Properties of radiation defects and threshold energy of displacement in zirconium hydride obtained by new deep-learning potential

Zirconium hydride(ZrH_(2)) is an ideal neutron moderator material. However, radiation effect significantly changes its properties, which affect its behavior and the lifespan of the reactor. The threshold energy of displacement is an important quantity of the number of radiation defects produced, which helps us to predict the evolution of radiation defects in ZrH_(2).Molecular dynamics(MD) and ab initio molecular dynamics(AIMD) are two main methods of calculating the threshold energy of displacement. The MD simulations with empirical potentials often cannot accurately depict the transitional states that lattice atoms must surpass to reach an interstitial state. Additionally, the AIMD method is unable to perform largescale calculation, which poses a computational challenge beyond the simulation range of density functional theory. Machine learning potentials are renowned for their high accuracy and efficiency, making them an increasingly preferred choice for molecular dynamics simulations. In this work, we develop an accurate potential energy model for the ZrH_(2) system by using the deep-potential(DP) method. The DP model has a high degree of agreement with first-principles calculations for the typical defect energy and mechanical properties of the ZrH_(2) system, including the basic bulk properties, formation energy of point defects, as well as diffusion behavior of hydrogen and zirconium. By integrating the DP model with Ziegler–Biersack–Littmark(ZBL) potential, we can predict the threshold energy of displacement of zirconium and hydrogen in ε-ZrH_(2).

Few-shot working condition recognition of a sucker-rod pumping system based on a 4-dimensional time-frequency signature and meta-learning convolutional shrinkage neural network

The accurate and intelligent identification of the working conditions of a sucker-rod pumping system is necessary. As onshore oil extraction gradually enters its mid-to late-stage, the cost required to train a deep learning working condition recognition model for pumping wells by obtaining enough new working condition samples is expensive. For the few-shot problem and large calculation issues of new working conditions of oil wells, a working condition recognition method for pumping unit wells based on a 4-dimensional time-frequency signature (4D-TFS) and meta-learning convolutional shrinkage neural network (ML-CSNN) is proposed. First, the measured pumping unit well workup data are converted into 4D-TFS data, and the initial feature extraction task is performed while compressing the data. Subsequently, a convolutional shrinkage neural network (CSNN) with a specific structure that can ablate low-frequency features is designed to extract working conditions features. Finally, a meta-learning fine-tuning framework for learning the network parameters that are susceptible to task changes is merged into the CSNN to solve the few-shot issue. The results of the experiments demonstrate that the trained ML-CSNN has good recognition accuracy and generalization ability for few-shot working condition recognition. More specifically, in the case of lower computational complexity, only few-shot samples are needed to fine-tune the network parameters, and the model can be quickly adapted to new classes of well conditions.

Vehicle Abnormal Behavior Detection Based on Dense Block and Soft Thresholding

With the rapid advancement of social economies,intelligent transportation systems are gaining increasing atten-tion.Central to these systems is the detection of abnormal vehicle behavior,which remains a critical challenge due to the complexity of urban roadways and the variability of external conditions.Current research on detecting abnormal traffic behaviors is still nascent,with significant room for improvement in recognition accuracy.To address this,this research has developed a new model for recognizing abnormal traffic behaviors.This model employs the R3D network as its core architecture,incorporating a dense block to facilitate feature reuse.This approach not only enhances performance with fewer parameters and reduced computational demands but also allows for the acquisition of new features while simplifying the overall network structure.Additionally,this research integrates a self-attentive method that dynamically adjusts to the prevailing traffic conditions,optimizing the relevance of features for the task at hand.For temporal analysis,a Bi-LSTM layer is utilized to extract and learn from time-based data nuances.This research conducted a series of comparative experiments using the UCF-Crime dataset,achieving a notable accuracy of 89.30%on our test set.Our results demonstrate that our model not only operates with fewer parameters but also achieves superior recognition accuracy compared to previous models.

Hyperparameter on-line learning of stochastic resonance based threshold networks

Aiming at training the feed-forward threshold neural network consisting of nondifferentiable activation functions, the approach of noise injection forms a stochastic resonance based threshold network that can be optimized by various gradientbased optimizers. The introduction of injected noise extends the noise level into the parameter space of the designed threshold network, but leads to a highly non-convex optimization landscape of the loss function. Thus, the hyperparameter on-line learning procedure with respective to network weights and noise levels becomes of challenge. It is shown that the Adam optimizer, as an adaptive variant of stochastic gradient descent, manifests its superior learning ability in training the stochastic resonance based threshold network effectively. Experimental results demonstrate the significant improvement of performance of the designed threshold network trained by the Adam optimizer for function approximation and image classification.

Threshold Filtering Semi-Supervised Learning Method for SAR Target Recognition

The semi-supervised deep learning technology driven by a small part of labeled data and a large amount of unlabeled data has achieved excellent performance in the field of image processing.However,the existing semisupervised learning techniques are all carried out under the assumption that the labeled data and the unlabeled data are in the same distribution,and its performance is mainly due to the two being in the same distribution state.When there is out-of-class data in unlabeled data,its performance will be affected.In practical applications,it is difficult to ensure that unlabeled data does not contain out-of-category data,especially in the field of Synthetic Aperture Radar(SAR)image recognition.In order to solve the problem that the unlabeled data contains out-of-class data which affects the performance of the model,this paper proposes a semi-supervised learning method of threshold filtering.In the training process,through the two selections of data by the model,unlabeled data outside the category is filtered out to optimize the performance of the model.Experiments were conducted on the Moving and Stationary Target Acquisition and Recognition(MSTAR)dataset,and compared with existing several state-of-the-art semi-supervised classification approaches,the superiority of our method was confirmed,especially when the unlabeled data contained a large amount of out-of-category data.

Iterative learning based fault diagnosis for discrete linear uncertain systems

In order to detect and estimate faults in discrete lin-ear time-varying uncertain systems, the discrete iterative learning strategy is applied in fault diagnosis, and a novel fault detection and estimation algorithm is proposed. And the threshold limited technology is adopted in the proposed algorithm. Within the chosen optimal time region, residual signals are used in the proposed algorithm to correct the introduced virtual faults with iterative learning rules, making the virtual faults close to these occurred in practical systems. And the same method is repeated in the rest optimal time regions, thereby reaching the aim of fault diagnosis. The proposed algorithm not only completes fault detection and estimation for discrete linear time-varying uncertain systems, but also improves the reliability of fault detection and decreases the false alarm rate. The final simulation results verify the validity of the proposed algorithm.

Statistic Learning-based Defect Detection for Twill Fabrics

Template matching methods have been widely utilized to detect fabric defects in textile quality control. In this paper, a novel approach is proposed to design a flexible classifier for distinguishing flaws from twill fabrics by statistically learning from the normal fabric texture. Statistical information of natural and normal texture of the fabric can be extracted via collecting and analyzing the gray image. On the basis of this, both judging threshold and template are acquired and updated adaptively in real-time according to the real textures of fabric, which promises more flexibility and universality. The algorithms are experimented with images of fault free and faulty textile samples.

Sound transducer calibration of ambulatory audiometric system utilizing delta learning rule

An efficient calibration algorithm for an ambulatory audiometric test system is proposed. This system utilizes a personal digital assistant (PDA) device to generate the correct sound pressure level (SPL) from an audiometric transducer such as an earphone. The calibrated sound intensities for an audio-logical examination can be obtained in terms of the sound pressure levels of pure-tonal sinusoidal signals in eight-banded frequency ranges (250, 500, 1 000, 2 000, 3 000, 4 000, 6 000 and 8 000 Hz), and with mapping of the input sound pressure levels by the weight coefficients that are tuned by the delta learning rule. With this scheme, the sound intensities, which evoke eight-banded sound pressure levels by 5 dB steps from a minimum of 25 dB to a maximum of 80 dB, can be generated without volume displacement. Consequently, these sound intensities can be utilized to accurately determine the hearing threshold of a subject in the ambulatory audiometric testing environment.

A Double Threshold Energy Detection-Based Neural Network for Cognitive Radio Networks

In cognitive radio networks(CoR),the performance of cooperative spectrum sensing is improved by reducing the overall error rate or maximizing the detection probability.Several optimization methods are usually used to optimize the number of user-chosen for cooperation and the threshold selection.However,these methods do not take into account the effect of sample size and its effect on improving CoR performance.In general,a large sample size results in more reliable detection,but takes longer sensing time and increases complexity.Thus,the locally sensed sample size is an optimization problem.Therefore,optimizing the local sample size for each cognitive user helps to improve CoR performance.In this study,two new methods are proposed to find the optimum sample size to achieve objective-based improved(single/double)threshold energy detection,these methods are the optimum sample size N^(*)and neural networks(NN)optimization.Through the evaluation,it was found that the proposed methods outperform the traditional sample size selection in terms of the total error rate,detection probability,and throughput.

WiCare:一种非接触式的老人如厕跌倒监测模型

老人在卫生间内的跌倒行为存在因救助及时性差而导致严重危害的风险,因此高效快捷的如厕跌倒监测研究具有重要意义。针对当前基于Wi-Fi感知的跌倒监测方法中存在的受噪声影响大而特征提取不充分、监测精度有限的问题,提出了一种基于多级离散小波变换和软阈值处理的信号降噪算法,及一种融合卷积神经网络、双向长短期记忆网络及自注意力机制的非接触式如厕跌倒监测模型WiCare。首先,从原始CSI数据中提取振幅作为基础数据;其次,使用多级离散小波变换和软阈值处理进行感知数据降噪;然后,将感知数据进行多维重构,以更准确地表征跌倒行为特征;最后,利用WiCare提取感知数据中的有效特征,进而实现卫生间如厕跌倒行为监测功能。实验结果表明,WiCare在居家卫生间环境下对跌倒行为监测的准确率为99.41%,与其他同类模型相比,WiCare的识别准确率高,模型复杂度低,且泛化能力更强。

基于深度强化学习和隐私保护的群智感知动态任务分配策略

在移动群智感知(Mobile Crowd Sensing,MCS)中,动态任务分配的结果对提高系统效率和确保数据质量至关重要。然而,现有的大部分研究在处理动态任务分配时,通常将其简化为二分匹配模型,该简化模型未充分考虑任务属性与工人属性对匹配结果的影响,同时忽视了工人位置隐私的保护问题。针对这些不足,文章提出一种基于深度强化学习和隐私保护的群智感知动态任务分配策略。该策略首先通过差分隐私技术为工人位置添加噪声,保护工人隐私;然后利用深度强化学习方法自适应地调整任务批量分配;最后使用基于工人任务执行能力阈值的贪婪算法计算最优策略下的平台总效用。在真实数据集上的实验结果表明,该策略在不同参数设置下均能保持优越的性能,同时有效地保护了工人的位置隐私。

基于增强多头注意力机制的Optuna-BiGRU测井岩性识别

测井岩性识别是油气勘探开发中至关重要的内容。针对现有算法模型在处理测井曲线数据时,无法有效捕获曲线内部深层关联和深度方向关系、拟合能力较弱、难以准确提取关键特征、噪声干扰以及模型超参数调优过程复杂困难等问题,提出了一种通过Optuna超参数优化双向门循环单元(Optuna-BiGRU)结合增强多头注意力机制(EMHA)的测井岩性识别模型——Optuna-BiGRU-EMHA模型。该模型引入残差机制和层归一化以改进多头注意力机制模块,并结合双向门循环单元(BiGRU)解决了处理测井数据时的问题,同时使用Optuna超参数优化框架和小波包自适应阈值方法分别解决了超参数调优和噪声干扰问题。首先通过交会图分析和敏感性箱线图分析选取自然伽马、深感应电阻率、中子-密度孔隙度、平均中子-密度孔隙度和岩性密度5个特征参数的测井数据,通过小波包自适应阈值方法对数据进行去噪,并将测井数据分割成数据块,然后利用Optuna框架优化BiGRU-EMHA模型超参数,最后通过实验对比K-近邻算法(KNN)、随机森林(RF)、极端梯度提升算法(XGBoost)、长短期记忆(LSTM)神经网络、BiGRU、双向长短期记忆(BiLSTM)神经网络、BiGRU-MHA、Optuna-BiGRU-EMHA等8种模型在测井岩性识别中的精度。结果表明:Optuna-BiGRU-EMHA模型识别准确率达到80%,相对于传统机器学习模型和深度学习模型,综合岩性识别准确率分别提高15.94%~23.14%和3.93%~15.94%,该模型为常规测井岩性识别提供了坚实的理论支持。

基于多残差注意力深度收缩网络的超微光图像增强方法

超微光成像可在极度黑暗的环境中给观察者提供近乎白昼的视觉体验,在许多民用和军事应用中起着至关重要的作用。超微光环境下拍摄的图像和视频通常存在亮度与对比度极低、噪声水平高、场景细节和色彩严重缺失等固有缺陷,近年来,深度学习为超微光成像的研究带来了新的机遇。文中采集并提供了一组实用性更强的超微光训练数据集,提出了一种多残差注意力深度收缩网络(Multi Residual Attention Shrinkage Network),以此实现了一种新的超微光成像方法。通过成功研制的小型化样机证实了该方法的工业量产前景。实现了基于通道注意力和空间注意力的残差内注意力机制,以及基于深度软阈值收缩的外注意力机制,不仅可以有效提取并还原极低照度环境下的图像细节信息,恢复场景真实色彩,而且可以有效去除此类环境下由成像设备感光不足带来的巨量噪声。实测效果显示该方法可对极低照度环境进行有效的增强且实时性高。通过与多种业界最新方法比较,文中方法在主观视觉体验以及客观参数两方面均表现更好。

基于自反馈阈值学习的半监督皮肤癌诊断模型

为解决监督学习皮肤癌诊断模型的训练需要大量数据标注,且医学专家标注工作成本高、耗时长、易疲劳等问题,提出了一种基于自反馈阈值学习(Self-Feedback Threshold Learning,SFTL)的半监督皮肤癌诊断方法.在标注数据预训练的ResNet网络基础上,引入全局和局部类别间伪标签自反馈阈值学习机制动态筛选ResNet预测概率大于自反馈阈值的无标记样本,引入无监督阈值学习损失和分类交叉熵损失进行模型训练,在标记样本稀缺的情况下深入挖掘无标记数据的鉴别诊断信息,显著降低模型在无标记皮肤病变图像中的误判率.选取公开数据集HAM10000的皮肤病变图像展开实验验证,在仅需50%标记数据下实现了0.8229的准确率和0.7651的F1分数,证明所提出的SFTL模型在半监督场景下可有效解决皮肤癌诊断任务,相比其他同类方法具有更好的分类性能.

遥感影像水体识别研究进展及热点分析

洪涝灾害是严重危害人类社会发展的自然灾害之一,从遥感影像中准确识别水体信息对提升突发洪涝灾害事件的应急处理、灾害预警具有重要意义,一直受到专家学者的关注。针对国内外专家学者利用遥感影像进行水体识别的相关研究成果,文章从传统方法以及深度学习法两大方面,对前人提出的水体识别理论和方法进行分析研究;在此基础上,利用文献计量的方法对水体识别当前的研究热点进行梳理分析,针对现存问题和当前热点对基于遥感影像的水体识别未来发展趋势进行了展望。

基于深度SVDD-CVAE的轴承自适应阈值故障检测

通过状态监测进行轴承故障报警,能有效避免设备灾难性事故的发生。基于数据时序特征重构的故障检测法由于仅采用正常数据进行训练,能有效避免故障数据不足而导致的模型检测精度下降。然而,此类方法的故障阈值确定依赖于大量的历史数据,且对检测精度有着极大的影响。为此,提出基于深度SVDD-CVAE的轴承自适应阈值故障检测方法。针对时序信号特征增强提取构建ConvLSTM作为基础单元的CVAE特征压缩提取框架,有效提取轴承故障微弱特征;结合SVDD自适应学习特征空间超球面,实现故障检测阈值的自适应确定;最后,通过全局误差损失反向传播对深度SVDD-CVAE框架进行迭代优化。实验结果表明:所提出的方法能有效提取轴承微弱故障特征、自适应确定阈值,并在IMS轴承数据集上取得97.7%的检测准确率。

基于高光谱成像的烤烟着生部位识别

【目的】采用高光谱成像技术结合机器学习方法,建立烤烟着生部位(上部、中部、下部)的识别模型。【方法】首先,通过分析烟叶在水、氮敏感波段下的强度分布特征,采用了一种结合OTSU和Sauvola图像分割算法的双阈值感兴趣区(ROI)选取方法,然后对比分析不同预处理方法对数据建模的影响规律,采用支持向量机(SVM)、极限梯度提升(XGBoost)算法进行判别模型的建立,通过参数寻优进行模型的优化。使用遗传算法(GA)和遗传算法结合连续投影算法(GA-SPA)进行特征波长的选择,建立简化模型。【结果】(1)建立的双阈值感兴趣区选取方法能准确高效地实现烤烟叶片正常叶面区域的选取(2)不同数据预处理方法对识别模型影响较为显著,基于一阶导和萨维莱茨-戈莱平滑(1Der+SG)预处理光谱数据,结合GA选取的特征波长建立的XGBoost着生部位识别模型具有最佳的分类效能,其准确率高达97.78%。【结论】研究建立的基于高光谱成像技术结合机器学习方法的部位模型可满足烤烟着生部位的高效准确识别。

基于深度学习的细骨料图像实时分割提取

文章基于深度学习算法对细骨料投影图像进行分割,通过对比传统阈值分割与PSPNet、DeepLab V3+、U-Net深度学习网络模型算法的分割结果对4种模型进行评价分析,同时实验对比细骨料2种等效粒径计算方法(单面投影法、双面投影法)的粒径和级配分布结果。结果表明:深度学习模型算法中U-Net网络模型的准确率、召回率、F平衡分数和交并比分别达到99.8%、88.1%、84.9%、84.3%,均优于对比组模型;对于3种不同粒径的单粒段细骨料,采用双面投影法计算出的细骨料等效粒径D d与实际细骨料粒径的偏差分别为1.40%、2.10%、3.12%;对于混合粒段骨料,采用等效粒径D d计算出的级配分布曲线更接近筛分法的实验结果,具有普遍适用性。研究结果可为细骨料径粒径和粒型参数的计算提取提供新的思路。

A novel non-intrusive load monitoring technique using semi-supervised deep learning framework for smart grid

Non-intrusive load monitoring(NILM)is a technique which extracts individual appliance consumption and operation state change information from the aggregate power consumption made by a single residential or commercial unit.NILM plays a pivotal role in modernizing building energy management by disaggregating total energy consumption into individual appliance-level insights.This enables informed decision-making,energy optimization,and cost reduction.However,NILM encounters substantial challenges like signal noise,data availability,and data privacy concerns,necessitating advanced algorithms and robust methodologies to ensure accurate and secure energy disaggregation in real-world scenarios.Deep learning techniques have recently shown some promising results in NILM research,but training these neural networks requires significant labeled data.Obtaining initial sets of labeled data for the research by installing smart meters at the end of consumers’appliances is laborious and expensive and exposes users to severe privacy risks.It is also important to mention that most NILM research uses empirical observations instead of proper mathematical approaches to obtain the threshold value for determining appliance operation states(On/Off)from their respective energy consumption value.This paper proposes a novel semi-supervised multilabel deep learning technique based on temporal convolutional network(TCN)and long short-term memory(LSTM)for classifying appliance operation states from labeled and unlabeled data.The two thresholding techniques,namely Middle-Point Thresholding and Variance-Sensitive Thresholding,which are needed to derive the threshold values for determining appliance operation states,are also compared thoroughly.The superiority of the proposed model,along with finding the appliance states through the Middle-Point Thresholding method,is demonstrated through 15%improved overall improved F1micro score and almost 26%improved Hamming loss,F1 and Specificity score for the performance of individual appliance when compared to the benchmarking techniques that also used semi-supervised learning approach.

基于通用学习均衡优化器的多阈值图像分割

传统的元启发式多阈值图像分割算法计算复杂度高且容易陷入局部最优,通用学习均衡优化器在搜索过程中使粒子从不同维度的候选粒子中学习,在求解复杂问题最优解时有很强的能力,克服了容易陷入局部最优的问题。提出将通用学习均衡优化算法优化最大类间方差法来实现多阈值图像分割,实验选择标准灰度图像,以峰值信噪比、结构相似度、运行时间和适应度值为评价标准,将该算法与均衡优化算法、粒子群优化算法进行了比较。结果表明,基于通用学习均衡优化器的多阈值图像分割算法结果的峰值信噪比、结构相似度在绝大多数情况下优于另外两个算法,并且收敛速度快,执行效率高。