Proposed Network Model Research Articles

Multi-Scale Ship Detection Algorithm Based on a Lightweight Neural Network for Spaceborne SAR Images

The current limited spaceborne hardware resources and the diversity of ship target scales in SAR images have led to the requirement of on-orbit real-time detection of ship targets in spaceborne synthetic aperture radar (SAR) images. In this paper, we propose a lightweight ship detection network based on the YOLOv4-LITE model. In order to facilitate the network migration to the satellite, the method uses MobileNetv2 as the backbone feature extraction network of the model. To solve the problem of ship target scale diversity in SAR images, an improved receptive field block (RFB) structure is introduced, enhancing the feature extraction ability of the network, and improving the accuracy of multi-scale ship target detection. A sliding window block method is designed to detect the whole SAR image, which can solve the problem of image input. Experiments on the SAR ship dataset SSDD show that the detection speed of the improved lightweight network could reach up to 47.16 FPS, with the mean average precision (mAP) of 95.03%, and the model size is only 49.34 M, which demonstrates that the proposed network can accurately and quickly detect ship targets. The proposed network model can provide a reference for constructing a spaceborne real-time lightweight ship detection network, which can balance the detection accuracy and speed of the network.

A cad system design based on HybridMultiscale convolutional Mantaray network for pneumonia diagnosis.

Pneumonia is one of the diseases that people may encounter in any period of their lives. Recently, researches and developers all around the world are focussing on deep learning and image processing strategies to quicken the pneumonia diagnosis as those strategies are capable of processing numerous X-ray and computed tomography (CT) images. Clinicians need more time and appropriate experiences for making a diagnosis. Hence, a precise, reckless, and less expensive tool to detect pneumonia is necessary. Thus, this research focuses on classifying the pneumonia chest X-ray images by proposing a very efficient stacked approach to improve the image quality and hybridmultiscale convolutional mantaray feature extraction network model with high accuracy. The input dataset is restructured with the sake of a hybrid fuzzy colored and stacking approach. Then the deep feature extraction stage is processed with the aid of stacking dataset by hybrid multiscale feature extraction unit to extract multiple features. Also, the features and network size are diminished by the self-attention module (SAM) based convolutional neural network (CNN). In addition to this, the error in the proposed network model will get reduced with the aid of adaptivemantaray foraging optimization (AMRFO) approach. Finally, the support vector regression (SVR) is suggested to classify the presence of pneumonia. The proposed module has been compared with existing technique to prove the overall efficiency of the system. The huge collection of chest X-ray images from the kaggle dataset was emphasized to validate the proposed work. The experimental results reveal an outstanding performance of accuracy (97%), precision (95%) and f-score (96%) progressively.

Back-Propagation Neural Network Optimized by K-Fold Cross-Validation for Prediction of Torsional Strength of Reinforced Concrete Beam.

Due to the limitation of sample size in predicting the torsional strength of Reinforced Concrete (RC) beams, this paper aims to discuss the feasibility of employing a novel machine learning approach with K-fold cross-validation in a small sample range, which combines the advantages of a Genetic Algorithm (GA) and a Neural Network (NN) to predict the torsional strength of RC beams. This research study not only utilizes the application of a Back Propagation (BP) neural network and the Gene Algorithm-Back Propagation (GA-BP) neural network in the prediction of the torsional strength of the RC beam, but it also investigates neural network parameter optimization, including connection weights and thresholds, using K-fold cross-validation. The root mean square error (RMSE), mean absolute error (MAE), mean square error (MSE), mean absolute percentage error (MAPE) and correlation coefficient (R2) are among the evaluation metrics used to assess the performance of the trained model. To elaborate on the superiority of the proposed network models in predicting the torsional strength of RC beams, a parametric study is conducted by comparing the proposed model to three commonly used empirical formulae from existing design codes. The comparative findings of this research study demonstrate that the performance of the BP neural network is highly similar to that of design codes; however, its accuracy is inadequate. After improving the weights and thresholds by k-fold cross-validation and GA, the prediction of the BP neural network shows higher consistency with the actual measured values. The outcome of this study can be used as a theoretical reference for the optimal design of RC beams in practical applications.

A fully convolutional network (FCN) based automated ischemic stroke segment method using chemical exchange saturation transfer imaging.

Chemical exchange saturation transfer (CEST) MRI is a promising imaging modality in ischemic stroke detection due to its sensitivity in sensing postischemic pH alteration. However, the accurate segmentation of pH-altered regions remains difficult due to the complicated sources in water signal changes of CEST MRI. Meanwhile, manual localization and quantification of stroke lesions are laborious and time-consuming, which cannot meet the urgent need for timely therapeutic interventions. The goal of this study was to develop an automatic lesion segmentation approach of the ischemic region based on CEST MR images. A novel segmentation framework based on the fully convolutional neural network was investigated in our task. Z-spectra from 10 rats were manually labeled as ground truth and split into two datasets, where the training dataset including 3 rats was used to generate a segmentation model, and the remaining rats were used as test datasets to evaluate the model's performance. Then a 1D fully convolutional neural network equipped with bottleneck structures was set up, and a Grad-CAM approach was used to produce a coarse localization map, which can reflect the relevancy to the "ischemia" class of each pixel. As compared with the ground truth, the proposed network model achieved satisfying segmentation results with high values of evaluation metrics including specificity (SPE), sensitivity (SEN), accuracy (ACC), and Dice similarity coefficient (DSC), especially in some intractable situations where conventional MRI modalities and CEST quantitative method failed to distinguish between ischemic and normal tissues; also the model with augmentation was robust to input perturbations. The Grad-CAM maps performed clear tissue change distributions and interpreted the segmentations, showed a strong correlation with the quantitative method, and gave extended thinking to the function of networks. The proposed method can segment ischemia region from CEST images, with the Grad-CAM maps giving access to interpretative information about the segmentations, which demonstrates great potential in clinical routines.

Application of Generalized Frequency Response Functions and Improved Convolutional Neural Network to Fault Diagnosis of Heavy-duty Industrial Robot

The combination of nonlinear spectrum and convolutional neural network (CNN) is efficient for fault diagnosis of nonlinear system . However, in traditional method, the nonlinear spectrum calculation was accomplished by identification algorithm outside the CNN, which reduced the diagnosis efficiency. To solve this problem, a novel CNN with the function of spectrum calculation and fault diagnosis is designed, in which the spectrum calculation network and the fault diagnosis network are connected in series. By extracting the optimized parameters of network, the nonlinear spectrum based on generalized frequency response function (GFRF) is obtained in the former network. Then, the GFRF spectrum is automatically put into the latter network for feature extraction and diagnosis. Hence, after determining the structure of the CNN, only by system input and output, the fault diagnosis can be realized, which avoids the complex process in traditional method. What's more, a new error cost function model is designed to guide the network parameters optimization in the direction of feature classification , which is conductive to improve the diagnosis accuracy. The proposed network model is applied to the heavy-duty industrial robot system, and the best performance is demonstrated by several experiments.

A Bayesian Regularization Approach to Predict the Quality of Injection-Moulded Components by statistical SVM for Online Monitoring system

To evaluate the quality of injection-molded components, conventional approaches are costly, time-consuming, or based on statistical process control characteristics that are not always accurate. Machine learning might be used to categorise components based on their quality. In order to accurately estimate the quality of injection moulded components, this study uses a SVM classifier. In addition, the form of the spare components after the working method product in simulation is classified as "qualified" or "unqualified". The quality indicators have an excellent association with data recordings from the original database of various sensors such as pressure and temperature used in the proposed network model for online prediction. The outliers are removed from the input original data to minimize the deviation of precision or prediction accuracy of the model performance metrics. Data points in the "to-be-confirmed" region (which is in the fit line area) may be misjudged by this statistical SVM model since it is placed between the "qualified" and "unqualified" areas. This statistical procedure in the proposed SVM model also uses Bayesian regularisation to classify final components into distinct quality levels.

A deep learning method for the recognition of solar radio burst spectrum.

Solar radiation is the excitation source that affects the weather in the atmosphere of the earth, and some solar activities such as flares and coronal mass ejections are often accompanied by radio bursts. The spectrum of solar radio bursts is helpful for astronomers to explore the mechanism of radio bursts. With the development and progress of solar radio spectrum observation methods, the observation of the Sun can be done at almost all times of day. How to quickly and automatically identify the small proportion of burst data from the huge corpus of observation data has become an important research direction. The innovation of this study is to enhance the original radio spectrum dataset with unbalanced sample distribution, and a neural network model for solar radio spectrum image classification is proposed on this basis. This hybrid structure of joint convolution and a memory unit overcomes the shortcoming of the traditional convolution or memory model, which can only extract one-sided features of an image. By extracting the frequency structure features and time-series features at the same time, the sensitivity to the small features of the spectrum image can be enhanced. Based on the data of the Solar Broadband Radio Spectrometer (SBRS) in China, the proposed network model can improve the average classification accuracy of the spectrum image to 98.73%, which will be helpful for related astronomical research.

Intelligent Detection of Steel Defects Based on Improved Split Attention Networks.

The intelligent monitoring and diagnosis of steel defects plays an important role in improving steel quality, production efficiency, and associated smart manufacturing. The application of the bio-inspired algorithms to mechanical engineering problems is of great significance. The split attention network is an improvement of the residual network, and it is an improvement of the visual attention mechanism in the bionic algorithm. In this paper, based on the feature pyramid network and split attention network, the network is improved and optimised in terms of data enhancement, multi-scale feature fusion and network structure optimisation. The DF-ResNeSt50 network model is proposed, which introduces a simple modularized split attention block, which can improve the attention mechanism of cross-feature graph groups. Finally, experimental validation proves that the proposed network model has good performance and application prospects in the intelligent detection of steel defects.

Novel large scale brain network models for EEG epileptic pattern generations

BackgroundUnlike normal EEG patterns, the epileptiform abnormal pattern is characterized by different morphologies such as the high-frequency oscillations (HFOs) of ripples on spikes, spikes and waves, continuous and sporadic spikes, and ploy2 spikes. Several studies have reported that HFOs can be novel biomarkers in human epilepsy study. MethodTo regenerate and investigate these patterns, we have proposed three large scale brain network models (BNMS) by linking the neural mass model (NMM) of Stefanescu-Jirsa 2D (S-J 2D) with our own structural connectivity derived from the realistic biological data, so called, large-scale connectivity connectome. These models include multiple network connectivity of brain regions at different lobes from both hemispheres (left and right). The network nodes of these models were simulated based on the local dynamics of the S-J 2D model, which were generated by adjusting the global coupling between the excitatory and inhibitory populations. The connection strength between the inhibitory and excitatory neurons of the local model was also adjusted to investigate different morphology patterns. ResultsThe proposed network models were developed and evaluated by simulations. Different abnormal patterns of EEG brain activities such as HFOS ripples on spikes, spikes, continuous spikes, sporadic spikes and ploy2 spikes ranging from 94 to 144 Hz were regenerated. Different morphology patterns of abnormality were generated from novel BNMs and the epileptiform abnormal pattern obtained in actual EEG and other computational models were also compared. SignificantThis study is able to assist researchers and clinical doctors in the field of epilepsy to better understand the complex neural mechanisms behind the abnormal oscillatory activities, which may lead to the discovery of new clinical interventions in epilepsy.

Secrecy Capacity Analysis of Reconfigurable Intelligent Surface Based Vehicular Networks

As Vehicular Networks based technologies are in the close proximity of deployment for various wireless applications under proposal worldwide, this research paper proposes secrecy capacity analysis for Reconfigurable Intelligent Surface (RIS) based Vehicular Network. The proposed network model has a fixed infrastructure comprising of source node, destination node incorporated with single antenna and passive eavesdropper forming the scenario. RIS based Vehicular communication links are modelled by Rayleigh fading for source-to RIS link and RIS to destination Vehicle, whereas Eavesdropper channel links are Double-Rayleigh amplitude distribution, induced by double scattering in the channel. For this scenario, we derive the closed-form expressions for the average Secrecy Capacity and Secrecy Outage Probability (SOP) of the considered system. Though, Secrecy Capacity analysis is an excellent performance metric for assessing eavesdropper based system, it has been reported by various research works, this research paper differentiates from other research papers by considering different secrecy rates and different distances of eavesdropper as presented in simulation. Further, to validate the obtained simulation results, theoretical results are also derived for assessing performance of SOP for various secrecy rates which is the highlight of this research paper and it can be used as benchmark for various research works to proceed further.

Sparse DNN Model for Frequency Expanding of Higher Order Ambisonics Encoding Process

The performance of higherorder Ambisonics (HOA) signals obtained using spherical harmonics decomposition method is disturbed by two primary sources of errors, the noise pollution in low-frequency band and the spatial aliasing in high-frequency band. Inspired by the HOA signals upscale method, which is performed using the sparse character of the sound field, this paper propose a sound field decomposition model based on a sparse deep neural network that offers HOA signals with wider frequency bandwidth. We use the frequency domain multi-scale convolutional network to realize the spherical harmonics decomposition, as well as learning the spatial aliasing pattern, based on which the aliasing-free HOA signals can be derived. Besides, we apply a sparse encoding network to cpature the sparse feature of the sound field which will improve the model performance when the sparse condition is satisfied. The experiments results prove that the proposed model can obtain HOA signals with wider frequency range of operation under multiple sources (up to 10 sources) and low reverberant environments (<inline-formula><tex-math notation="LaTeX">$T_{60}\le$</tex-math></inline-formula> 400 ms). When the sparsity feature cannot be satisfied (<inline-formula><tex-math notation="LaTeX">$T_{60} =$</tex-math></inline-formula> 800 ms), the proposed network model still maintain the same performance as the traditional methods.

Super-Resolution and Large Depth of Field Model for Optical Microscope Imaging

Due to the limitation of numerical aperture (NA) in a microscope, it is very difficult to obtain a clear image of the specimen with a large depth of field (DOF). We propose a deep learning network model to simultaneously improve the imaging resolution and DOF of optical microscopes. The proposed M-Deblurgan consists of three parts: (i) a deblurring module equipped with an encoder-decoder network for feature extraction, (ii) an optimal approximation module to reduce the error propagation between the two tasks, and (iii) an SR module to super-resolve the image from the output of the optimal approximation module. The experimental results show that the proposed network model reaches the optimal result. The peak signal-to-noise ratio (PSNR) of the method can reach 37.5326, and the structural similarity (SSIM) can reach 0.9551 in the experimental dataset. The method can also be used in other potential applications, such as microscopes, mobile cameras, and telescopes.

A Domain Adaptive Person Re-Identification Based on Dual Attention Mechanism and Camstyle Transfer

Due to the variation in the image capturing process, the difference between source and target sets causes a challenge in unsupervised domain adaptation (UDA) on person re-identification (re-ID). Given a labeled source training set and an unlabeled target training set, this paper focuses on improving the generalization ability of the re-ID model on the target testing set. The proposed method enforces two properties at the same time: (1) camera invariance is achieved through the positive learning formed by unlabeled target images and their camera style transfer counterparts; and (2) the robustness of the backbone network feature extraction is improved, and the accuracy of feature extraction is enhanced by adding a position-channel dual attention mechanism. The proposed network model uses a classic dual-stream network. Comparative experimental results on three public benchmarks prove the superiority of the proposed method.

Classification and Inverse Design of Metasurface Absorber in Visible Band

AbstractMetasurface absorber (MA) has been a research hotspot in the field of artificial electromagnetic structural material due to its dual advantages of high performance and compact design. Usually, the design of MA depends on the designer's professional knowledge, experience and physical inspiration. The desired optical response can be obtained by using electromagnetic simulation software to carry out hundreds or thousands of numerical calculations. Thus, it is still a challenge to quickly retrieve the optimal structure according to the desired optical response and realize the on‐demand inverse design. Besides, limited by the inner physics of the MA, it is not always possible to find the structural parameters corresponding to the desired spectrum. This paper not only takes the planar geometry and thickness of the structures into account but also realizes the probability classification of the desired spectra through the classification network. According to the classification results, the prediction network of the corresponding is selected to realize the on‐demand inverse design of MA. The proposed network model can design MA rapidly and accurately in a data‐driven way and can be flexibly applied to the design of other data‐enabled photonic devices, which is promising to become a comprehensive and effective design tool.

Semantic Sequence Labeling Model of Power Dispatching Based on Deep Long Short Term Memory Network

As the decision-making brain for power system operation, grid regulation and operation is a comprehensive decision-making control that combines a large amount of data, mechanism analysis, operating procedures and professional experience, and a new generation of artificial intelligence development ideas and evolution characterized by data-driven and knowledge-guided. The directions are very close. However, the current scheduling control is still based on experience and manual analysis. The massive and diverse data of the control center and the lack of logical models between the plans require a large amount of experience and knowledge associations by the control personnel. There are more repetitive human brain labor and relatively low intelligence. Therefore, deep learning is applied to the learning of power control knowledge, and a semantic understanding network based on deep Long Short Term Memory is proposed. It uses sequence labeling to extract in-depth semantic related information of different keywords and query questions, and finds key information about language problems in order to achieve fine-grained and precise query. Experiments show that the proposed network model is superior to the previous methods, and it achieves better performance in the joint extraction of fine-grained evaluation words and evaluation objects, extracts the key information and deep semantic information of query problems and corresponding cases, and realizes power scheduling based on voice interaction The model can be effectively applied in the field of power dispatching and solve a large number of problems in power dispatching and control.

Geometry and boundary condition adaptive data-driven model of fluid flow based on deep convolutional neural networks

We develop a deep neural network-based reduced-order model (ROM) for rapid prediction of the steady-state velocity field with arbitrary geometry and various boundary conditions. The input matrix of the network is composed of the nearest wall signed distance function (NWSDF), which contains more physical information than the signed distance function (SDF) and binary map; the boundary conditions are represented by specifically designed values and fused with NWSDF. The network architecture comprises convolutional and transpose-convolutional layers, and convolutional layers are employed to encode and extract the physical information from NWSDF. The highly encoded information is decoded by transpose-convolutional layers to estimate the velocity fields. Furthermore, we introduce a pooling layer to innovatively emphasize/preserve information of boundary conditions, which are gradually flooded by other features during the convolutional operation. The network model is trained using several simple geometries and tested with more complex cases. The proposed network model shows excellent adaptability to arbitrary complex geometry and variable boundary conditions. The average prediction error of the network model on the testing dataset is less than 6%, and the prediction speed is two orders faster than that of the numerical simulation. In contrast to the current model, the average error of the network model with the input matrix of the binary map, traditional SDF, and model without pooling layers is around 12%, 11%, and 11%, respectively. The outstanding performance of the proposed network model indicates the potential of the deep neural network-based ROM for real-time control and rapid optimization, while encouraging further investigation to achieve practical application.

Deep learning approach for automatic segmentation of ulna and radius in dual-energy X-ray imaging

BackgroundSegmentation of the ulna and radius is a crucial step for the measurement of bone mineral density (BMD) in dual-energy X-ray imaging in patients suspected of having osteoporosis.PurposeThis work aimed to propose a deep learning approach for the accurate automatic segmentation of the ulna and radius in dual-energy X-ray imaging.Methods and materialsWe developed a deep learning model with residual block (Resblock) for the segmentation of the ulna and radius. Three hundred and sixty subjects were included in the study, and five-fold cross-validation was used to evaluate the performance of the proposed network. The Dice coefficient and Jaccard index were calculated to evaluate the results of segmentation in this study.ResultsThe proposed network model had a better segmentation performance than the previous deep learning-based methods with respect to the automatic segmentation of the ulna and radius. The evaluation results suggested that the average Dice coefficients of the ulna and radius were 0.9835 and 0.9874, with average Jaccard indexes of 0.9680 and 0.9751, respectively.ConclusionThe deep learning-based method developed in this study improved the segmentation performance of the ulna and radius in dual-energy X-ray imaging.

Data-Driven Modeling of Geometry-Adaptive Steady Heat Convection Based on Convolutional Neural Networks

Heat convection is one of the main mechanisms of heat transfer, and it involves both heat conduction and heat transportation by fluid flow; as a result, it usually requires numerical simulation for solving heat convection problems. Although the derivation of governing equations is not difficult, the solution process can be complicated and usually requires numerical discretization and iteration of differential equations. In this paper, based on neural networks, we developed a data-driven model for an extremely fast prediction of steady-state heat convection of a hot object with an arbitrary complex geometry in a two-dimensional space. According to the governing equations, the steady-state heat convection is dominated by convection and thermal diffusion terms; thus the distribution of the physical fields would exhibit stronger correlations between adjacent points. Therefore, the proposed neural network model uses convolutional neural network (CNN) layers as the encoder and deconvolutional neural network (DCNN) layers as the decoder. Compared with a fully connected (FC) network model, the CNN-based model is good for capturing and reconstructing the spatial relationships of low-rank feature spaces, such as edge intersections, parallelism, and symmetry. Furthermore, we applied the signed distance function (SDF) as the network input for representing the problem geometry, which contains more information compared with a binary image. For displaying the strong learning and generalization ability of the proposed network model, the training dataset only contains hot objects with simple geometries: triangles, quadrilaterals, pentagons, hexagons, and dodecagons, while the testing cases use arbitrary and complex geometries. According to the study, the trained network model can accurately predict the velocity and temperature field of the problems with complex geometries, which has never been seen by the network model during the model training; and the prediction speed is two orders faster than the CFD. The ability of accurate and extremely fast prediction of the network model suggests the potential of applying reduced-order network models to the applications of real-time control and fast optimization in the future.

A 3D macro-segment network model for vanadium redox flow battery with serpentine flow field

This paper presents a 3D macro-segment network model for a vanadium redox flow battery with serpentine flow field. The proposed network model is coupled of electrolyte flow module, species transfer module, charge transfer module. In flow resistance network module, the characteristics of electrolyte flow in the serpentine flow channel and under-rib convection in the porous electrode are all considered. In addition, the electrode intrusion and the electrode non-uniform structure caused by compression are taken into account. In species transfer network module, the convection, diffusion and migration between adjacent segments and reversible electrochemical reactions and self-discharge reactions inside the segment are all analyzed in the ion conservation equation. In charge transfer network module, each battery segment, which is composed of the channel, electrode and membrane segment in series, is connected in parallel and has the same output voltage. In this paper, the battery performance including charge-discharge voltages and cell pressure drop under different electrode compression ratios are validated. In the following, the battery performance under single and multiple charge-discharge cycle are investigated using the proposed network model. Besides, the field distribution of key parameters in terms of velocity, pressure, ions concentration and current density are analyzed and validated with finite element method model data. The proposed 3D macro-segment network model is not only able to effectively consider the distribution difference inside the battery caused by the flow field, but also capable of reducing the computational resources, which renders the network model is suitable for the fast prediction of battery performance.

ECG Signal-Enabled Automatic Diagnosis Technology of Heart Failure.

Usually, heart failure occurs when heart-related diseases are developed and continue to deteriorate veins and arteries. Heart failure is the final stage of heart disease, and it has become an important medical problem, particularly among the aging population. In medical diagnosis and treatment, the examination of heart failure contains various indicators such as electrocardiogram. It is one of the relatively common ways to collect heart failure or attack related information and is also used as a reference indicator for doctors. Electrocardiogram indicates the potential activity of patient's heart and directly reflects the changes in it. In this paper, a deep learning-based diagnosis system is presented for the early detection of heart failure particularly in elderly patients. For this purpose, we have used two datasets, Physio-Bank and MIMIC-III, which are publicly available, to extract ECG signals and thoroughly examine heart failure. Initially, a heart failure diagnosis model which is based on attention convolutional neural network (CBAM-CNN) is proposed to automatically extract features. Additionally, attention module adaptively learns the characteristics of local features and efficiently extracts the complex features of the ECG signal to perform classification diagnosis. To verify the exceptional performance of the proposed network model, various experiments were carried out in the realistic environment of hospitals. Influence of signal preprocessing on the performance of model is also discussed. These results show that the proposed CBAM-CNN model performance is better for both classifications of ECG signals. Likewise, the CBAM-CNN model is sensitive to noise, and its accuracy is effectively improved as soon as signal is refined.