Epileptic seizure prediction based on electroencephalogram (EEG) signals is one of the critical applications of medical artificial intelligence (AI), with considerable clinical potential for improving the quality of life of patients through early warnings. However, existing prediction models face dual challenges: insufficient feature representation and limited explainability of the decision. To address these challenges, this study proposes a dynamic multiscale cross-band fusion filter network (MCFNet) for end-to-end seizure prediction. Specifically, the model first decomposes EEG signals into multiscale components and incorporates a cross-band fusion attention mechanism to achieve multi-granularity signal fusion. Subsequently, the synchronous spectral filtering network, comprising both static and dynamic filtering modules, is designed to capture the periodic components and cross-channel dependencies in EEG signals. Notably, two explainable methods are introduced: a joint feature visualization strategy and an efficient feature ablation analysis, helping to bridge the gap between the "black-box" nature of deep learning and clinical needs. Evaluated on the CHB-MIT dataset, MCFNet achieves a sensitivity of 97.13%, a specificity of 97.22%, and a false positive rate (FPR) of 0.0326/h. Experimental results show that MCFNet not only exhibits superior predictive performance but also maintains a low FPR, offering a feasible scheme for clinical application of EEG-based seizure prediction.

Multi-scale orthogonal Gabor filters based ConvNets for illumination robust single sample face recognition

Abstract Contactless fingerprint acquisition systems encounter significant challenges due to image quality variations, illumination conditions, and evaluation ambiguities that compromise biometric identification accuracy. These challenges constitute Multi-Criteria Decision Making ( $$\text{MCDM}$$ ) problems involving both qualitative and quantitative assessment criteria, which traditional enhancement methods inadequately address due to limited uncertainty quantification capabilities. This paper proposes a novel multi-scale Neutrosophic Similarity Scale ( $$\textrm{NSS}$$ ) enhancement framework for contactless fingerprint recognition. The methodology transforms images into the neutrosophic domain using three membership functions: truth (T), indeterminacy (I), and falsity (F). The framework systematically evaluates multiple filter scales ( $$3 \times 3$$ , $$5 \times 5$$ , $$7 \times 7$$ , $$9 \times 9$$ ) to optimize enhancement performance across varying degrees of image degradation and noise conditions. Comprehensive evaluation integrates objective measures with subjective quality assessments using 5-point fuzzy scales, validated through Intraclass Correlation Coefficient (ICC) and Analysis of Variance (ANOVA). The proposed NSS method with $$3 \times 3$$ filter configuration achieves the highest overall subjective evaluation score of 4.28. Benchmark comparisons with state-of-the-art techniques including U-Net, Feature Pyramid Network (FPN), ResNet, and GAN-based methods demonstrate superior performance in fingerprint feature clarity. Sensitivity analysis and ablation studies validate individual contributions of enhancement criteria $$\mathcal {C}_{\alpha }$$ , $$\mathcal {C}_{\beta }$$ , and $$\mathcal {C}_{\gamma }$$ . Results demonstrate significant improvements in managing vagueness and uncertainty while preserving structural information, establishing the method’s suitability for reliable contactless biometric identification systems.

MSAF: Multi-scale adaptive filter for object tracking

Myelinated fiber labeling and orientation mapping of the human brain with light-sheet fluorescence microscopy

Peach firmness is a critical quality attribute, yet conventional destructive measurement methods are unsuitable for batch detection in industrial settings. This study investigated a noncontact method for firmness assessment across multiple peach cultivars based on acoustic vibration technology. Three peach cultivars were mechanically excited via a controlled air jet, and the resulting acoustic vibration responses were captured noninvasively using a laser Doppler vibrometer. The frequency-domain acoustic vibration spectra were used as input for firmness prediction models developed using partial least squares regression (PLSR), support vector regression (SVR), and a one-dimensional convolutional neural network (ISNet-1D) that incorporated Inception and squeeze-and-excitation modules. Comparative analysis demonstrated that the ISNet-1D substantially outperformed the conventional linear and nonlinear methods on an independent test set, achieving superior predictive accuracy, with a coefficient of determination ( RP2) of 0.8069, a root mean square error (RMSEP) of 0.9206 N/mm, and a residual prediction deviation ( RPDP) of 2.2879. The good performance of the ISNet-1D can be attributed to the integration of multi-scale convolutional filters with a channel-wise attention mechanism. This integration allows the network to adaptively prioritize discriminative spectral features, thereby enhancing its prediction accuracy. A hierarchical transfer learning strategy was proposed to improve model generalizability, offering a practical and cost-effective means to adapt to diverse cultivars. In summary, the combination of noncontact acoustic vibration and deep learning presents a robust, accurate, and nondestructive methodology for assessing peach firmness, demonstrating considerable potential for cross-cultivar application in industrial sorting and quality control.

The tongue is a critical soft tissue of the upper airway; excessive enlargement can narrow or collapse the airway, contributing to obstructive sleep apnea (OSA). To our knowledge, no existing methods quantify OSA-related tongue geometric features using deep learning segmentation. We present a deep learning model (TOSA-Net) capable of accurately segmenting and measuring tongue geometry, offering a more efficient approach for future OSA tongue research. A dataset (n = 207) of front and profile tongue images, along with manually segmented and quantified tongue dimensions, was used for model development and evaluation. We modified a U-Net architecture using multi-scale convolution filters for feature extraction. Automated segmentations were used to calculate tongue geometric features. Dice coefficient, Pearson correlation, agreement analyses, and expert-derived clinical parameters assessed segmentation and measurement accuracy between deep learning and manual methods. Five-fold cross-validation yielded a mean Dice coefficient of 0.870 across all subjects, with the highest mean Dice (0.914) in OSA patients. Tongue features (area, length, thickness, curvature) showed strong correlation with manual measurements, with no statistically significant differences [Formula: see text]). The high accuracy of automated segmentation and measurement indicates the proposed method's potential to replace time-consuming manual tasks in clinical and large-scale OSA research.

Graphite nodularity is a key indicator for evaluating the microstructure quality of ductile iron and plays a crucial role in ensuring product quality and enhancing manufacturing efficiency. Existing research often only focuses on a single type of feature and fails to utilize multi-source information in a coordinated manner. Single-feature methods are difficult to comprehensively capture microstructures, which limits the accuracy and robustness of the model. This study proposes a hybrid estimation model for the graphite nodularity of ductile cast iron based on multi-source feature extraction. A comprehensive feature engineering pipeline was established, incorporating geometric, color, and texture features extracted via Hue-Saturation-Value color space (HSV) histograms, gray level co-occurrence matrix (GLCM), Local Binary Pattern (LBP), and multi-scale Gabor filters. Dimensionality reduction was performed using Principal Component Analysis (PCA) to mitigate redundancy. An improved watershed algorithm combined with intelligent filtering was used for accurate particle segmentation. Several machine learning algorithms, including Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Random Forest (RF), Gradient Boosting Regressor (GBR), eXtreme Gradient Boosting (XGBoost) and Categorical Boosting (CatBoost), are applied to estimate graphite nodularity based on geometric features (GFs) and feature extraction. Experimental results demonstrate that the CatBoost model trained on fused features achieves high estimation accuracy and stability for geometric parameters, with R-squared (R2) exceeding 0.98. Furthermore, introducing geometric features into the fusion set enhances model generalization and suppresses overfitting. This framework offers an efficient and robust approach for intelligent analysis of metallographic images and provides valuable support for automated quality assessment in casting production.

MERFusion: A multiscale edge-preserving filter combined with Retinex enhancement for infrared and visible image fusion

ABSTRACT As the core component of the railway train, the healthy state of the bogie bearing is essential for the safe operation of the train. The traditional bearing fault diagnosis methods typically rely on a single source signal, which cannot fully capture fault feature information, resulting in low diagnosis accuracy. To address this problem, this paper proposes a new diagnosis framework for train bogie bearings based on multi-source data fusion and multi-scale residual network (MF-MSRNet). Firstly, a multi-source data fusion method is designed to extract fault feature information from voiceprint, acoustic emission, and vibration sensor data, effectively extracting the low-dimensional features embedded in high-dimensional nonlinear data and fusing them into RGB images. Secondly, a new dual-scale residual block is presented to learn both profound and shallow features at various scales, thereby capturing bearing fault information in different spatial dimensions and enhancing the capacity to extract fusion features. Finally, to enhance the adaptability of the residual network in noisy scenes, a new denoising module is designed to help the network explore multi-scale features and filter irrelevant information. The experimental outcomes indicate that the classification accuracy of MSRNet in traction motor bearing datasets can reach up to 99.75%, and its comprehensive fault diagnosis performance is the best among all models.

Palmprint recognition, as a biometric recognition technology, has unique individual recognition and high accuracy, and is broadly utilized in fields such as identity verification and security monitoring. Therefore, a palm print recognition model that integrates regions of interest and Gabor filters has been proposed to solve the problem of difficulty in feature extraction caused by factors such as noise, lighting changes, and acquisition angles that often affect palm print images during the acquisition process. This model extracts standardized feature regions of palmprint images through the region of interest method, enhances texture features through multi-scale Gabor filters, and finally uses support vector machines for classification. The experiment findings denote that the region of interest model performs better than other methods in terms of signal-to-noise ratio and root mean square error, with a signal-to-noise ratio of 0.89 on the GPDS dataset and 0.97 on the CASIA dataset. The proposed model performs the best in recognition accuracy and error convergence speed, with a final accuracy of 95%. The proposed model has the shortest running time, less than 0.4 seconds in all groups, especially less than 0.3 seconds in Group 4, demonstrating high recognition efficiency. The research conclusion shows that the palmprint recognition method combining regions of interest and Gabor filters has high efficiency and performance, and can effectively improve recognition accuracy.

BackgroundIt is challenging to identify Papillary Thyroid Cancer (PTC) which shows atypia of undetermined significance (AUS) by Fine-needle Aspiration (FNA). This study aims to seek the meaningful quantitative biomarkers of the microvasculature and construct a classification model for PTC with AUS based on these new biomarkers and Thyroid Imaging Reporting and Data System (TI-RADS).MethodsThis prospective study enrolled 281 patients with 300 thyroid nodules showing AUS. These cases were divided into two groups with the largest dimension (LD) of 10 mm, A (< 10 mm) and B (≥ 10 mm). Firstly, an open-source artifact suppression algorithm, which combined a multi-scale Frangi filter and TOPHAT operation, was proposed for the segmentation of micro-vessels in Ultra Micro-Angiography (UMA) images. Then, 18 quantitative biomarkers were calculated and analyzed through Mann-Whitney test (U-test), while LASSO regression was utilized to remove collinear features. Finally, two different classification models were built using logistic regression through the selected biomarkers combined with Chinese TI-RADS (C TI-RADS) or American College of Radiology TI-RADS (ACR TI-RADS). The performances were evaluated using the mean Area Under the Curve (AUC) value and the DeLong test, through a 5-fold cross-validation experiment.ResultsGroup A comprised 58 benign nodules and 104 PTCs, while Group B consisted of 60 benign nodules and 78 PTCs. Four biomarkers were selected in Group A. The 5-fold cross-validation experiment showed that the mean Area Under Curve (AUC) improved from 0.725 with ACR TI-RADS to 0.851 (P < 0.05), while the mean AUC improved from 0.809 with C TI-RADS to 0.882 (P < 0.05). In Group B, four different biomarkers were selected, and the classification models showed improvements from 0.841 with ACR TI-RADS to 0.874 and from 0.894 with C TI-RADS to 0.936.ConclusionsThis study demonstrated the potential value of microvasculature in the prediction of PTC in AUS Cases and improved the performance of ultrasound examination. Moreover, the morphology of microvasculature showed different changes at different LD groups.

The convoluted network of myelinated fibers that supports behavior, cognition, and sensory processing in the human brain is the source of its extraordinary complexity. Advancements in tissue optical clearing, 3D fluorescence microscopy, and automated image analysis have enabled unprecedented insights into the architecture of these networks. Here, we investigate the multiscale organization of myelinated fibers in human brain tissue from the brainstem, Broca’s area, hippocampus, and primary visual cortex by exploiting a specific fiber staining method, light-sheet fluorescence microscopy (LSFM), and an advanced spatial orientation analysis tool. Using an optimized protocol that integrates tissue clearing with the lipophilic DiD probe to achieve uniform and deep myelinated fiber labeling, we generate micrometerresolution volumetric reconstructions of multiple brain regions through an inverted LSFM. Automated image processing, employing unsupervised 3D multiscale Frangi filters, provides orientation distribution functions and local orientation dispersion maps. This enables precise characterization of the directionality of white matter bundles, linking mesoscopic structural properties to orientation details computed at the native micrometric resolution of the LSFM apparatus. The presented workflow illustrates a robust platform for large-scale, high-resolution brain mapping, which may facilitate the investigation of pathological alterations with unparalleled spatial resolution and, furthermore, the validation of other neuroimaging modalities.

Weighted multiscale combined difference morphological filter for incipient bearing fault diagnosis

To address the challenges of low accuracy and the difficulty in balancing a large field of view and long distance when tracking high-speed moving targets with a single sensor, an ROI adaptive digital zoom tracking method is proposed. In this paper, we discuss the impact of ROI on image processing and describe the design of the ROI adaptive digital zoom tracking system. Additionally, we construct an adaptive ROI update model based on normalized target information. To capture target changes effectively, we introduce the multi-scale regional measure and propose an improved particle filter algorithm, referred to as the improved multi-scale regional measure resampling particle filter (IMR-PF). This method enables high temporal resolution processing efficiency within a high-resolution large field of view, which is particularly beneficial for high-resolution videos. The IMR-PF can maintain high temporal resolution within a wide field of view with high resolution. Simulation results demonstrate that the improved target tracking method effectively improves tracking robustness to target motion changes and reduces the tracking center error by 20%, as compared to other state-of-the-art methods. The IMR-PF still maintains good performance even when confronted with various interference factors and in real-world scenario applications.

Sinkholes are significant geohazards in karst regions that pose risks to landscapes and infrastructure by disrupting geological stability. Usually, sinkholes are mapped by field surveys, which is very cost-intensive with regard to vast coverages. One possible solution to derive sinkholes without entering the area is the use of high-resolution digital terrain models, which are also expensive with respect to remote areas. Therefore, this study focusses on the mapping of sinkholes in arid regions from open-access remote sensing data. The case study involves data from the Sentinel missions over the Mangystau region in Kazakhstan provided by the European Space Agency free of cost. The core of the technique is a multi-scale curvature filter bank that highlights sinkholes (and takyrs) by their very special illumination pattern in Sentinel-2 images. Marginal confusions with vegetation shadows are excluded by consulting the newly developed Combined Vegetation Doline Index based on Sentinel-1 and Sentinel-2. The geospatial analysis reveals distinct spatial correlations among sinkholes, takyrs, vegetation, and possible surface discharge. The generic and, therefore, transferable approach reached an accuracy of 92%. However, extensive reference data or comparable methods are not currently available.

Faulty-Feeder Detection Based on Sparse Waveform Encoding and Simple Convolutional Neural Network with Multi-Scale Filters and One Layer of Convolution

Graph node anomaly detection has important applications in practical scenarios. Although many graph neural networks (GNNs) have been proposed, how to design tailored spectral filters for node anomaly detection to fully mine high-frequency signals in the graph is still a challenge. Most GNNs are equivalent to low-pass filters and mine multiorder signals through a series structure. The computational cost increases as the number of layers increases and further leads to an over-smoothing problem. They mainly focus on low-frequency signals and suppress high-frequency signals, thus smoothing the differences between abnormal and normal nodes, making them indistinguishable. Due to the difficulty in mining high-frequency signals, the poorly distinguishable feature representations learned by low-pass GNNs can even harm the performance of data augmentation. To solve the above challenges, in this article, we propose a or-gate mixup multiscale spectral GNN (MMGNN) from the spectral domain. Specifically, we design multiorder multiscale bandpass filters through the superposition of polynomial spectral filters and then decompose them into preprocessing parts and training parts to form a double-parallel structure, which can effectively mine high-frequency signals in the graph and reduce computational cost. Finally, we propose or-gate mixup to perform data augmentation in the spectral space to improve model generalization. Experimental results on four real-world datasets demonstrate the effectiveness of the proposed MMGNN against the state-of-the-art methods.

BackgroundThe subcellular localization of mRNA plays a crucial role in gene expression regulation and various cellular processes. However, existing wet lab techniques like RNA-FISH are usually time-consuming, labor-intensive, and limited to specific tissue types. Researchers have developed several computational methods to predict mRNA subcellular localization to address this. These methods face the problem of class imbalance in multi-label classification, causing models to favor majority classes and overlook minority classes during training. Additionally, traditional feature extraction methods have high computational costs, incomplete features, and may lead to the loss of critical information. On the other hand, deep learning methods face challenges related to hardware performance and training time when handling complex sequences. They may suffer from the curse of dimensionality and overfitting problems. Therefore, there is an urgent need for more efficient and accurate prediction models.ResultsTo address these issues, we propose a multi-label classifier, EDCLoc, for predicting mRNA subcellular localization. EDCLoc reduces training pressure through a stepwise pooling strategy and applies grouped convolution blocks of varying sizes at different levels, combined with residual connections, to achieve efficient feature extraction and gradient propagation. The model employs global max pooling at the end to further reduce feature dimensions and highlight key features. To tackle class imbalance, we improved the focal loss function to enhance the model’s focus on minority classes. Evaluation results show that EDCLoc outperforms existing methods in most subcellular regions. Additionally, the position weight matrix extracted by multi-scale CNN filters can match known RNA-binding protein motifs, demonstrating EDCLoc’s effectiveness in capturing key sequence features.ConclusionsEDCLoc outperforms existing prediction tools in most subcellular regions and effectively mitigates class imbalance issues in multi-label classification. These advantages make EDCLoc a reliable choice for multi-label mRNA subcellular localization. The dataset and source code used in this study are available at https://github.com/DellCode233/EDCLoc.

This paper introduces a hybrid technique designed to identify fabric defects in the early stages of textile manufacturing production. Manual detection of fabric abnormalities is a challenging task, prompting the development of an intelligent system that leverages techniques such as multi-scale Gabor transform and gray level co-occurrence matrix (GGCM) for fabric defect diagnosis. In the preprocessing phase, the input image undergoes down-sampling and quantization. Subsequently, efficient noise removal is achieved using a median filter, followed by the extraction of the region of interest through histogram matching. During the segmentation stage, a band-pass filter is formed by manipulating a 1-D Gaussian filter in frequency space, creating what is known as a Circular Gaussian Filter (CGF) by rotating it off-center. A CGF is distinguished by its unique definition involving the specification of both a central frequency and a frequency band. Typically, fabric flaws appear as brighter regions in the processed image. Defect regions are identified by dynamically establishing a threshold through histogram analysis of the CGF-filtered image. The subsequent step involves extracting defects by leveraging gray level co-occurrence matrix (GLCM) features. This method is rigorously compared with current mainstream algorithms across various types of fabric defects, accompanied by an analysis of its strengths and weaknesses. The experimental results demonstrate that the algorithm achieves a high detection success rate and precision in defect marking, indicating promising potential for practical applications.