<?xml version="1.0" encoding="utf-8"?>
    <rss version="2.0">
      <channel xmlns:content="http://purl.org/rss/1.0/modules/content/">
        <title>Frontiers in Signal Processing | New and Recent Articles</title>
        <link>https://www.frontiersin.org/journals/signal-processing</link>
        <description>RSS Feed for Frontiers in Signal Processing | New and Recent Articles</description>
        <language>en-us</language>
        <generator>Frontiers Feed Generator,version:1</generator>
        <pubDate>2026-09-16T16:01:47.763+00:00</pubDate>
        <ttl>60</ttl>
        <item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1904505</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1904505</link>
        <title><![CDATA[PDSA-DBUNet: dual-branch U-Net for abdominal medical image segmentation driven by polynomial dynamic sparse attention]]></title>
        <pubdate>2026-09-16T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Jiatong Shang</author><author>Guolin He</author><author>Xuyang Li</author>
        <description><![CDATA[Abdominal medical image segmentation is essential for computer-aided diagnosis, lesion localization, quantitative organ analysis, and treatment planning. However, abdominal computed tomography (CT) images often contain large anatomical variations, complex tissue structures, low contrast, blurred boundaries, and small tumor regions, which make accurate and efficient segmentation challenging. Convolutional neural networks (CNNs) are effective in extracting local textures, edges, and spatial details, but their limited receptive fields restrict long-range dependency modeling. Transformers can capture global contextual relationships, yet dense token modeling introduces redundant background information and high computational cost, while fine-grained local details may be insufficiently preserved. To address these limitations, we propose a Polynomial Dynamic Sparse Attention-driven Dual-Branch U-Net, termed PDSA-DBUNet, for abdominal medical image segmentation. The network adopts a parallel CNN–Transformer architecture, where the CNN branch preserves local boundary and texture information, and the Transformer branch models global semantic dependencies. A dynamic sparse visual token generation module adaptively aggregates dense image tokens into a compact set of discriminative sparse tokens, emphasizing organ regions and blurred boundaries while suppressing irrelevant background responses. A PoM-based polynomial mixing mechanism is further incorporated into the dynamically generated sparse token set to replace conventional multi-head self-attention, enabling efficient global contextual interaction with reduced computational and memory costs. In addition, a Dual-Scale Parallel Attention Fusion (DSPAF) module integrates CNN-derived local details with Transformer-derived global semantics to enhance boundary localization and small-target representation. Experiments on the LiTS and KiTS19 datasets show that PDSA-DBUNet achieves Dice scores of 96.79% and 86.13% on LiTS-Liver and LiTS-Tumors, and 94.03% and 89.06% on Kidney and Kidney-Tumor segmentation tasks, respectively. The model requires 72.87 GFLOPs, contains 41.62M learnable parameters, and has a checkpoint storage size of 129.93 MB. These results demonstrate that PDSA-DBUNet achieves a favorable balance between segmentation accuracy and computational efficiency.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1864603</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1864603</link>
        <title><![CDATA[Computationally efficient hybrid multi-domain transform fusion for secure image steganography using rate–distortion optimization and Toom–Cook acceleration]]></title>
        <pubdate>2026-09-14T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Abhrendu Bhattacharya</author><author>Raktim Kumar Dey</author><author>Amit R. Welekar</author><author>Rajesh Bose</author><author>Sandip Roy</author><author>Arfat Ahmad Khan</author><author>Muhammad Shoaib Ayub</author><author>Insoo Koo</author>
        <description><![CDATA[In the era of extensive developmental growth, multimedia communication systems demand protective and indiscernible information-hiding techniques that can maintain a balance between robustness, capacity, and computational efficacy. The research gap analysis highlights the necessity for a mathematically defined steganographic strategy that integrates rate–distortion optimization, knowledge-based adaptive embedding, hybrid domain modeling, and statistical invisibility constraints in a non-heuristic approach toward scalable high-capacity transmission systems. The novelty of the proposed strategy lies in the combination of rate–distortion optimization, adaptive CNN-based control for embedding, and histogram preservation of frequency channel embedding within a consolidated information-hiding architecture, within illuminated components of color areas. This ensures observable fidelity and configurational stability. To generate an adaptive embedding-based probability map, a lightweight convolutional neural network that enables a content-aware embedding dependent on local picture complexity is incorporated. Moreover, incorporating a histogram-preservation restriction reduces sensitivity to classical and learning-guided steganalysis and increases analytical security. Furthermore, Toom–Cook accelerated convolution is utilized to achieve improved computational scalability through accelerated transform-domain convolution, ensuring scalability for high-resolution images. Experimental analysis illustrates excellent performance with superior imperceptibility in average values obtained for a peak signal-to-noise ratio (PSNR) maintained above 50 dB (≈51.40 dB) and average structural similarity index measure (SSIM) ≈0.9816, low bit error rates, and near-random detectability (area under curve ≈0.5). Comparative and ablation analyses confirm that the proposed framework outperforms existing spatial, transform-domain, and deep learning-based methods. The proposed approach provides a unified, theoretically constrained, and scalable solution for secure high-capacity image steganography.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1842851</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1842851</link>
        <title><![CDATA[AdaTent++: reliability-guided hybrid test-time adaptation for traffic sign detection under distribution shift]]></title>
        <pubdate>2026-09-09T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Xiaomei Gai</author><author>Meichun Wang</author><author>Yantong Guo</author><author>Shiyu Wang</author><author>Ye Li</author><author>Rui Fu</author>
        <description><![CDATA[Test-time adaptation (TTA) can improve detector robustness under distribution shift, but indiscriminate online updating may amplify errors from unreliable frames. We present AdaTent++, a reliability-guided hybrid TTA framework for traffic sign detection. For each unlabeled frame, the framework combines predictive entropy and cross-view agreement to select among a smoothed batched normalization (BN)-statistics update, a guarded BN-affine gradient update, and no adaptation. Exponential moving average (EMA) threshold tracking, prototype regularization, gradient control, and state restoration are used as supporting safeguards rather than claimed as independent novelties. On one public traffic sign dataset evaluated under 12 corrupted domains, AdaTent++ achieves a mean average precision (mAP50) of 68.87%, compared with 68.03% for the strongest baseline, T3A, corresponding to an improvement of 0.84 percentage points. It ranks first in three of the twelve domains and is less effective under several structure-destructive corruptions. The results indicate a favorable average trade-off in the studied setting, while cross-dataset, cross-architecture, embedded-device, and long-horizon validation remain open.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1911510</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1911510</link>
        <title><![CDATA[TF-Mamba-DPHNet: a time-frequency state-space and dual-path interaction framework with hybrid cross-scale feature calibration for speech enhancement]]></title>
        <pubdate>2026-09-01T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Shaik AreefaBegam</author><author>Sunnydayal Vanambathina</author>
        <description><![CDATA[IntroductionSpeech enhancement plays a fundamental role in improving the perceptual quality and intelligibility of speech signals degraded by noise. Conventional U-Net architectures effectively capture local spectral information but struggle to model long-range dependencies and may allow residual noise to propagate through skip connections. Transformer-based networks can model global contextual dependencies and generate high-quality enhanced speech; however, their limited capacity to preserve fine-grained high-frequency spectral details can limit their suitability for real-time applications.MethodsTo address these limitations, this paper proposes TF-Mamba-DPHNet, a novel encoder-decoder speech enhancement framework integrating Multi-Scale Feature Extraction (MSFE), Time-Frequency Mamba (TF-Mamba), Dual-Path Higher-Order Information Interaction with Time-Frequency Attention (DPH-TFA), and Hybrid Cross-Scale Feature Calibration (H-CS-FC) modules. The MSFE blocks in the encoder and decoder extract local patterns across multiple receptive fields, capturing fine-grained and global time-frequency cues. The TF-Mamba module models global time-frequency dependencies using selective state-space modeling for effective long-term sequence understanding. At the bottleneck, four stacked DPH-TFA blocks capture long-range dependencies in both the temporal and spectral domains. These global features are fused with hierarchical encoder outputs through H-CS-FC modules, which perform cross-scale-guided feature recalibration to suppress noise leakage in the skip pathways and improve decoder reliability.ResultsEvaluations conducted on the Common Voice and LibriSpeech datasets demonstrate that TF-Mamba-DPHNet effectively improves speech quality and robustness compared with earlier models.DiscussionThe results indicate that combining multi-scale local feature extraction, selective state-space modeling, dual-path time-frequency interaction, and cross-scale feature calibration enables effective modeling of complementary local and global information for robust speech enhancement.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1909928</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1909928</link>
        <title><![CDATA[Modeling the post-extrasystolic potentiation to determine heart failure]]></title>
        <pubdate>2026-08-28T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Roel J. H. Montree</author><author>Elisabetta Peri</author><author>Lukas R. C. Dekker</author><author>Xi Long</author><author>Reinder Haakma</author><author>Rik Vullings</author>
        <description><![CDATA[IntroductionPremature ventricular contractions (PVC) occur naturally in the human heart and lead to a well-studied increase in cardiac contractility, referred to as post-extrasystolic potentiation (PESP), which behaves independently of many otherwise confounding factors. It is well-known that the degree of potentiation differs between healthy and failing hearts. In this paper, we present a method for non-invasive and non-obtrusive assessment of the PESP, leveraging photoplethysmography (PPG) measurements, revealing information on the degree of potentiation and thereby cardiac functioning.MethodsWe modeled the PESP as a second-order system and we compared the parameters of this system between patients with differing ejection fractions. The method was developed on an internal dataset acquired during percutaneous, electrophysiological procedures. The parameters of the proposed second-order system are determined for each of these patients and the results show a correlation between cardiac ejection fraction and the damping factor, one of the parameters of the system.ResultsThis model-based approach was demonstrated on the earlier mentioned non-obtrusive PPG measurements, as well as invasive blood pressure measurements. Modeling the PESP as a second-order model using information obtained from the non-invasive PPG signal reveals information showing a promising step towards enabling continuous monitoring of cardiac contractility using non-invasive means.DiscussionThe presented method could be applied in non-invasive, non-obtrusive PPG recording devices such as a smartwatch. This provides a continuously monitored estimate of cardiac contractility as an additional clinical parameter, as well as decreasing the time between reduction of contractility and potential hospital admission.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1884114</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1884114</link>
        <title><![CDATA[Seismic phase picking of coal mine microseismic data based on lightweight CNN]]></title>
        <pubdate>2026-08-21T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Haichen Yin</author><author>Xiufeng Zhang</author><author>Atao Li</author><author>Wei Zhu</author><author>Yuhua Hou</author><author>Yang Chen</author><author>Caiping Lu</author>
        <description><![CDATA[IntroductionIn coal mining, the high-noise underground environment poses significant challenges to the accuracy and stability of microseismic (MS) signal first-arrival picking.MethodsThis study proposes a high-robustness automatic phase picking method based on a lightweight Convolutional Neural Network (CNN). The method uses a multi-layer convolution and adaptive pooling structure to efficiently extract waveform features across temporal scales, and introduces a regression prediction strategy to directly output the first-arrival time, eliminating the quantization error induced by classification-based methods. The model was validated on 1,791 manually labelled single-component records of 597 microseismic events from the 6305 working face of Xinjulong Coal Mine, Shandong Province.ResultsOn the test set, the proportion of first-arrival errors not exceeding 100 ms reached 80.27%. Compared with the traditional STA/LTA method, the overall relative error was significantly reduced, with the average relative error decreasing by approximately 50 ms. The model maintained high picking accuracy and stability under high-noise and complex geological conditions, and under high-SNR conditions it effectively reduced the subjective bias of manual picking.DiscussionThe findings provide reliable technical support for high-precision localization of microseismic events and the early warning of dynamic hazards in coal mines.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1876242</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1876242</link>
        <title><![CDATA[Physiological and subjective evaluation of music processing strategies for hearing-impaired listeners: a pilot study]]></title>
        <pubdate>2026-08-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Duncan A. H. Williams</author><author>Rory Waterhouse-Stocks</author><author>Rebecca Vos</author><author>Trevor Cox</author>
        <description><![CDATA[IntroductionHearing loss affects over 2.5 billion people worldwide and significantly degrades music enjoyment. While advances in digital signal processing (DSP) aim to improve listening experiences, their combined effects on perceived audio quality and physiological engagement remain underexplored.MethodsThis pilot study evaluated hearing-impairment-optimised DSP models from the CADENZA challenge. Eleven participants (eight normal-hearing, three hearing-aid users) listened to musical excerpts processed via four DSP algorithms using circumaural open-back headphones in a semi-anechoic chamber. Signal delivery was tailored using personalized audiogram baselines and volume attenuation. Subjective metrics, Basic Audio Quality (BAQ) and likeability, were collected alongside continuous physiological recordings (electrodermal activity [EDA] and photoplethysmography [PPG]) captured with a medical-grade wrist wearable.ResultsDSP model E021 consistently achieved the highest BAQ and likeability scores across both participant groups, with BAQ and likeability demonstrating a strong positive correlation. Physiological monitoring revealed exploratory associations between autonomic arousal and variability metrics: specifically, phasic EDA variability and inter-beat interval (IBI) variability positively correlated with both BAQ and likeability.DiscussionVariability-based physiological markers present viable pathways for evaluating subjective listener responses in small cohorts. Integrating wearable-derived biometric signals with subjective appraisal provides a framework for developing adaptive, listener-aware DSP strategies that enhance music accessibility for individuals with hearing impairment.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1893420</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1893420</link>
        <title><![CDATA[An explainable AI framework integrating deep learning and large language model for student’s mental health]]></title>
        <pubdate>2026-08-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Rupali D. Kasar</author><author>Garima Shukla</author>
        <description><![CDATA[Mental health disorders such as anxiety, depression, and Mild Cognitive Impairment (MCI) are increasingly prevalent among young adults aged 18–30, significantly affecting academic performance, cognitive functioning, and overall wellbeing. Traditional diagnostic approaches depend on subjective assessments and limited clinical observations, making early and accurate detection challenging. To address these limitations, this research proposes an interpretable deep learning-based multimodal method for comprehensive mental health prediction and personalized intervention. The framework integrates heterogeneous data sources, including demographic, cognitive, behavioral, physiological, and neurocognitive indicators collected from clinical settings. Data preprocessing includes imputation, normalization, encoding, and text transformation. A Cross-Directional Feature Learning Network (CDFLN) is employed for robust multimodal feature extraction, followed by a Multi-model Progressive Dense Self-Attention for Cross Domain (MPDSA-CD) architecture for classification of anxiety, depression, and MCI, along with cognitive risk and severity assessment. Model performance is further enhanced by the Starfish Optimization Algorithm for hyperparameter tuning and parameter refinement. To ensure clinical transparency, SHapley Additive exPlanations (SHAP) are utilized to interpret model predictions and identify key risk factors influencing mental health outcomes. The proposed method achieves an accuracy of 99.8%, precision of 99.7%, recall of 99.9%, and F1-score of 99.8%, demonstrating strong robustness, generalization ability, and clinical applicability for early detection and effective psychological intervention in young adults.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1769553</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1769553</link>
        <title><![CDATA[Identification of eye diseases with small fundus image datasets using a new unsupervised deep learning approach]]></title>
        <pubdate>2026-08-03T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Al Rafi Aurnob</author><author>Sharia Arfin Tanim</author><author>Farook Sattar</author>
        <description><![CDATA[Ocular diseases are a severe global problem, particularly in countries that lack the technology or economy to treat them. Automated analysis of retinal fundus images could enable early screening, but supervised deep learning methods require large, annotated datasets that are expensive and time-consuming to obtain. This study introduces an unsupervised deep learning framework for learning and clustering of retinal image representation that achieves strong performance on small-scale fundus image datasets using only image-level disease labels to guide contrastive pair construction and classification loss, without requiring pixel-level annotations for clustering. The proposed method, NF-HAE-UDL (normalizing flow-guided hybrid autoencoder-based unsupervised deep learning), combines a hybrid autoencoder architecture (ResNet-50 and Swin Transformer) with normalizing flow and contrastive learning to learn discriminative latent representations. The autoencoder generates meaningful feature representations, upon which the normalizing flow maps the encoded features to a more expressive and structured latent space. Meanwhile, contrastive learning encourages the separation of patterns targeted to specific diseases through data augmentation. Evaluation on the STARE dataset (81 images, multi-label) and the IDRiD subset (211 images, binary) demonstrates superior unsupervised clustering performance compared to traditional methods, achieving ARI scores of 0.58±0.10 on STARE and 0.58±0.06 on IDRiD. Auxiliary classification validation on frozen encoder representations confirms that learned embeddings preserve disease discriminative information (88% in-sample accuracy on STARE, 98% on IDRiD), though these results represent quality assessment rather than diagnostic performance. Furthermore, Grad-CAM visualizations validated against expert annotations (IoU =0.52±0.11) highlight clinically meaningful attention to pathological features. The framework demonstrates a promising direction for effective, interpretable, unsupervised retinal image analysis with limited data.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1862897</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1862897</link>
        <title><![CDATA[Hybrid CUDA acceleration of wavelet-based MRI image denoising]]></title>
        <pubdate>2026-07-09T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Lesia Hentosh</author><author>Yana Vyklyuk</author><author>Oleh Berezsky</author>
        <description><![CDATA[BackgroundMagnetic resonance imaging is a key modality in medical diagnostics, but its quality is often degraded by noise, which reduces image clarity and may affect subsequent analysis. Wavelet-based denoising is effective for preserving anatomical structures, but its practical use is limited by computational cost. This study aimed to develop a hybrid graphics processing unit execution method for wavelet-based denoising of two-dimensional magnetic resonance images with high throughput and stable reconstruction quality.MethodsA hybrid CUDA-based implementation of wavelet denoising was developed using a standard discrete wavelet transform with adaptive soft-thresholding as the denoising core. The proposed execution method combines data-level and task-level parallelism within a unified graphics processing unit pipeline and includes dual-stream vertical-pass scheduling, fused reconstruction kernels, constant-memory filter taps, symmetric boundary handling, and selective loop unrolling. Experiments were performed on 3,064 grayscale brain magnetic resonance images of size ×512 512 using synthetic Gaussian noise and simulated Rician noise and motion artifacts. An additional sensitivity analysis was conducted to assess the influence of wavelet basis, filter length, and threshold scaling on PSNR, SSIM, mean squared error, and runtime. Reconstruction quality was evaluated using peak signal-to-noise ratio, structural similarity index, and mean squared error, while computational performance was assessed using runtime and throughput. Results were summarized over repeated runs using mean values, standard deviations, and 95% confidence intervals.ResultsUnder Gaussian noise, the proposed method achieved peak signal-to-noise ratio values of approximately 29–30 dB and structural similarity values around 0.90 while maintaining a throughput of 215–220 images per second. Compared with a reproduced data-parallel graphics processing unit wavelet baseline, the proposed implementation reduced runtime by 20%–30%. For Rician noise and motion-related artifacts, the method maintained stable slice-level throughput and consistent reconstruction behavior across the evaluated degradation types.ConclusionThe proposed hybrid CUDA execution method provides an effective high-throughput solution for wavelet-based denoising of two-dimensional magnetic resonance images. By combining dual-stream scheduling, fused reconstruction, and memory-aware execution, the method improves runtime performance while preserving reconstruction quality.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1829279</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1829279</link>
        <title><![CDATA[Extended Kalman filtering on Stiefel manifolds]]></title>
        <pubdate>2026-06-30T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Jordi-Lluís Figueras</author><author>Aron Persson</author><author>Lauri Viitasaari</author>
        <description><![CDATA[A generalization of the extended Kalman filter for Stiefel manifold-valued measurements is presented. Our approach is based on translating between manifold-valued measurements and the ambient space through an injective mapping between scalar true variance on the manifold and the variance of the noise. This mapping inherently takes into account the curvature of the manifold corresponding to the higher order derivatives. To illustrate our approach, we provide simulations on the 2-sphere and the space of orthogonal 4-by-2 matrices which show significant improvement of the Extended Kalman Filter compared to only relying on raw measurements. We also compare our approach to classical Extended Kalman Filter and second order Extended Kalman Filter, and complement our results by physically informed application related to spacecraft measurements. Our results indicate that our algorithm has similar performance as the second order Extended Kalman Filter without requiring the costly computations of the second order derivatives.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1853106</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1853106</link>
        <title><![CDATA[An explainable AI-driven hybrid SE-transformer architecture for robust knee osteoporosis classification from X-ray data]]></title>
        <pubdate>2026-06-26T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Sudhir Kumar Sangula</author><author>Siddique Ibrahim S. P.</author>
        <description><![CDATA[BackgroundEarly and accurate classification of knee osteoporosis from radiographic images is crucial for timely diagnosis and treatment, yet existing deep learning models often lack both contextual understanding and interpretability.ObjectivesTo develop an explainable hybrid deep learning framework that combines Squeeze-and-Excitation (SE) networks, Transformer architecture, BiLSTM, and BiGRU for improved knee osteoporosis classification.MethodsThe proposed model integrates SE blocks for channel-wise feature enhancement, a Transformer for capturing long-range spatial dependencies, and BiLSTM/BiGRU layers for sequential feature learning. Grad-CAM is employed to provide visual explanations of model predictions. Performance is evaluated using a stratified 70:20:10 training, validation, and testing split.ResultsThe framework achieved an accuracy of 91.0%, precision of 91.0%, recall of 90.2%, and an F1-score of 90.4% on the test set. The reported results are based on a single-dataset evaluation and comparisons under identical experimental conditions.ConclusionThe proposed SE-Transformer–BiLSTM–BiGRU framework delivers accurate and interpretable knee osteoporosis classification, demonstrating its potential for computer-aided diagnosis while warranting further validation on external datasets.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1802483</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1802483</link>
        <title><![CDATA[Energy-aware bitrate ladder for VVC streaming with client-side super-resolution]]></title>
        <pubdate>2026-06-24T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Amritha Premkumar</author><author>Vignesh V. Menon</author><author>Christian Herglotz</author>
        <description><![CDATA[Modern adaptive video streaming must jointly optimize perceptual quality, bitrate efficiency, latency, and energy consumption across heterogeneous client devices. However, existing bitrate-ladder design approaches, whether fixed, per-title, or machine-learned, typically assume that decoded video is directly consumed, overlooking the increasing use of client-side Video Super-Resolution (VSR) and its associated computational and energy costs. Moreover, while the Versatile Video Coding (VVC) standard achieves substantial compression gains, its increased decoding complexity further complicates energy-efficient deployments. To address these challenges, we introduce E2-Stream, a fully measurement-driven framework that integrates VVC encoding, multi-model VSR enhancement, and client-side decoding and enhancement energy profiling for enhancement-aware bitrate-ladder construction. Instead of relying on predictions or heuristics, E2-Stream exhaustively evaluates each resolution–bitrate–VSR configuration on real hardware to capture the true rate–quality–energy behavior, and selects representations that balance perceptual quality and client-side energy consumption. A Just Noticeable Difference (JND)-aware pruning strategy further removes perceptually redundant representations, yielding compact ladders with reduced storage overhead. Unlike prior work focusing solely on spatial super-resolution, we evaluate both lightweight frame-wise SISR models (FSRCNN, ESPCN) and a temporally-aware VSR model (EDVR), enabling direct comparison of quality gains, temporal consistency, and inference cost. We further evaluate E2-Stream on two heterogeneous client devices: (i) a CPU-only system with software VVC decoding and CPU-based inference, and (ii) a GPU-enabled system with hardware-accelerated decoding and GPU-accelerated VSR. Experiments on the Inter-4K dataset show that enhancement-aware ladders preserve sub-JND perceptual quality while achieving 12%–28% bitrate savings. On the CPU-only device, decoding and VSR inference increase energy consumption by up to 85%, highlighting the cost of software-based enhancement. In contrast, on the GPU-enabled device, the same configurations reduce total client-side energy by 25%–60% due to efficient hardware acceleration, while maintaining strong bitrate reductions (up to 40%). Additionally, JND-based pruning reduces storage requirements by up to 58.98% without compromising perceptual quality. These results demonstrate that VSR fundamentally reshapes the rate–quality–energy trade-off and that optimal bitrate ladders are inherently device-dependent. E2-Stream provides a principled and reproducible framework for energy- and enhancement-aware streaming in the VVC era.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1844194</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1844194</link>
        <title><![CDATA[Classification of infrasonic signals based on time-frequency representation]]></title>
        <pubdate>2026-06-23T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Hao Yin</author><author>Yu Lu</author><author>Yunfen Chang</author><author>Kai Zhang</author><author>Yunhui Wu</author><author>Fan Yang</author><author>Xuexu Li</author><author>Xinliang Pang</author><author>Peng Li</author>
        <description><![CDATA[Accurate classification of infrasound signals is an important research topic in the fields of nuclear explosion monitoring and natural disaster early warning. Current studies often rely on empirical selection of time-frequency analysis methods to characterize denoised signals, lacking a systematic evaluation of different methods under noisy conditions within a unified framework. To address this issue, this study takes six types of non-denoised infrasound signals—including nuclear tests, chemical explosions, and lightning—as research objects, and systematically compares the comprehensive performance of Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Hilbert–Huang Transform (HHT) in a convolutional neural network (CNN) classification model. Experimental results show that STFT achieves the best overall performance under the configuration of a Hanning window with a length of 1 s and NFFT = 1,024, with a classification accuracy of 98.35%, significantly outperforming CWT and HHT. Meanwhile, STFT requires the shortest computation time (0.85 s) and exhibits the fastest model convergence. Further analysis reveals that the inherent smoothing and consistency provided by the fixed window function of STFT contribute to stronger robustness and feature stability in noisy environments. This study simplifies the classification pipeline, avoids information loss and computational overhead caused by denoising preprocessing, emphasizes the importance of selecting time-frequency methods under noisy conditions, and provides a more efficient and reliable engineering solution for real-time infrasound monitoring systems, demonstrating strong practical application value.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1821900</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1821900</link>
        <title><![CDATA[SwinPyramidNet: a deep learning network for pear leaf disease classification using hybrid convolutional feature pyramids and swin transformer]]></title>
        <pubdate>2026-06-19T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Tejas Sharma</author><author>Nikhil Shinde</author><author>J. Mukundh</author><author>R. Karthik</author><author>K. Suganthi</author>
        <description><![CDATA[IntroductionPear is rich in dietary fiber, vitamins, and minerals, but its cultivation has been affected by the adverse effects of climate change. Leaf health serves as an early indicator of potential fruit damage, making timely disease detection essential to prevent substantial economic losses. Conventional diagnosis methods, relying on manual visual inspection, are often labor-intensive, costly, and prone to errors. Recent advances in deep learning have enabled accurate image-based disease classification, providing an effective alternative for automated disease detection in precision agriculture.MethodsThis study introduces SwinPyramidNet, a novel dual-track deep learning architecture for pear leaf disease classification. The network integrates the strengths of Convolutional Neural Networks (CNNs) and Transformers to enhance classification performance. The Swin Transformer branch captures global contextual information by modeling long-range dependencies and complex patterns, while the Hybrid Convolutional Feature Pyramid Network (HCFPNet) branch focuses on extracting fine-grained local features through multi-scale convolutional processing. The outputs from both branches are fused and refined using a coordinate attention mechanism, which encodes positional information to emphasize spatially relevant features across channels. This complementary feature extraction strategy enables the network to effectively learn both global and local disease characteristics from pear leaf images.ResultsThe proposed network achieved a classification accuracy of 92.69% on the Diamos dataset. The experimental findings demonstrate the effectiveness of combining Transformer-based global feature learning with CNN-based local feature extraction for pear leaf disease classification. The coordinate attention module further contributed to improved feature representation by highlighting disease-relevant regions.DiscussionThe obtained results indicate that SwinPyramidNet can serve as a reliable and efficient approach for automated pear leaf disease diagnosis. The integration of global contextual information and fine-grained local features enhances the network’s ability to distinguish between disease classes with high accuracy. These findings highlight the applicability of hybrid CNN–Transformer architectures in smart agriculture and plant disease management systems.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1831207</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1831207</link>
        <title><![CDATA[Detection of bacterial and viral pneumonia in pediatric chest radiographs using fusion of mediastinum and lung imaging biomarkers]]></title>
        <pubdate>2026-05-29T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Sukanta Kumar Tulo</author><author>Shilpi Ruchi Kerketta</author><author>O. Rahul Manohar</author><author>Pramod Martha</author>
        <description><![CDATA[IntroductionDiagnosis of pneumonia in pediatric patients remains challenging due to the similarity of manifestations across different pneumonia types. Evaluation of variations in two clinically significant regions, the lung and the mediastinum, on chest radiographs could assist in accurate disease identification. In this work, the morphological characteristics of the mediastinum and lungs are analyzed, and multiple wrapper-based biomarker fusion techniques are employed to enhance the differentiation of bacterial and viral pneumonia.MethodsThe pediatric radiographic images are acquired from a publicly accessible dataset. A hybrid segmentation model combining edge and region-based level set techniques is employed to segment the lungs and mediastinum. Furthermore, morphological imaging biomarkers such as geometric and Hu moments are extracted from the segmented masks and statistically analyzed. Multiple wrapper-based biomarker fusion methods are implemented using Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) algorithms. Further, the optimal fused imaging biomarkers are fed to LDA and SVM classifiers to differentiate the conditions.Results and DiscussionResults indicate that the employed hybrid model could segment both lungs and mediastinum regions from raw radiographic images. The extracted imaging biomarkers effectively characterize the morphological variations. In bacterial pneumonia, the mean lung area is reduced, whereas the mean mediastinum area is increased compared to viral pneumonia. The SVM classifier provided better F-measures of 75.5%, 81.3%, and 82.9% to differentiate bacterial and viral pneumonia using individual mediastinum, lung, and fused biomarkers, respectively, compared to the LDA classifier. Further, enhanced F-measures of 76.5%, 82.0%, and 87.3% are obtained using the LDA-based wrapper selected mediastinum, lung, and fused biomarkers, respectively. The findings indicate that fusion of imaging biomarkers from the lung and mediastinum regions achieves better performance than individual biomarkers.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1812987</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1812987</link>
        <title><![CDATA[InvZW: invariant feature learning via noise-adversarial training for robust image zero-watermarking]]></title>
        <pubdate>2026-05-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>Abdullah All Tanvir</author><author>Frank Y. Shih</author><author>Xin Zhong</author>
        <description><![CDATA[This paper introduces a novel deep learning framework for robust image zero-watermarking based on distortion-invariant feature learning. As a zero-watermarking scheme, our method leaves the original image unaltered and learns a reference signature through optimization in the feature space. The proposed framework consists of two key modules. In the first module, a feature extractor is trained via noise-adversarial learning to generate representations that are both invariant to distortions and semantically expressive. This is achieved by combining adversarial supervision against a distortion discriminator and a reconstruction constraint to retain image content. In the second module, we design a learning-based multibit zero-watermarking scheme where the trained invariant features are projected onto a set of trainable reference codes optimized to match a target binary message. Extensive experiments on diverse image datasets and a wide range of distortions show that our method achieves state-of-the-art robustness in both feature stability and watermark recovery. Comparative evaluations against existing self-supervised and deep watermarking techniques further highlight the superiority of our framework in generalization and robustness.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1787092</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1787092</link>
        <title><![CDATA[Multi-scale pyramid spatial atrous channel attention enhanced ResNet50V2 for explainable pulmonary fibrosis diagnosis]]></title>
        <pubdate>2026-05-13T00:00:00Z</pubdate>
        <category>Original Research</category>
        <author>K. Mahapackialakshmi</author><author>G. Jaffino</author>
        <description><![CDATA[IntroductionPulmonary fibrosis is a progressive interstitial lung disease where delayed or erroneous diagnosis may result in severe clinical consequences. While chest CT imaging serves as the principal method for evaluation, manual interpretation is time-consuming and susceptible to observer variability. Moreover, most modern technologies operate as “black boxes,” resulting in diminished transparency in clinical decision-making. To overcome these limitations, this study proposes a novel Pyramid Spatial Atrous Channel Attention (PSACA)-based feature enhancement module integrated into the ResNet50V2 deep transfer learning framework for automated pulmonary fibrosis classification from chest CT images.MethodsThe proposed model integrates with PSACA into the ResNet50V2 backbone, prioritizing spatial attention before channel attention, with Atrous Spatial Pyramid Pooling added between them to enhance discriminative feature representation in disease-relevant regions. A multi-level pyramid model, which is designed on parallel dilated convolutions, provides hierarchical contextual detail by incrementally increasing receptive fields, providing the ability to model robustly local-to-global pulmonary patterns. By combining spatial attention, atrous spatial pyramid pooling, and channel attention, the fibrosis-relevant regions and informative feature channels are selectively enhanced and lead to a better detection of normal and fibrotic lung tissues. This combined block is embedded in Stage 3 and Stage 4 of ResNet50V2 to maximize high-level fibrosis representation while keeping computational complexity low.Results and discussionExperimental evaluation on a balanced CT dataset demonstrates that the proposed method achieves a classification accuracy of 99.83%, sensitivity 99.93%, specificity 99.72%, F1-score 99.83%, and precision 99.72%. Grad-CAM++ explainability yields clinically meaningful heatmaps that highlight fibrotic abnormalities, enhancing radiologist confidence and interpretability. Overall, the proposed architecture presents a novel multi-scale attention mechanism designed for pulmonary fibrosis, providing enhanced feature discrimination, greater localization, and enhanced clinical explainability.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1795809</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1795809</link>
        <title><![CDATA[Explainable artificial intelligence for the analysis of histopathological images of breast cancer: methods, interpretability and emerging directions]]></title>
        <pubdate>2026-05-12T00:00:00Z</pubdate>
        <category>Mini Review</category>
        <author>Raimondo Fanale</author><author>Barbara Martini</author><author>Filippo Sciarrone</author><author>Roberto Caldelli</author>
        <description><![CDATA[Recent advances in image processing and artificial intelligence have significantly improved the analysis of histopathological images for breast cancer. Deep learning models applied to high-resolution histopathological images (Whole Slide Images, WSIs) are capable of capturing complex, multiscale morphological patterns, achieving high performance in classification, segmentation and risk stratification tasks. However, the increasing architectural complexity of these models has introduced major interpretability issues, limiting their reliability and adoption in sensitive medical image processing applications. Explainable Artificial Intelligence (XAI) has emerged as a key research area within signal and image processing, aiming to make the internal representations and decision-making mechanisms of complex models more transparent and understandable. In the histopathological context, signal- and feature-level interpretability is essential to verify that model predictions are driven by meaningful morphological and textural patterns rather than by spurious correlations or acquisition artefacts. This review work analyses and synthesizes the main XAI techniques applied to the analysis of breast cancer histopathological images, including saliency-based methods, feature attribution approaches, concept-based techniques and intrinsically interpretable architectures. The strengths and limitations of each approach are discussed from an image processing perspective, with particular emphasis on conceptual aspects of spatial localization, multiscale coherence and signal fidelity. Finally, an emerging trend towards multidimensional and composite approaches to explainability is highlighted, which may support the development of standardized evaluation strategies and the design of explainable-by-design image analysis systems.]]></description>
      </item><item>
        <guid isPermaLink="true">https://www.frontiersin.org/articles/10.3389/frsip.2026.1797749</guid>
        <link>https://www.frontiersin.org/articles/10.3389/frsip.2026.1797749</link>
        <title><![CDATA[Explainable artificial intelligence approaches in cardiovascular imaging: methodological advances and clinical implications]]></title>
        <pubdate>2026-05-04T00:00:00Z</pubdate>
        <category>Mini Review</category>
        <author>Wentao Yan</author><author>Rui Sun</author><author>Li Shen</author>
        <description><![CDATA[Cardiovascular diseases remain the leading cause of mortality worldwide, making accurate and efficient imaging-based diagnosis indispensable. Modern modalities such as Coronary Computed Tomography Angiography, Cardiac Magnetic Resonance, Echocardiography, and Chest X-Ray enable rich structural and functional assessment; however, the rapid growth of imaging data strains traditional analysis. Deep learning has markedly improved performance across cardiovascular imaging tasks, yet its “black box” nature limits interpretability, clinician trust, and clinical adoption. eXplainable Artificial Intelligence (XAI) addresses this gap by exposing the decision logic of models in human-understandable forms. This review provides a structured synthesis of recent progress in XAI for cardiovascular imaging. We outline the core principles of perturbation-based and backpropagation-based methods, and survey their applications across major modalities for disease characterization, lesion discrimination, and risk stratification. We further analyze current evaluation challenges and methodological limitations, and propose future directions toward robust, trustworthy, and clinically deployable XAI systems.]]></description>
      </item>
      </channel>
    </rss>