Deep Learning Neural Network Research Articles

Automated High-Precision Recognition of Solar Filaments Based on an Improved U2-Net

Solar filaments are a significant solar activity phenomenon, typically observed in full-disk solar observations in the H-alpha band. They are closely associated with the magnetic fields of solar active regions, solar flare eruptions, and coronal mass ejections. With the increasing volume of observational data, the automated high-precision recognition of solar filaments using deep learning is crucial. In this study, we processed full-disk H-alpha solar images captured by the Chinese H-alpha Solar Explorer in 2023 to generate labels for solar filaments. The preprocessing steps included limb-darkening removal, grayscale transformation, K-means clustering, particle erosion, multiple closing operations, and hole filling. The dataset containing solar filament labels is constructed for deep learning. We developed the Attention U2-Net neural network for deep learning on the solar dataset by introducing an attention mechanism into U2-Net. In the results, Attention U2-Net achieved an average Accuracy of 0.9987, an average Precision of 0.8221, an average Recall of 0.8469, an average IoU of 0.7139, and an average F1-score of 0.8323 on the solar filament test set, showing significant improvements compared to other U-net variants.

Distance-based loss function for deep feature space learning of convolutional neural networks

Convolutional Neural Networks (CNNs) have been on the forefront of neural network research in recent years. Their breakthrough performance in fields such as image classification has gathered efforts in the development of new CNN-based architectures, but recently more attention has been directed to the study of new loss functions. Softmax loss remains the most popular loss function due mainly to its efficiency in class separation, but the function is unsatisfactory in terms of intra-class compactness. While some studies have addressed this problem, most solutions attempt to refine softmax loss or combine it with other approaches. We present a novel loss function based on distance matrices (LDMAT), softmax independent, that maximizes interclass distance and minimizes intraclass distance. The loss function operates directly on deep features, allowing their use on arbitrary classifiers. LDMAT minimizes the distance between two distance matrices, one constructed with the model’s deep features and the other calculated from the labels. The use of a distance matrix in the loss function allows a two-dimensional representation of features and imposes a fixed distance between classes, while improving intra-class compactness. A regularization method applied to the distance matrix of labels is also presented, that allows a degree of relaxation of the solution and leads to a better spreading of features in the separation space. Efficient feature extraction was observed on datasets such as MNIST, CIFAR10 and CIFAR100.

Flotation separation of lithium–ion battery electrodes predicted by a long short-term memory network using data from physicochemical kinetic simulations and experiments

Anode and cathode active materials from spent lithium–ion batteries may be recovered and potentially used in new batteries to promote recycling and resource circulation. Froth flotation was applied to pristine active materials and the black mass obtained from pretreated spent batteries. Flotation kinetics was simulated with the use of computational fluid dynamics and surface chemistry. Bubble surface coverage and entrainment in the flotation kinetics model were selected and optimized by systematic fitting to experimental data. Entrainment influences the recovery and grade of the active materials. The optimized flotation kinetics model was used for generating additional data that, along with the fitted data, were used to train a deep learning neural network. The trained network was validated using anode–cathode and black mass flotation experiments, and its predictions showed a maximum residual error of 0.18 ± 0.11 recovery. The simulation framework was developed into a desktop application that predicts the flotation behavior of active materials. It provides information for estimating results following operational and physicochemical changes and for optimizing flotation processes. Finally, recovered anode active materials from black mass were selected for coin cell tests. The coulombic efficiency of these coin cells was initially lower (86.8 %) than that of cells made with pristine graphite particles (98.4 %).

Deep learning-based segmentation for high-dose-rate brachytherapy in cervical cancer using 3D Prompt-ResUNet

Objective.To develop and evaluate a 3D Prompt-ResUNet module that utilized the prompt-based model combined with 3D nnUNet for rapid and consistent autosegmentation of high-risk clinical target volume (HRCTV) and organ at risk (OAR) in high-dose-rate brachytherapy for cervical cancer patients.Approach.We used 73 computed tomography scans and 62 magnetic resonance imaging scans from 135 (103 for training, 16 for validation, and 16 for testing) cervical cancer patients across two hospitals for HRCTV and OAR segmentation. A novel comparison of the deep learning neural networks 3D Prompt-ResUNet, nnUNet, and segment anything model-Med3D was applied for the segmentation. Evaluation was conducted in two parts: geometric and clinical assessments. Quantitative metrics included the Dice similarity coefficient (DSC), 95th percentile Hausdorff distance (HD95%), Jaccard index (JI), and Matthews correlation coefficient (MCC). Clinical evaluation involved interobserver comparison, 4-grade expert scoring, and a double-blinded Turing test.Main results.The Prompt-ResUNet model performed most similarly to experienced radiation oncologists, outperforming less experienced ones. During testing, the DSC, HD95% (mm), JI, and MCC value (mean ± SD) for HRCTV were 0.92 ± 0.03, 2.91 ± 0.69, 0.85 ± 0.04, and 0.92 ± 0.02, respectively. For the bladder, these values were 0.93 ± 0.05, 3.07 ± 1.05, 0.87 ± 0.08, and 0.93 ± 0.05, respectively. For the rectum, they were 0.87 ± 0.03, 3.54 ± 1.46, 0.78 ± 0.05, and 0.87 ± 0.03, respectively. For the sigmoid, they were 0.76 ± 0.11, 7.54 ± 5.54, 0.63 ± 0.14, and 0.78 ± 0.09, respectively. The Prompt-ResUNet achieved a clinical viability score of at least 2 in all evaluation cases (100%) for both HRCTV and bladder and exceeded the 30% positive rate benchmark for all evaluated structures in the Turing test.Significance.The Prompt-ResUNet architecture demonstrated high consistency with ground truth in autosegmentation of HRCTV and OARs, reducing interobserver variability and shortening treatment times.

Length-scale study in deep learning prediction for non-small cell lung cancer brain metastasis

Deep learning-assisted digital pathology has demonstrated the potential to profoundly impact clinical practice, even surpassing human pathologists in performance. However, as deep neural network (DNN) architectures grow in size and complexity, their explainability decreases, posing challenges in interpreting pathology features for broader clinical insights into physiological diseases. To better assess the interpretability of digital microscopic images and guide future microscopic system design, we developed a novel method to study the predictive feature length-scale that underpins a DNN’s predictive power. We applied this method to analyze a DNN’s capability in predicting brain metastasis from early-stage non-small-cell lung cancer biopsy slides. This study quantifies DNN’s attention for brain metastasis prediction, targeting features at both the cellular scale and tissue scale in H&E-stained histological whole slide images. At the cellular scale, the predictive power of DNNs progressively increases with higher resolution and significantly decreases when the resolvable feature length exceeds 5 microns. Additionally, DNN uses more macro-scale features associated with tissue architecture and is optimized when assessing visual fields greater than 41 microns. Our study computes the length-scale requirements for optimal DNN learning on digital whole-slide microscopic images, holding the promise to guide future optical microscope designs in pathology applications and facilitating downstream deep learning analysis.

Identifying defects and varieties of Malting Barley Kernels

This study introduces a comprehensive approach for classifying individual malting barley kernels, involving dual-sided kernel imaging, a specifically designed image processing algorithm, an optimized deep neural network architecture, and a mechanical sorting system. The proposed method achieves precise classification into multiple classes, aligning with quality standards for malting material assessment. Throughout the study, various image analysis techniques were assessed, including traditional feature engineering, established transfer learning deep neural network architectures, and our custom-designed convolutional neural network tailored for barley kernel image analysis. Comparative analysis underscores the superior performance of our network model. The study reveals that our proposed deep learning network achieves a 94% accuracy in classifying barley kernel defects and varieties, outperforming well-established transfer learning models to complex architectures that attain 93% accuracy. Additionally, it surpasses the traditional machine learning approach involving feature extraction and support vector machine classifiers, which achieve accuracy below 90% in detecting defective kernels and below 70% in varietal classification. However, we also noted the traditional approach’s advantage in morphological feature recognition. This observation guides new research toward integrating morphological feature extraction techniques with modern convolutional networks. This paper presents a deep neural network designed specifically for the analysis of cereal kernel images in two applications: defect and variety classification. It emphasizes the importance of standardizing kernel orientation and merging images from both sides of the kernel, and introduces a device for image acquisition that fulfills this need.

Multifunctional Human-Computer Interaction System Based on Deep Learning-Assisted Strain Sensing Array.

Continuous and reliable monitoring of gait is crucial for health monitoring, such as postoperative recovery of bone joint surgery and early diagnosis of disease. However, existing gait analysis systems often suffer from large volumes and the requirement of special space for setting motion capture systems, limiting their application in daily life. Here, we develop an intelligent gait monitoring and analysis prediction system based on flexible piezoelectric sensors and deep learning neural networks with high sensitivity (241.29 mV/N), quick response (66 ms loading, 87 ms recovery), and excellent stability (R2 = 0.9946). The theoretical simulations and experiments confirm that the sensor provides exceptional signal feedback, which can easily acquire accurate gait data when fitted to shoe soles. By integrating high-quality gait data with a custom-built deep learning model, the system can detect and infer human motion states in real time (the recognition accuracy reaches 94.7%). To further validate the sensor's application in real life, we constructed a flexible wearable recognition system with human-computer interaction interface and a simple operation process for long-term and continuous tracking of athletes' gait, potentially aiding personalized health management, early detection of disease, and remote medical care.

Wind Field Reconstruction Method Using Incomplete Wind Data Based on Vision Mamba Decoder Network

Accurate meteorological information is crucial for the safety of civil aviation flights. Complete wind field information is particularly helpful for planning flight routes. To address the challenge of accurately reconstructing wind fields, this paper introduces a deep learning neural network method based on the Vision Mamba Decoder. The goal of the method is to reconstruct the original complete wind field from incomplete wind data distributed along air routes. This paper proposes improvements to the Vision Mamba model to fit our mission, showing that the developed model can accurately reconstruct the complete wind field. The experimental results demonstrate a mean absolute error (MAE) of wind speed of approximately 1.83 m/s, a mean relative error (MRE) of around 7.87%, an R-square value of about 0.92, and an MAE of wind direction of 5.78 degrees.

Jet Tagging with More-Interaction Particle Transformer

Abstract In this study, we introduce the More-Interaction Particle Transformer (MIParT), a novel deep learning neural network designed for jet tagging. This framework incorporates our own design, the More-Interaction Attention (MIA) mechanism, which increases the dimensionality of particle interaction embeddings. We tested MIParT using the top tagging and quark-gluon datasets. Our results show that MIParT not only matches the accuracy and AUC of LorentzNet and a series of Lorentz-equivariant methods, but also significantly outperforms the ParT model in background rejection. Specifically, it improves background rejection by approximately 25\% at a 30\% signal efficiency on the top tagging dataset and by 3\% on the quark-gluon dataset. Additionally, MIParT requires only 30\% of the parameters and 53\% of the computational complexity needed by ParT, proving that high performance can be achieved with reduced model complexity. For very large datasets, we double the dimension of particle embeddings, referring to this variant as MIParT-Large (MIParT-L). We find that MIParT-L can further capitalize on the knowledge from large datasets. From a model pre-trained on the 100M JetClass dataset, the background rejection performance of the fine-tuned MIParT-L improved by 39\% on the top tagging dataset and by 6\% on the quark-gluon dataset, surpassing that of the fine-tuned ParT. Specifically, the background rejection of fine-tuned MIParT-L improved by an additional 2\% compared to the fine-tuned ParT. The results suggest that MIParT has the potential to advance efficiency benchmarks for jet tagging and event identification in particle physics.Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. Article funded by SCOAP3 and published under licence by Chinese Physical Society and the Institute of High Energy Physics of the Chinese Academy of Science and the Institute of Modern Physics of the Chinese Academy of Sciences and IOP Publishing Ltd

Enhanced plasmonic scattering imaging via deep learning-based super-resolution reconstruction for exosome imaging.

Exosome analysis plays pivotal roles in various physiological and pathological processes. Plasmonic scattering microscopy (PSM) has proven to be an excellent label-free imaging platform for exosome detection. However, accurately detecting images scattered from exosomes remains a challenging task due to noise interference. Herein, we proposed an image processing strategy based on a new blind super-resolution deep learning neural network, named ESRGAN-SE, to improve the resolution of exosome PSI images. This model can obtain super-resolution reconstructed images without increasing experimental complexity. The trained model can directly generate high-resolution plasma scattering images from low-resolution images collected in experiments. The results of experiments involving the detection of light scattered by exosomes showed that the proposed super-resolution detection method has strong generalizability and robustness. Moreover, ESRGAN-SE achieved excellent results of 35.52036, 0.09081, and 8.13176 in terms of three reference-free image quality assessment metrics, respectively. These results show that the proposed network can effectively reduce image information loss, enhance mutual information between pixels, and decrease feature differentiation. And, the single-image SNR evaluation score of 3.93078 also showed that the distinction between the target and the background was significant. The suggested model lays the foundation for a potentially successful approach to imaging analysis. This approach has the potential to greatly improve the accuracy and efficiency of exosome analysis, leading to more accurate cancer diagnosis and potentially improving patient outcomes.

Enhancing Multi-Object Detection in Ultrasound Images Through Semi-Supervised Learning, Focal Loss and Relation of Frame

ObjectiveTo identify musculoskeletal anatomical structures in real time by using deep learning techniques. MethodsAn automated annotation system based on deep learning neural networks was designed to aid in the real-time identification of anatomical structures. Additionally, novel algorithms aimed at diminishing model training duration while enhancing accuracy were introduced. In this study, we proposed a semi-supervised learning (SSL) approach that substantially reduced annotation time. We also adopted the focal loss (FL) method to enhance the accuracy of challenging structures. Additionally, during the inference stage, we harnessed the temporal continuity of video frames, which involved leveraging information from preceding frames to facilitate recognition of structures in the current image. Training the model through a combination of SSL and FL yielded superior performance compared with supervised learning, while also substantially mitigating any expense linked to annotations. During inference, the incorporation of frame continuity helped to avoid discontinuity and bolster accuracy. ResultsForearm tissue detection was demonstrated by properly configuring the SSL approach, including FL and the filtering threshold. Comparable performance with supervised learning was achieved while only using 30% of the training data. The real-time experimental results also demonstrated that implementing relation of frame reduced the number of missing frames during inference and successfully increased the confidence scores of detected objects. ConclusionThis proposed system has the potential to aid medical professionals in efficiently and effectively diagnosing musculoskeletal disorders, ultimately leading to enhanced patient outcomes.

Balance Controller Design for Inverted Pendulum Considering Detail Reward Function and Two-Phase Learning Protocol

As a complex nonlinear system, the inverted pendulum (IP) system has the characteristics of asymmetry and instability. In this paper, the IP system is controlled by a learned deep neural network (DNN) that directly maps the system states to control commands in an end-to-end style. On the basis of deep reinforcement learning (DRL), the detail reward function (DRF) is designed to guide the DNN learning control strategy, which greatly enhances the pertinence and flexibility of the control. Moreover, a two-phase learning protocol (offline learning phase and online learning phase) is proposed to solve the “real gap” problem of the IP system. Firstly, the DNN learns the offline control strategy based on a simplified IP dynamic model and DRF. Then, a security controller is designed and used on the IP platform to optimize the DNN online. The experimental results demonstrate that the DNN has good robustness to model errors after secondary learning on the platform. When the length of the pendulum is reduced by 25% or increased by 25%, the steady-state error of the pendulum angle is less than 0.05 rad. The error is within the allowable range. The DNN is robust to changes in the length of the pendulum. The DRF and the two-phase learning protocol improve the adaptability of the controller to the complex and variable characteristics of the real platform and provide reference for other learning-based robot control problems.

Automatyczna diagnostyka elementów toru kolejowego z użyciem sieci neuronowych głębokiego uczenia

The article discusses a method for automatic diagnostics of a railway track. It consists in automatic evaluation of the technical condition of selected track elements, such as rails, wooden and concrete sleepers, fasteners and turnouts. It was carried out on the basis of analysis of video images of railroad track elements recorded by two line cameras placed on the diagnostic carriage. The selected FCN-8 deep learning neural network was used to assess the technical condition of the surveyed elements, and the effectiveness of the applied algorithm was determined on the basis of such measures as IoU, Precision, Recall. Conclusions on the application of the FCN-8 network in the automatic classification of features of selected railroad track elements are presented. The results obtained were compared with other methods used in vision diagnostics.

External validation of a deep learning model for automatic segmentation of skeletal muscle and adipose tissue on abdominal computed tomography images.

Body composition assessment using computed tomography (CT) images at the L3-level is increasingly applied in cancer research. Robust high-throughput automated segmentation is key to assess large patient cohorts and to support implementation of body composition analysis into routine clinical practice. We trained and externally validated a deep learning neural network (DLNN) to automatically segment L3-CT images. Expert-drawn segmentations of visceral and subcutaneous adipose tissue (VAT/SAT) and skeletal muscle (SM) of L3-CT-images of 3,187 patients undergoing abdominal surgery were used to train a DLNN. The external validation cohort was comprised of 2,535 patients with abdominal cancer. DLNN performance was evaluated with (geometric) Dice Similarity (DS) and Lin's Concordance Correlation Coefficient. There was a strong concordance between automatic and manual segmentations with median DS for SM, VAT, and SAT of 0.97 (interquartile range, IQR: 0.95-0.98), 0.98 (IQR: 0.95-0.98), and 0.95 (IQR: 0.92-0.97), respectively. Concordance correlations were excellent: SM 0.964 (0.959-0.968), VAT 0.998 (0.998-0.998), and SAT 0.992 (0.991-0.993). Bland-Altman metrics indicated only small and clinically insignificant systematic offsets; SM radiodensity: 0.23 hounsfield units (0.5%), SM: 1.26 cm2.m-2 (2.8%), VAT: -1.02 cm2.m-2 (1.7%), and SAT: 3.24 cm2.m-2 (4.6%). A robustly-performing and independently externally validated DLNN for automated body composition analysis was developed. CT-based body composition analysis is highly prognostic for long-term overall survival in oncology. This DLNN was succesfully trained and externally validated on several large patient cohorts and will therefore enable large scale population studies and implementation of body composition analysis into clinical practice.

An improved man-in-the-middle (MITM) attack detections using convolutional neural networks

The increasing reliance on digital communication networks has made information security a critical concern for individuals, organizations, and governments worldwide. Man-in-the-middle (MITM) attacks are significant, prevalent, and damaging concerning cyber-attacks. Detecting MitM attacks is complex due to their stealthy nature and the sophisticated methods employed by attackers. There is the need for researchers to address this issue using current and novel methods like artificial intelligence. In this paper, an improved MitM attack detection approach using the Convolutional Neural Network (CNN) deep learning algorithm is developed, resulting in an overall detection accuracy of 0.986%. The results confirms that the proposed model is very efficient in comparision to other proposed solutions by other authors.

Advancements and applications of AI technologies in pathology: a scoping review

ABSTRACT The benefits of AI in pathology are demonstrated by the growing potential of diagnostic accuracy and efficiency in tissue segmentation, cancer detection and grading. In this scoping review, we studied how artificial intelligence (AI) technologies have been developed for usage in pathology. In the study, a literature search on databases and publisher portals PubMed, ScienceDirect, ProQuest, and CINAHL was conducted to collect peer-reviewed studies published between 2019 and 2023. Then after a thorough database search, 57 articles covering the topic were included out of 19 185 potential ones and were categorised into groups by their main purposes, such as pre-processing, segmentation, analysis, and prediction. The findings demonstrate that AI technologies like deep learning and convolutional neural networks (CNNs) enhance medical imaging for tissue segmentation and cancer detection by reducing human error and variability.

Improved Cascade-RCNN for automatic detection of coronary artery plaque in multi-angle fusion CPR images

Coronary heart disease is a disease that seriously endangers human health and life which is caused by plaque formation due to coronary artery atherosclerosis or spasm. The detection of coronary plaques through medical imaging is a non-destructive and fast diagnostic method, which holds significant medical and clinical value for the diagnosis of coronary heart disease. In this paper, We proposed an automatic detecting method for coronary artery vulnerable plaque based on CTA image sequence. We collected CTA image sequences from 132 patients and obtained the coronary artery tree by performing three-dimensional (3D) segmentation. The coronary artery centerline was extracted using the skeleton thinning method. To address the issue of insufficient features in a single CPR image, we constructed Curved Planar Reconstructed (CPR) images from multiple directions perpendicular to the coronary artery, following the centerline. In total, we established 8797 images in the training dataset and 1553 images in the test dataset. We proposed an attentive detection method that fused a new Cascade RCNN ResNet50-FPN neural network model with a convolutional attention module (CBAM). This method achieved automatic recognition and detection of plaques on the two-dimensional multi-angle fused CPR images. The proposed ResNet50-FPN network performed feature extraction, and CBAM was added after the first and last convolutional layers to enhance the network’s attention to plaque features. Our experimental results demonstrate that deep learning neural networks can provide a feasible approach to automatically detecting coronary artery plaques. Our method achieved a plaque detection accuracy of 94.6%, outperforming the RCNN method proposed by Zriek.

Removing nonrigid refractive distortions for underwater images using an attention-based deep neural network

Refractive distortions in underwater images usually occur when these images are captured through a dynamic refractive water surface, such as unmanned aerial vehicles capturing shallow underwater scenes from the surface of water or autonomous underwater vehicles observing floating platforms in the air. We propose an end-to-end deep neural network for learning to restore real scene images for removing refractive distortions. This network adopts an encoder-decoder architecture with a specially designed attention module. The use of the attention image and the distortion field generated by the proposed deep neural network can restore the exact distorted areas in more detail. Qualitative and quantitative experimental results show that the proposed framework effectively eliminates refractive distortions and refines image details. We also test the proposed framework in practical applications by embedding it into an NVIDIA JETSON TX2 platform, and the results demonstrate the practical value of the proposed framework.

Detection of bone cancer based on a four-phase framework generative deep belief neural network in deep learning

The medical image processing plays an important role in analyzing and detecting tumors and detecting cancer cells in their early stages. Microscopic images were used as a deep learning approach to recognize bone features. These images were acquired using microscopy, a time-consuming and labor-intensive process that required repeating. Cancer cells cannot be identified by conventional techniques because of the noise in the images. To address these challenges, this paper introduces the Gated Convolutional Neural Network (GCNN) techniques used to identify multiclass bone cancer, including non-tumor, non-viable, and viable tumors. As a first stage, the most basic part of image Preprocessing is to deposing without interrupting the diagnostics information during the removal of noise. The second stage is edge detection, used for image brightness discontinuity detection based on Scaled Region Edge Detector (SRED). It is used for image segmentation and data extraction within regions based on the Screen Cluster Area Segmentation (SCAS) to segment the Region of Interest (RoI) based on Relative Scalar Color Histogram (RSCH). The fourth step is Feature extraction is to find out the features and match the image feature from the Bone Cancer Detection object based on the maximum threshold weights using GCNN with Maximum Standard External Regions (MSER) used for features extracting applied to improve feature quality analysis and Feature support rate. This results in higher classification accuracy, Precision, recall under F-measure, and lower error rates. This enables a more achieved classification than other existing methods.

Deep learning neural networks for monitoring early-age concrete strength through a surface-bonded PZT sensor configuration

Monitoring the immediate evolution of concrete’s strength is essential to ensure structural integrity and construction efficiency, requiring Forecasting to avoid unforeseen and severe failures during construction. The present investigation introduces an Equivalent structural parametric (ESP) study using a surface-bonded piezo sensor and electromechanical impedance (EMI) methodology to monitor and forecast the strength of concrete by implementing machine learning and deep learning methods. The concrete hydration processes are simulated using COMSOL™ 5.5. The concrete cube’s hydration is represented by altering Young’s modulus and damping ratio to show the rate of curing. Concrete strength development is examined in terms of conductance resonant frequency (CRF) and Conductance resonant peak (CRP). Continuous conductance signature monitoring and data analysis show that CRF and CRP increase with compressive strength. The system’s mechanical impedance is measured, and EMI signatures vs. frequency plots within a specific frequency range are compared to healthy impedance graphs. A mass-spring-damper system with identical properties is identified, and relevant structural parameters are computed. Modern ML algorithms include linear regression (LR), interaction LR, fine, medium, and coarse Gaussian SVM, etc. reliably predict strength with an error rate of less than 2%. Convolutional neural networks (CNN) have advanced image-based recognition, but their usage in EMI-based structural strength assessment is still being studied. A unique strategy using 2D CNN, 2D CNN– Long short-term memory (LSTM), and 2D CNN Bidirectional-LSTM to forecast concrete structure compressive strength shows the potential of deep learning. The proposed 2D CNN-Bi-LSTM model excels in compressive strength prediction, obtaining an R2 value of 0.99 and stabilizing loss error over a few epochs.