesion segmentation is vital for the early detection, diagnosis, and treatment of melanoma, yet it remains challenging due to significant variations in lesion attributes (e.g., color, size, shape), ambiguous boundaries, and noise interference. Recent advancements have focused on capturing contextual information and incorporating boundary priors to handle challenging lesions. However, there has been limited exploration on the explicit analysis of the inherent patterns of skin lesions, a crucial aspect of the knowledge-driven decision-making process used by clinical experts. In this work, we introduce a novel approach called Probabilistic Attribute Learning (PAL), which leverages knowledge of lesion patterns to achieve enhanced performance on challenging lesions. Recognizing that the lesion patterns exhibited in each image can be properly depicted by disentangled attributes, we begin by explicitly estimating the distributions of these attributes as distinct Gaussian distributions, with mean and variance indicating the most likely pattern of that attribute and its variation. Using Monte Carlo Sampling, we iteratively draw multiple samples from these distributions to capture various potential patterns for each attribute. These samples are then merged through an effective attribute fusion technique, resulting in diverse representations that comprehensively depict the lesion class. By performing pixel-class proximity matching between each pixel-wise representation and the diverse class-wise representations, we significantly enhance the model's robustness. Extensive experiments on two public skin lesion datasets and one unified polyp lesion dataset demonstrate the effectiveness and strong generalization ability of our method. Codes are available at https://github.com/IsYuchenYuan/PAL.

WiCG: Heartbeat Sensing Using COTS WiFi Devices with Common Antenna

Vital sign detection, based on Channel State Information (CSI) from commercial off-the-shelf (COTS) WiFi devices, has become a popular research area. Previous works in this field mainly focus on respiration, while heartbeat sensing has not been well studied yet, because its signal is very weak and overwhelmed by hardware noises and the respiration signal. Different from existing research that exploits directional antenna, the proposed WiCG ( Wi Fi C ardio G ram) system uses common antennas, and not only accurately senses heartbeat rate but also provides the heartbeat signal for further analysis in complex real-life home scenes. Specifically, we first propose an effective denoising solution for Wi-Fi CSI by exploiting its spatial structure, which exhibits strong correlation among the In-phase/Quadrature components. Leveraging this characteristic with Principal Component Analysis (PCA) achieves effective reduction of ambient noise in both the amplitude and phase of the CSI. Then, we introduce a heartbeat enhancement scheme that utilizes the periodicity of the heartbeat signal. By applying Singular Spectrum Analysis (SSA), the complex effects of residual noise and respiratory interference are effectively mitigated. Extensive experiments have proven that WiCG can effectively sense the heartbeat rate. In a real deployment environment, the average detection error can be reduced to 0.28 bpm, close to current commercial heartbeat sensors.

Design and Fabrication of Multi-Frequency and Low-Quality-Factor Capacitive Micromachined Ultrasonic Transducers

Capacitive micromachined ultrasonic transducers (CMUTs) have been developed for air-coupled applications to address key challenges such as noise, prolonged ringing, and side-lobe interference. This study introduces an optimized CMUT design that leverages the squeeze-film damping effect to achieve a low-quality factor, enhancing resolution and temporal precision for imaging as one of the suggested airborne application. The device was fabricated using the PolyMUMPs process, ensuring high structural accuracy and consistency. Finite element analysis (FEA) simulations validated the optimized parameters, demonstrating improved displacement, reduced side-lobe artifacts, and sharper main lobes for superior imaging performance. Experimental validation, including Laser Doppler Vibrometer (LDV) measurements of membrane displacement and mode shapes, along with ring oscillation tests to assess Q-factor and signal decay, confirmed the device’s reliability and consistency across four CMUT arrays. Additionally, this study explores the implementation of multi-frequency CMUT arrays, enhancing imaging versatility across different air-coupled applications. By integrating multiple frequency bands, the proposed CMUTs enable adaptable imaging focus, improving their suitability for diverse diagnostic scenarios. These advancements highlight the potential of the proposed design to deliver a superior performance for airborne applications, paving the way for its integration into advanced diagnostic systems.

Periodic-noise-tolerant neurodynamic approach for kWTA operation applied to opinions evolution.

Periodic-noise-tolerant neurodynamic approach for kWTA operation applied to opinions evolution.

Multifractal-Aware Convolutional Attention Synergistic Network for Carbon Market Price Forecasting

Accurate carbon market price prediction is crucial for promoting a low-carbon economy and sustainable engineering. Traditional models often face challenges in effectively capturing the multifractality inherent in carbon market prices. Inspired by the self-similarity and scale invariance inherent in fractal structures, this study proposes a novel multifractal-aware model, MF-Transformer-DEC, for carbon market price prediction. The multi-scale convolution (MSC) module employs multi-layer dilated convolutions constrained by shared convolution kernel weights to construct a scale-invariant convolutional network. By projecting and reconstructing time series data within a multi-scale fractal space, MSC enhances the model’s ability to adapt to complex nonlinear fluctuations while significantly suppressing noise interference. The fractal attention (FA) module calculates similarity matrices within a multi-scale feature space through multi-head attention, adaptively integrating multifractal market dynamics and implicit associations. The dynamic error correction (DEC) module models error commonality through variational autoencoder (VAE), and uncertainty-guided dynamic weighting achieves robust error correction. The proposed model achieved an average R2 of 0.9777 and 0.9942 for 7-step ahead predictions on the Shanghai and Guangdong carbon price datasets, respectively. This study pioneers the interdisciplinary integration of fractal theory and artificial intelligence methods for complex engineering analysis, enhancing the accuracy of carbon market price prediction. The proposed technical pathway of “multi-scale deconstruction and similarity mining” offers a valuable reference for AI-driven fractal modeling.

Research on adaptive filter of Orthogonal Frequency Division Multiplexing Signals Based on LMS Algorithm

This research focuses on processing signals in orthogonal frequency division multiplexing (OFDM) systems, especially adaptive filter techniques for multipath effects and noise interference. By deeply discussing the application of the LMS algorithm in the OFDM system, this paper proposes an adaptive filter scheme founded upon the LMS algorithm, which adaptively changes filter parameters to increase signal transmission efficiency and speed. The principles of the LMS algorithm and that of the adaptive filter, as well as the applicability and performance of the filter in the OFDM system, are analyzed and studied. Simulation experiments are conducted by MATLAB for the purpose of verifying the efficiency of the proposed scheme in reducing signal interference and improving system performance. The results indicate adaptive filter technology founded upon LMS algorithm has a wide application prospect in OFDM system, and plays a strong support for the development of wireless communication technology.

DFN-YOLO: Detecting Narrowband Signals in Broadband Spectrum

With the rapid development of wireless communication technologies and the increasing demand for efficient spectrum utilization, broadband spectrum sensing has become critical in both civilian and military fields. Detecting narrowband signals under broadband environments, especially under low-signal-to-noise-ratio (SNR) conditions, poses significant challenges due to the complexity of time–frequency features and noise interference. To this end, this study presents a signal detection model named deformable feature-enhanced network–You Only Look Once (DFN-YOLO), specifically designed for blind signal detection in broadband scenarios. The DFN-YOLO model incorporates a deformable channel feature fusion network (DCFFN), replacing the concatenate-to-fusion (C2f) module to enhance the extraction and integration of channel features. The deformable attention mechanism embedded in DCFFN adaptively focuses on critical signal regions, while the loss function is optimized to the focal scaled intersection over union (Focal_SIoU), improving detection accuracy under low-SNR conditions. To support this task, a signal detection dataset is constructed and utilized to evaluate the performance of DFN-YOLO. The experimental results for broadband time–frequency spectrograms demonstrate that DFN-YOLO achieves a mean average precision (mAP50–95) of 0.850, averaged over IoU thresholds ranging from 0.50 to 0.95 with a step of 0.05, significantly outperforming mainstream object detection models such as YOLOv8, which serves as the benchmark baseline in this study. Additionally, the model maintains an average time estimation error within s and provides preliminary center frequency estimation in the broadband spectrum. These findings underscore the strong potential of DFN-YOLO for blind signal detection in broadband environments, with significant implications for both civilian and military applications.

Generalized Adaptive Huber Loss Driven Robust Twin Support Vector Machine Learning Framework for Pattern Classification

Aiming at the problem that the traditional Twin Support Vector Machine (TSVM) is sensitive to noise and outliers, this paper proposes a twin support vector machine model based on generalized adaptive Huber loss function (GAHTSVM). Via dynamically adjusting the robust parameters s, the model can effectively suppress the adverse effects of noisy data points on the decision function, and improve the sparsity of the model by introducing insensitive region design. For non-convex optimization problems, concave-convex process (CCCP) is adopted to avoid quadratic programming problems and significantly improve the efficiency of the algorithm. Experiments show that GAHTSVM performs well in the scenarios where Gaussian noise is added to real-world datasets and outliers are introduced to artificial datasets: it is significantly better than other algorithms in low noise environment; In the high noise environment, it still maintains a leading position, and most datasets are ranked in the top; It performs well on two artificial datasets and is significantly superior to other algorithms.In conclusion, the model effectively overcomes the noise and outlier interference by dynamically adjusting the outlier penalty, and provides an efficient solution to the pattern recognition problem in high noise environment.

Motor Fault Diagnosis Under Strong Background Noise Based on Parameter-Optimized Feature Mode Decomposition and Spatial–Temporal Features Fusion

As the mining motor is used long-term in a complex multi-source noise environment composed of equipment group coordinated operations and high-frequency start–stop, its vibration signal has the features of significant strong noise interference, weak fault features, and the superposition of multiple working conditions coupling, which makes it arduous to efficiently extract and identify mechanical fault features. To address this issue, this study introduces a high-performance fault diagnosis approach for mining motors operating under strong background noise by integrating parameter-optimized feature mode decomposition (WOA-FMD) with the RepLKNet-BiGRU-Attention dual-channel model. According to the experimental results, the average accuracies of the proposed method were 97.7% and 93.38% for the noise-added CWRU bearing fault dataset and the actual operation dataset of the mine motor, respectively, which are significantly better than those of similar methods, showing that the approach in this study is superior in fault feature extraction and identification.

A Novel FMCW LiDAR Multi-Target Denoising Method Based on Optimized CEEMDAN with Singular Value Decomposition

Frequency-modulated continuous-wave (FMCW) LiDAR systems frequently experience noise interference during multi-target measurements in real-world applications, resulting in target overlapping and diminished detection accuracy. Conventional denoising approaches—such as Empirical Mode Decomposition (EMD) and wavelet thresholding—are often constrained by challenges like mode mixing and the attenuation of weak target signals, which limits their detection precision. To address these limitations, this study presents a novel denoising framework that integrates an optimized Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) algorithm and singular value decomposition (SVD). The CEEMDAN algorithm’s two critical parameters—the noise standard deviation and the number of noise additions—are optimally determined using particle swarm optimization (PSO), with the envelope entropy of the intrinsic mode functions (IMFs) serving as the fitness criterion. IMFs are subsequently selected based on spectral and amplitude comparisons with the original signal to facilitate initial signal reconstruction. Following CEEMDAN-based decomposition, SVD is employed with a normalized soft thresholding technique to further suppress residual noise. Validation using both synthetic and experimental datasets demonstrates the superior performance of the proposed approach over existing methods in multi-target scenarios. Specifically, it reduces the root mean square error (RMSE) by 45% to 59% and the mean square error (MSE) by 34% to 69%, and improves the signal-to-noise ratio (SNR) by 1.85–4.38 dB and the peak signal-to-noise ratio (PSNR) by 1.18–6.94 dB. These results affirm the method’s effectiveness in enhancing signal quality and target distinction in noisy FMCW LiDAR measurements.

Semantic ECG hash similarity graph

Graph-based methods have made significant progress in addressing the dependent correlations among ECG time series variables. However, most existing graph structures primarily focus on local similarity while overlooking global semantic correlation. Additionally, the adjacency matrix is highly susceptible to noise interference, leading to unreliable node connections. In this paper, we present a novel graph generation learning framework that incorporates semantic hash coding to capture the intricate associations both within and between ECG signals, thereby significantly enhancing the retrieval efficiency of subsequent graph-based deep learning models. Specifically, the Semantic Hash Similarity Graph (SHSG) initially leverages the similarity within the label space to generate a hash representation for the supervised signal. Subsequently, a lightweight linear hash function is utilized to produce a hash representation for the unseen signal. Thereafter, a comprehensive global hash dictionary is systematically constructed. Finally, the graph topology is meticulously assembled by leveraging Hamming similarity. Additionally, to ensure the maintenance of semantic similarity, we propose an iterative optimization approach in the orthogonal domain for generating hash representations. To validate the efficacy of the generated graph, we utilized a fast Graph Convolutional Network (GCN) for ECG recognition. The experimental outcomes on multiple publicly available ECG datasets corroborate the robustness and effectiveness of our proposed method.

Longitudinal Tire Force Estimation Method for 4WIDEV Based on Data-Driven Modified Recursive Subspace Identification Algorithm

For the longitudinal tire force estimation problem of four-wheel independent drive electric vehicles (4WIDEVs), traditional model-based observers have limitations such as high modeling complexity and strong parameter sensitivity, while pure data-driven methods are susceptible to noise interference and have insufficient generalization ability. Therefore, this study proposes a joint estimation framework that integrates data-driven and modified recursive subspace identification algorithms. Firstly, based on the electromechanical coupling mechanism, an electric drive wheel dynamics model (EDWM) is constructed, and multidimensional driving data is collected through a chassis dynamometer experimental platform. Secondly, an improved proportional integral observer (PIO) is designed to decouple the longitudinal force from the system input into a state variable, and a subspace identification recursive algorithm based on correction term with forgetting factor (CFF-SIR) is introduced to suppress the residual influence of historical data and enhance the ability to track time-varying parameters. The simulation and experimental results show that under complex working conditions without noise and interference, with noise influence (5% white noise), and with interference (5% irregular signal), the mean and mean square error of longitudinal force estimation under the CFF-SIR algorithm are significantly reduced compared to the correction-based subspace identification recursive (C-SIR) algorithm, and the comprehensive estimation accuracy is improved by 8.37%. It can provide a high-precision and highly adaptive longitudinal force estimation solution for vehicle dynamics control and intelligent driving systems.

β-Ga2O3/BP Heterojunction for Deep Ultraviolet and Infrared Narrowband Dual-Band Photodetection

Abstract The development of high-performance dual-band photodetectors (PDs) capable of simultaneous deep ultraviolet (DUV) and infrared (IR) detection is critical for advanced optoelectronic applications, particularly in missile warning and target identification systems. Conventional UV/IR PDs often suffer from Ultraviolet (320–400 nm) noise interference and limited responsivity due to narrow-bandgap semiconductors and self-powered operation modes. To overcome these challenges, high-quality β-Ga2O3 thin films were epitaxially grown on c-plane sapphire via metalorganic chemical vapor deposition (MOCVD), exhibiting excellent crystallinity and surface morphology. Unlike conventional heterojunctions (β-Ga2O3/graphene or β-Ga2O3/TMDs), the β-Ga2O3/BP structure leverages BP’s tunable bandgap and high carrier mobility while maintaining strong type-II band alignment, facilitating efficient charge separation under both UV and IR illumination. We present a high-sensitivity dual-band PD based on a β-Ga2O3/black phosphorus (BP) pn heterojunction. The ultrawide bandgap of β-Ga2O3 ensures selective deep ultraviolet DUV detection, effectively suppressing UVA noise, while BP provides a layer-dependent IR response. Photocurrent analysis reveals distinct carrier transport mechanisms, with electrons dominating under UV illumination and holes contributing predominantly under IR exposure. Systematic investigation of bias-dependent photoresponse demonstrates enhanced responsivity at higher voltages. Under 7 V bias, the device exhibits a high responsivity of 4.63 × 10-2 mA/W at 254 nm and 2.35 × 10-3 mA/W at 850 nm. This work not only provides a viable strategy for developing high-performance dual-band PDs but also advances the understanding of heterojunction-based optoelectronic devices for military and sensing applications.

Nodal Carbon Emission Factor Prediction for Power Systems Based on MDBO-CNN-LSTM

Carbon emission estimation for power systems is essential for identifying emission responsibilities and formulating effective mitigation measures. Current carbon emission prediction methods for power systems exhibit limited computational efficiency and inadequate noise immunity under complex operating conditions. In this study, we address these limitations by improving population initialization, search mechanisms, and iteration strategies and developing a hybrid strategy Modified Dung Beetle Optimization (MDBO) algorithm. This led to the development of an MDBO-enhanced Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) network hybrid prediction model for carbon emission prediction. Firstly, the theoretical calculation mechanism of carbon emission flow in power systems is analyzed. Subsequently, an MDBO-CNN-LSTM deep network architecture is constructed, with detailed explanations of its fundamental structure and operational principles. Then, the proposed MDBO-CNN-LSTM model is utilized to predict the nodal carbon emission factor of power systems with the integration of renewable energy sources. Comparative experiments with conventional CNN-LSTM models are conducted on modified IEEE 30-, 118-, and 300-bus test systems. The results show that the maximum mean squared error of the proposed method does not exceed 0.5734% in the strong-noise scenario for the 300-bus system, which is reduced by half compared with the traditional method. The proposed method exhibits enhanced robustness under strong noise interference, providing a novel technical approach for precise carbon accounting in power systems.

ScRDAN: a robust domain adaptation network for cell type annotation across single-cell RNA sequencing data.

Single-cell RNA sequencing technology facilitates the recognition of diverse cell types and subgroups, playing a crucial role in investigating cellular heterogeneity. Cell type annotation, a crucial process in single-cell RNA sequencing analysis, is often influenced by noise and batch effects. To address these challenges, we propose scRDAN, which is a robust domain adaptation network comprising three modules: the denoising domain adaptation module, the fine-grained discrimination module, and the robustness enhancement module. The denoising domain adaptation module mitigates noise interference through feature reconstruction in domains, while leveraging adversarial learning to align data distributions, improving annotation accuracy and robustness against batch effects. The fine-grained discrimination module maintains intra-class compactness and enhances inter-class separability, reducing feature overlap and improving cell type distinction. Finally, the robustness enhancement module introduces noise from various perspectives in both domains, enhancing robustness and generalization. We evaluate scRDAN on simulated, cross-platforms, and cross-species datasets, comparing it with advanced methods. Results demonstrate that scRDAN outperforms existing methods in handling batch effects and cell type annotation.

DAM-Faster RCNN: few-shot defect detection method for wood based on dual attention mechanism

In wood defect detection, factors such as few-shot sample scarcity, diverse defect types, and complex background interference severely limit the model’s recognition accuracy and generalization ability. To address the above issues, this paper proposes an improved Faster RCNN model based on a dual attention mechanism (DAM). The model integrates cross-attention and spatial attention modules to enhance the expression of key region features, suppresses texture noise interference; the improved Wood-Region Proposal Network (WRPs) module utilizes feature mean pooling and cross-layer fusion strategies to significantly improve the quality and robustness of candidate box generation; in addition, the Wood-Feature Reconstruction Head (WFRH) module effectively enhances the adaptability to new classes and few-shot defects through multi-branch classification and weighted fusion mechanisms. After synergistic optimization of all modules, the model demonstrates superior detection accuracy and category discrimination capability. Experimental results show that the proposed method achieves state-of-the-art performance on the PASCAL VOC and FSOD datasets, particularly in the identification of 17 types of wood defects, where AP50 and AP75 are improved by 25% and 7.9%, respectively, validating the significant advantages of the proposed DAM mechanism under few-shot and complex background conditions. The findings of this study provide practical technical references for intelligent and efficient few-shot detection in real-world wood quality inspection tasks.