Network Model Research Articles

Leakage detection of an acoustic emission pipeline based on an improved transformer network

Abstract Pipeline leakage detection is an integral part of pipeline integrity management. Combining AE (Acoustic Emission) with deep learning is currently the most commonly used method for pipeline leakage detection. However, this approach is usually applicable only to specific situations and requires powerful signal analysis and computational capabilities. To address these issues, this paper proposes an improved Transformer network model for diagnosing faults associated with abnormal working conditions in acoustic emission pipelines. First, the method utilizes the temporal properties of the GRU and the positional coding of the Transformer to capture and feature extract the data point sequence position information to suppress redundant information, and introduces the largest pooling layer into the Transformer model to alleviate the overfitting phenomenon. Second, while retaining the original attention learning mechanism and identity path in the original DRSN, a new soft threshold function is introduced to replace the ReLU activation function with a new threshold function, and a new soft threshold module and adaptive slope module are designed to construct the improved residual shrinkage unit (ASB-STRSBU), which is used to adaptively set the optimal threshold. Finally, pipeline leakage is classified. The experimental results show that the NDRSN model is able to make full use of global and local information when considering leakage signals and can automatically learn and acquire the important parameters of the input features in the spatial and channel domains. By optimizing the GRU improved Transformer network recognition model, the method significantly reduces the model training time and computational resource consumption while maintaining high leakage recognition accuracy. The average accuracy reached 93.97%. This indicates that the method has good robustness in acoustic emission pipeline leakage detection.

Enhanced bi-level optimal scheduling strategy for distribution network with multi-microgrids considering source-load uncertainties

With the increasing integrations of renewable energy resources into distribution networks (DNs) and microgrids (MGs), the imperative for an effective market scheduling mechanism becomes paramount to enhance the operational safety, reliability, and economic efficiency of distribution grids. Taking advantage of bi-level programming theory, this study meticulously formulates a comprehensive optimization scheduling model for the multi-MGs distribution network. The upper-level optimization objective is to minimize both the operational losses and total costs of the DN. Concurrently, the lower-level optimization pursues the maximization of daily operational revenue for MGs. Recognizing the pervasive impact of the inherent uncertainty associated with renewable energy sources on system safety and reliability, a cutting-edge scenario-based stochastic planning framework is introduced. The methodology integrates a heuristic matrix matching approach to effectively handle the intricate challenges posed by uncertainties from wind and photovoltaic generations. Moreover, in addressing the proposed nonlinear models, a sophisticated method is employed, utilizing the second-order cone relaxation and linearization methods. These methods meticulously transform the upper and lower-level models into second-order cone planning and mixed-integer linear programming issues, respectively. Finally, the proposed methodologies are rigorously scrutinized and validated with intricate case studies, providing a nuanced understanding of their efficacy. The empirical results underscore the theoretical feasibility and superiority of the proposed scheduling scheme. Notably, the operational performance of the DN as well as the economic viability of multiple MGs can also be significantly improved.

Time division multiplexing based multi-spectral semantic camera for LiDAR applications

The recent progress in the development of measurement systems for autonomous recognition had a substantial impact on emerging technology in numerous fields, especially robotics and automotive applications. In particular, time-of-flight (TOF) based light detection and ranging (LiDAR) systems enable to map the surrounding environmental information over long distances and with high accuracy. The combination of advanced LiDAR with an artificial intelligence platform allows enhanced object recognition and classification, which however still suffers from limitations of inaccuracy and misidentification. Recently, multi-spectral LiDAR systems have been employed to increase the object recognition performance by additionally providing material information in the short-wave infrared (SWIR) range where the reflection spectrum characteristics are typically very sensitive to material properties. However, previous multi-spectral LiDAR systems utilized band-pass filters or complex dispersive optical systems and even required multiple photodetectors, adding complexity and cost. In this work, we propose a time-division-multiplexing (TDM) based multi-spectral LiDAR system for semantic object inference by the simultaneous acquisition of spatial and spectral information. By utilizing the TDM method, we enable the simultaneous acquisition of spatial and spectral information as well as a TOF based distance map with minimized optical loss using only a single photodetector. Our LiDAR system utilizes nanosecond pulses of five different wavelengths in the SWIR range to acquire sufficient material information in addition to 3D spatial information. To demonstrate the recognition performance, we map the multi-spectral image from a human hand, a mannequin hand, a fabric gloved hand, a nitrile gloved hand, and a printed human hand onto an RGB-color encoded image, which clearly visualizes spectral differences as RGB color depending on the material while having a similar shape. Additionally, the classification performance of the multi-spectral image is demonstrated with a convolution neural network (CNN) model using the full multi-spectral data set. Our work presents a compact novel spectroscopic LiDAR system, which provides increased recognition performance and thus a great potential to improve safety and reliability in autonomous driving.

Improving the Concrete Crack Detection Process via a Hybrid Visual Transformer Algorithm

Inspections of concrete bridges across the United States represent a significant commitment of resources, given their biannual mandate for many structures. With a notable number of aging bridges, there is an imperative need to enhance the efficiency of these inspections. This study harnessed the power of computer vision to streamline the inspection process. Our experiment examined the efficacy of a state-of-the-art Visual Transformer (ViT) model combined with distinct image enhancement detector algorithms. We benchmarked against a deep learning Convolutional Neural Network (CNN) model. These models were applied to over 20,000 high-quality images from the Concrete Images for Classification dataset. Traditional crack detection methods often fall short due to their heavy reliance on time and resources. This research pioneers bridge inspection by integrating ViT with diverse image enhancement detectors, significantly improving concrete crack detection accuracy. Notably, a custom-built CNN achieves over 99% accuracy with substantially lower training time than ViT, making it an efficient solution for enhancing safety and resource conservation in infrastructure management. These advancements enhance safety by enabling reliable detection and timely maintenance, but they also align with Industry 4.0 objectives, automating manual inspections, reducing costs, and advancing technological integration in public infrastructure management.

Synthetic aperture radar automatic target recognition based on cost‐sensitive awareness generative adversarial network for imbalanced data

AbstractIn military contexts, synthetic aperture radar (SAR) automatic target recognition (ATR) models frequently encounter the challenge of imbalanced data, resulting in a noticeable degradation in the recognition performance. Therefore, the authors propose a cost‐sensitive awareness generative adversarial network (CAGAN) model, aiming to improve the robustness of ATR models under imbalanced data. Firstly, the authors introduce a convolutional neural network (DCNN) to extract features. Then, the synthetic minority over‐sampling technique (SMOTE) is applied to achieve feature‐level balancing for the minority category. Finally, a CAGAN model is designed to perform the final classification task. In this process, the GAN‐based adversarial training mechanism enriches the diversity of training samples, making the ATR model more comprehensive in understanding different categories. In addition, the cost matrix increases the penalty for misclassification results and further improves the classification accuracy. Simultaneously, the cost‐sensitive awareness can accurately adjust the cost matrix through training data, thus reducing dependence on expert knowledge and improving the generalisation performance of the ATR model. This model is an end‐to‐end ATR scheme, which has been experimentally validated on the MSTAR and OpenSARship datasets. Compared to other methods, the proposed method exhibits strong robustness in dealing with various imbalanced scenarios and significant generalisation capability across different datasets.

Basketball Sports Posture Recognition Technology Based on Improved Graph Convolutional Neural Network

Basketball has rapidly developed in recent years. Analysis of various moves in basketball can provide technical references for professional players and assist referees in judging games. Traditional technology can no longer provide modern basketball players with theoretical support. Therefore, using intelligent methods to recognize human body postures in basketball was a relatively innovative approach. To be able to recognize the basketball sports posture of players more accurately, the experiment proposes a basketball stance recognition model based on enhanced graph convolutional networks (GCN), that is, the basketball stance recognition model based on enhanced GCN and spatial temporal graph convolutional network (ST-GCN) model. This model combines the respective advantages of the GCN and temporal convolutional network and can handle graph-structured data with time-series relationships. The ST-GCN can be further deduced by realizing the convolution operation of the graph structure and establishing a spatiotemporal graph convolution model for the posture sequence of a person’s body. A dataset of technical basketball actions is constructed to verify the effectiveness of the ST-GCN model. The final experimental findings indicated that the final recognition accuracy of the ST-GCN model for basketball postures was approximately 95.58%, whereas the final recognition accuracy of the long short term memory + multiview re-observation skeleton action recognition (LSTM+MV+AC) model was about 93.65%.

ANN model of a triple-junction cell in concentrating photovoltaic system

Significant advancements in concentrating photovoltaic (CPV) systems have been achieved in recent years, also thanks to the definition of calculation methods of their energy performances in several operation conditions. Typically, the CPV systems electrical power is separately calculated or in terms of its temperature or concentration factor (C), but not simultaneously in terms of both variables. In this paper, an Artificial Neural Network model based on experimental data, linking electric power of CPV system with Direct Normal Irradiance and Triple-Junction cell temperature for different C values, is developed. Moreover, the model is also adopted to realize a feasibility analysis of point-focus CPV system used for different users: residential building and agricultural livestock farm. The optimal number of modules is determined to maximize the Net Present Value (NPV) of the investment. For the residential user, an optimal configuration of CPV system includes 16 modules, providing a peak power of 3.1 kW and covering an area of 130 m2. This configuration allows the maximization of NPV value, reaching 15.9 k€, with DPB of 9.8 years. As for the agricultural livestock, 36 modules, with peak power of 7.0 kW and covering an area of 292 m2, allow the maximization of NPV value equal to 16.3 k€, with DPB of 10.2 years.

Predicting Legal Systems: An Artificial Neural Network Approach with Statistical Analysis for Comparative Study of Civil Law and Common Law Countries

Abstract This study compares countries with common law with countries with civil law systems and investigates the possibility of predicting legal systems using artificial neural networks (ANNs). The OLS model, ANOVA, ANN, and Tensor Flow are used in the research to analyze the data. The goal is to find out how board characteristics and country legislative frameworks affect how European corporations disclose their social performance. The performance of a hidden layer with five nodes is best, according to the ANN model. The model’s accuracy throughout testing and validation is 0.750. The confusion matrix shows that, of the four observations in the test set, three were correctly categorized as “Civil law” and one was incorrectly categorized as “Common law.” To evaluate the model’s efficacy, evaluation metrics are computed. The model’s accuracy is 0.750, which represents a prediction success rate of 75%. For the “Civil law” class, the recall (true positive rate) is 1.0, indicating that all “Civil law” cases are correctly identified. Metrics for the “Common law” class, however, are not available due to the scant amount of data that is available. The prevalence of countries with common law and civil law systems is compared in the ANOVA analysis. As shown by the computed F-value of 0.482, there is less variance inside each legal system than there is between the two. There is no statistically significant difference in frequency between the two legal systems, according to the p-value of 0.495. Overall, the research’s conclusions imply that social performance disclosure between countries with common law and civil law systems differs only slightly. The neural network model’s network weights provide insight into the importance of different features in prediction.

The frustum network model based on clique extension

The frustum network model based on clique extension

The Effect of Hot Forming–Quenching and Heat Treatment Processes on the Mechanical Properties of AA6016 Aluminum Alloy Sheets

This study explored the impact of Hot Forming–Quenching (HFQ) and heat treatment processes on the mechanical properties of AA6016 sheets. The experimental findings demonstrated that at high-temperature pre-straining (HT-PS) of 15%, the strength performance of the AA6016 sheet exhibited enhancement, with a progressive increase in both the heat treatment temperature and duration. Conversely, under HT-PS conditions of 3% and 7%, the heat treatment process exhibited a relatively modest impact on the mechanical properties of the AA6016 sheet. Differential scanning calorimetry (DSC) was employed to understand the influence of different process conditions on the precipitated phases. By comparing the precipitation peaks of the β″ phase at HT-PS of 3% and 15%, it was observed that the precipitation peak of the β″ phase decreased with an increase in HT-PS. This indicated that HT-PS promoted the precipitation of the β″ phase. In order to forecast the mechanical performance of the AA6016 sheets after applying various pre-straining and heat treatment parameters, two models were used: a backpropagation (BP) neural network and a genetic algorithm (GA)-BP neural network. These models were evaluated for their fitting and predictive capabilities. The research findings demonstrated that the GA-BP neural network model exhibited superior fitting and predictive accuracy compared to the BP neural network model.

A Study on Speech Recognition by a Neural Network Based on English Speech Feature Parameters

In this study, from the perspective of English speech feature parameters, two feature parameters, the mel-frequency cepstral coefficient (MFCC) and filter bank (Fbank), were selected to identify English speech. The algorithms used for recognition employed the classical back-propagation neural network (BPNN), recurrent neural network (RNN), and long short-term memory (LSTM) that were obtained by improving RNN. The three recognition algorithms were compared in the experiments, and the effects of the two feature parameters on the performance of the recognition algorithms were also compared. The LSTM model had the best identification performance among the three neural networks under different experimental environments; the neural network model using the MFCC feature parameter outperformed the neural network using the Fbank feature parameter; the LSTM model had the highest correct rate and the highest speed, while the RNN model ranked second, and the BPNN model ranked worst. The results confirm that the application of the LSTM model in combination with MFCC feature parameter extraction to English speech recognition can achieve higher speech recognition accuracy compared to other neural networks.

Financialization of commodity markets: New evidence from temporal and spatial domains

AbstractTo address the ongoing contention surrounding the impact of financialization, this study adopts a ripple‐spreading network model to analyze the transmission of information across 13 globally significant commodity markets. By juxtaposing the pre‐ and postfinancialization periods, notable disparities in spillover magnitude are discerned, with overall effects registering at 58% and 85%, respectively. Moreover, the postfinancialization period exhibits accelerated spillover dynamics, necessitating a reduced timeframe (less than 1000 units) in contrast to the prefinancialization period (approximately 2000 units). Furthermore, a heightened interconnectedness among energy, metal, and agricultural futures is evident in the postfinancialization period. These findings furnish compelling evidence regarding the ramifications of financialization on commodity markets.

Early detection of glaucoma: feature visualization with a deep convolutional network

ABSTRACT Computer-aided diagnosis of ocular disorders like glaucoma can be effectively performed with retinal fundus images. Since the advent of machine learning and later deep learning techniques in medical image analysis, research focusing on automated and early detection of glaucoma has gained prominence. In this paper, we show how a deep Convolutional Neural Network model can be used in visualising the features related to Glaucoma that are consistent with (a subset of) those used by ophthalmologists for diagnosis. However, since the doctors do not necessarily depend entirely on these features but back them up with other important measurements such as visual field testing (HVF), any deep neural network such as CNN may not necessarily be expected to yield 100% accuracy. Further, other important parameters such as Intraocular Pressure (IOP), which may be built into a deep learning model, may not necessarily correlate well with the presence of Glaucoma. In this work, a deep CNN model has been trained and tested on a large number of high-quality fundus images, both normal and glaucomatous. We compare the results with transfer learning models such as VGG16, ResNet50, and MobileNetV2. On an average, we obtained an accuracy of 93.75% in identifying glaucoma, focusing only on the features of the fundus image.

Feasibility of decoding covert speech in ECoG with a Transformer trained on overt speech

Several attempts for speech brain–computer interfacing (BCI) have been made to decode phonemes, sub-words, words, or sentences using invasive measurements, such as the electrocorticogram (ECoG), during auditory speech perception, overt speech, or imagined (covert) speech. Decoding sentences from covert speech is a challenging task. Sixteen epilepsy patients with intracranially implanted electrodes participated in this study, and ECoGs were recorded during overt speech and covert speech of eight Japanese sentences, each consisting of three tokens. In particular, Transformer neural network model was applied to decode text sentences from covert speech, which was trained using ECoGs obtained during overt speech. We first examined the proposed Transformer model using the same task for training and testing, and then evaluated the model’s performance when trained with overt task for decoding covert speech. The Transformer model trained on covert speech achieved an average token error rate (TER) of 46.6% for decoding covert speech, whereas the model trained on overt speech achieved a TER of 46.3% (p>0.05;d=0.07)\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$( p > 0.05; d=0.07)$$\\end{document}. Therefore, the challenge of collecting training data for covert speech can be addressed using overt speech. The performance of covert speech can improve by employing several overt speeches.

Segmentation of ethnic clothing patterns with fusion of multiple attention mechanisms

AbstractTo address the challenges posed by diverse pattern-background elements, intricate details, and complex textures in the semantic segmentation of ethnic clothing patterns, this research introduces a novel semantic segmentation network model called MST-Unet (Mixed Swin Transformer U-net). The proposed model combines a U-shaped network structure with multiple attention mechanisms. The upper layers of the model employ classical convolutional operations, focusing on local relationships in the initial layers containing high-resolution details. In deeper layers, Swin Transformer modules are utilized, capable of efficient feature extraction with smaller spatial dimensions, maintaining performance while reducing computational burden. An attention gate mechanism is integrated into the decoder, contributing to enhanced performance in ethnic clothing pattern segmentation tasks by allowing the model to better capture crucial image features and achieve precise segmentation results. In visual comparisons of segmentation results, our proposed model demonstrates superior performance. The segmentation results exhibit more complete preservation of edge contours and fewer misclassifications in irrelevant regions within the images. In qualitative and quantitative experiments conducted on the ethnic clothing pattern dataset, our model achieves the highest Dice score for segmentation results in all four subclasses of ethnic clothing patterns. The average Dice score of our model reaches an impressive 89.80%, surpassing other algorithms in the same category. When compared to Deeplab_V3+, ResUnet, SwinUnet, and Unet networks, our model outperforms them by 7.72%, 5.09%, 5.05%, and 0.67%.

PSO-PSP-Net + InceptionV3: An optimized hyper-parameter tuned Computer-Aided Diagnostic model for liver tumor detection using CT scan slices

An automated diagnostic system leads to a supreme requirement in medical image analysis, greatly impacting the death rate due to the high spreading rate of liver tumors. However, the traditional histopathological diagnostic processes cause more disruption in the medical detection framework because of the procedure of manual estimation of CT scans and medical labor-intensive, which lead to the time consumption process. It is a highly challenging task for the segmentation of liver tumor because of the irregular appearance, distinct densities, patchy edges, and tumor sizes. To overcome this issue, an automated computer-aided diagnosis framework has been utilized to diagnose the liver cancer quickly and accurately with the less workload of expert radiologists.Presently, Deep learning-based techniques have been applied to identify the tumor regions and assist radiologists in the initial diagnosis and scrutinizing the severity level. The proposed work presents a modified pyramid scene parsing network (PSP-Net) model based on InceptionV3 deep network architecture with a particle swarm optimization algorithm, which increases the efficiency and effectiveness of the liver diagnostic model. Three publicly accessible benchmark datasets have been engaged with the pre-processing task accomplished by image normalization and data augmentation techniques. Moreover, a hyper-parameter tuned improved PSO-PSP-Net algorithm has analyzed the optimal set of gathered features. In the experimental evaluation, the proposed model outperformed than grey wolf optimization (GWO-PSP-Net), PSO-U-Net, GWO-U-Net, opposition based spotted hyena optimization (O-SHO-CRNN), and grey wolf-class topper optimization (GW-CTO) + U-Net by 8.42 %, 4.9 %, 9.54 %, 5.12 %, and 1.75 %, respectively in terms of accuracy for LiTS dataset. The developed PSO-PSP-Net model attained better accuracy than GWO-PSP-Net, PSO-U-Net, GWO-U-Net, and Residual Atrous U-Net (RA-U-Net) for the segmentation of Type-I and Type-II liver tumors by 6.88 %, 8.44 %, 13.75 %, 5.99 % and 6.12 %, 9.74 %, 17.3 %, and 5.11 %, respectively for 3DIRCAD-b1 dataset. The obtained accuracy of the developed model is 14.09 %, 7.54 %, 16.36 %, 2.64 %, and 14.9 %, 8.86 %, 17.76 %, and 4.87 % maximized than GWO-PSP-Net, PSO-U-Net, GWO-U-Net, and RA-U-Net, respectively, for the segmentation of the infected arteries and bones in the 3DIRCAD-b1 dataset. For the CHAOS dataset, the improved PSO-PSP-Net model outperformed GWO-PSP-Net, PSO-U-Net, and GWO-U-Net, respectively, in terms of overall accuracy by 7.44 %, 8.85 %, and 17.2 %, respectively. The investigational analysis validates the proposed PSO-PSP-Net model's effectiveness and performance in obtaining high segmentation accuracy in comparison with other state-of-the-art approaches.

Alzheimer's disease early screening and staged detection with plasma proteome using machine learning and convolutional neural network.

Alzheimer's disease (AD) stands as the prevalent progressive neurodegenerative disease, precipitating cognitive impairment and even memory loss. Amyloid biomarkers have been extensively used in the diagnosis of AD. However, amyloid proteins offer limited information about the disease process and accurate diagnosis depends on the presence of a substantial accumulation of amyloid deposition which significantly impedes the early screening of AD. In this study, we have combined plasma proteomics with an ensemble learning model (CatBoost) to develop a cost-effective and non-invasive diagnostic method for AD. A longitudinal panel has been identified that can serve as reliable biomarkers across the entire progression of AD. Simultaneously, we have developed a neural network algorithm that utilizes plasma proteins to detect stages of Alzheimer's disease. Based on the developed longitudinal panel, the CatBoost model achieved an area under the operating curve of at least 0.90 in distinguishing mild cognitive impairment from cognitively normal. The neural network model was utilized for the detection of three stages of AD, and the results demonstrated that the neural network model exhibited an accuracy as high as 0.83, surpassing that of the traditional machine learning model.

Accelerating high-entropy alloy discovery: efficient exploration via active learning

The exploration of the complex composition space for high entropy alloys (HEAs) is extremely challenging and resource intensive using traditional materials discovery approaches. Here, we apply Bayesian optimization in the form of active learning with a neural network model to efficiently explore the vast composition space of HEAs with a focus on predicting their stable phases. For example, by focusing data acquisition on the uncertain regions, we achieved a testing accuracy of 95% with 27% of an experimental HEA dataset, comparable to the accuracy (94.6%) of a random forest model trained with 80% of the full dataset (2198 quinary HEAs). Similarly, with just ∼2% of a CALPHAD dataset we achieved a testing accuracy for phase predictions above 96%, which is comparable to the accuracy (above 97%) achieved when an XGboost model is trained with nearly the full dataset (664,650 quinary HEAs). Thus, our approach greatly facilitates both computational and experimental exploration of HEA design spaces.

Systemic Risk Arising from Shadow Banking and Sustainable Development: A Study of Wealth Management Products in China

Shadow banking is a main way for the financial market to serve the real economy today, and this process is closely related to systemic risk. This study examines the impact of shadow banking associated with sustainable development in China’s banking on systemic risk. We analyze the data obtained from a rich sample of 31 listed commercial banks in China and shadow banking represented by wealth management products (WMPs) by constructing a dynamic complex interbank network model. The results show that the risks and vulnerabilities generated by shadow banking spread out through the interbank network and cause systemic risk to increase. The effect operates through increasing the number of default banks, reducing banks’ survival rate and profit, and forcing central bank bailout funds expansion. However, it has a positive impact in terms of augmenting liquidity and enhancing investment opportunities. Furthermore, the variability in the influence of different categories of shadow banking is assessed, emphasizing that short-term shadow banking exerts a more pronounced impact on systemic risk. In addition, the heterogeneity of the shadow banking effect on different types of commercial banks is explored, revealing that local and rural commercial banks experience a more conspicuous effect compared to state-owned and joint-stock banks. Our findings highlight that improving external supervision, promoting financial internal governance, and constraining credit linkages are vital for alleviating the increase in risks in shadow banking and maintaining the sustainable development of banking.

Sign language recognition based on dual-path background erasure convolutional neural network

Sign language is an important way to provide expression information to people with hearing and speaking disabilities. Therefore, sign language recognition has always been a very important research topic. However, many sign language recognition systems currently require complex deep models and rely on expensive sensors, which limits the application scenarios of sign language recognition. To address this issue, based on computer vision, this study proposed a lightweight, dual-path background erasing deep convolutional neural network (DPCNN) model for sign language recognition. The DPCNN consists of two paths. One path is used to learn the overall features, while the other path learns the background features. The background features are gradually subtracted from the overall features to obtain an effective representation of hand features. Then, these features are flatten into a one-dimensional layer, and pass through a fully connected layer with an output unit of 128. Finally, use a fully connected layer with an output unit of 24 as the output layer. Based on the ASL Finger Spelling dataset, the total accuracy and Macro-F1 scores of the proposed method is 99.52% and 0.997, respectively. More importantly, the proposed method can be applied to small terminals, thereby improving the application scenarios of sign language recognition. Through experimental comparison, the dual path background erasure network model proposed in this paper has better generalization ability.