基于深度学习的气溶胶荧光光谱识别应用研究

空气中的高危病原微生物对人类社会存在着极大威胁,而传统的监测方法无法对空气中的微生物实现准确的识别与分类。因此采用激光诱导荧光技术原理,以单光子探测器为核心器件,设计并搭建了一种高效的荧光光谱仪用于空气中高危病原微生物的识别与分类,并且该光谱仪可以实现对微生物浓度的预测,其对于环境安全具有重要意义。对于该光谱仪采集的数据,探索了以一维向量和二维矩阵2种输入形式来实现荧光光谱的识别与分类,并研究对比了主成分分析网络、卷积神经网络和全卷积网络等深度学习网络的识别与分类效果。实验结果表明以矩阵形式输入的卷积神经网络模型在测试集中识别分类准确率达到98.05%。采用矩阵形式输入的全卷积网络模型在测试集中微生物浓度预测准确率达到98.97%。     

基于深度残差网络的脱机手写汉字识别研究

摘 要: 手写汉字识别是模式识别与机器学习的重要研究方向和应用领域。近年来,随着深度学习理论方法的完善、新技术的层出不穷,深度神经网络在图像识别分类、图像生成等典型应用中取得了突破性的进展,其中,深度残差网络作为最新的研究成果,已成功应用于手写数字识别、图片识别分类等多个领域。本文将研究深度残差网络在脱机孤立手写汉字识别中的应用方法,通过改进残差学习模块的单元结构,优化深度残差网络性能,同时通过对训练集的预处理,从数据层面实现训练生成模型性能的提升,最后设计实验,验证深度残差网络、End-to-End模式在脱机手写汉字识别中的可行性,分析、总结存在的问题及今后的研究方向。     

激光测距在测量船舶近场辐射噪声中的应用

本文主要阐述在海上测量船舶的近场辐射噪声中，利用激光测距，快速，准确得到了被测船舶与测量水听器之间的正横距离。     

基于时延神经网络模型的舰船辐射噪声目标识别*

水声目标被动识别是水声信号处理领域的研究热点之一。海洋环境中存在的不规则噪声干扰,使得基于传统方法的水声目标被动识别技术在实际的应用场景中效果不佳。本文采用一种基于时延网络(Time Delay Neural Network,TDNN)模型的舰船辐射噪声目标识别方法,该方法利用目标的短时平稳特性和长时关联特性对目标的声纹特征进行建模,使用梅尔谱图提取目标信号的初级特征,再通过融合注意力机制和时延神经网络的深度学习模型实现高级特性提取,最后再利用余弦相似度实现不同目标的类别划分。该方法在ShipsEar数据集和自行采集的数据进行测试验证,目标识别准确率分别达到79.2%和73.9%,可证明本文方法的有效性。     

基于深度学习的故障检测方法

针对传统故障诊断方法中特征提取技术难度大、故障样本获取困难等问题,在深度学习计算框架下提出了一种半监督训练的故障检测方法,利用深度信念网络中的受限波茨曼机堆栈结构实现了数据高层特征的自动提取,结合支持向量数据描述方法实现了异常数据检测,只需利用正常工况的数据样本进行网络训练和模型拟合,无需故障样本数据,也无需人工干预进行信号特征提取,即能实现对故障数据进行的实时检测和判别。经采用标准轴承实验数据的三组故障数据进行验证,故障识别率达到100%,具有很强的工程应用价值。     

深度学习在超声检测缺陷识别中的应用与发展*

深度学习（Deep Learning）是目前最强大的机器学习算法之一,其中卷积神经网络（Convolutional Neural Network, CNN）模型具有自动学习特征的能力,在图像处理领域较其他深度学习模型有较大的性能优势。本文先简述了深度学习的发展史,然后综述了深度学习在超声检测缺陷识别中的应用与发展,从早期浅层神经网络到现在深度学习的应用现状,并借鉴医学影像识别和射线图像识别领域的方法,分析了卷积神经网络对超声图像缺陷识别的适用性。最后,探讨归纳了目前在超声检测图像识别中使用CNN存在的一些问题,及其主要应对策略的研究方向。     

基于深度学习的252Cf源驱动核材料浓度识别技术

针对核武器/核材料识别系统中核材料浓度识别的关键技术问题,采用Monte Carlo方法,通过建立252Cf源驱动核材料裂变中子信号样本库,模拟分析了随探测器距离和角度及核材料浓度变化的裂变脉冲中子信号特点,基于深度学习之卷积神经网络,构建了一种252Cf源驱动核材料浓度识别方法,实现了对测试样本浓度的识别,且还与BP神经网络和K最近邻方法进行了对比试验研究。结果表明,卷积神经网络算法进行核材料浓度识别,得到了高达92.05%识别准确率,不仅解决了因探测器距离和角度变化时对核材料浓度识别准确率影响的难题,而且还获得了优于BP神经网络和K最近邻算法对核材料浓度识别的认识,这为252Cf源驱动核材料浓度识别提供了一种新的途径。     

基于深度学习的两分量BEC中量子相变点的识别

识别物质的相变是物理学研究中一个重要问题.本文采用了一种混淆标签方案的卷积神经网络算法来识别两分量玻色-爱因斯坦凝聚(BEC)中量子相变点,通过计算神经网络输出的准确率,得到W型性能曲线,此性能曲线中间的极大值对应着量子相变的临界点.研究结果表明,深度学习得到的量子相变点与解析计算值吻合度较高.此混淆标签方案的深度学习研究方法可以应用到存在两种相的相变体系.     

利用盲反卷积和混沌振子增强船舶辐射噪声解调线谱

针对经典解调方法中心频率、带宽选择困难和解调线谱受带外噪声干扰难以分辨等问题,提出采用盲反卷积和混沌振子方法抑制带外噪声,增强船舶噪声解调线谱。该方法通过Duffing振子预检宽带船舶噪声低频弱周期信号,随后将相应频率作为最小噪声幅值比反卷积(MNAD)方法的先验参数,利用MNAD方法自适应搜索解调频带得到可清晰分辨的高信噪比解调线谱。仿真和实测数据分析表明,该方法较经典解调方法和其他盲反卷积方法,可获取更佳的中心频率和滤波带宽,所得解调线谱的窄带信噪比DF值最高。     

船舶辐射噪声信号仿真及评价

船舶辐射噪声信号仿真应尽可能多地仿真特定船舶辐射噪声的信号特征,达到具体应用对信号仿真度的要求。对船舶辐射噪声仿真信号进行评价,是调整仿真参数和判定仿真信号是否达到要求的必要步骤。该文给出了一套对实测船舶辐射噪声信号进行重构仿真及评价的方法,仿真过程考虑了信号的听觉、连续谱、调制谱和线谱特征,评价方法对各特征进行指标提取和量化,并通过实测与仿真信号之间的相似度计算评价仿真信号的仿真度。对实测船舶辐射噪声信号进行重构仿真的试验结果显示,仿真信号具有较高的特征仿真度,与计算的仿真度评价数值一致。     

基于深度学习的ELM实时识别研究

基于深度学习的方法,在HL-2A装置上开发出了一套边缘局域模(ELM)实时识别算法。算法使用5200次放电数据(约24.19万数据切片)进行学习,得到一个深度为22层的卷积神经网络。为衡量算法的识别能力,识别了HL-2A装置自2009年实现稳定ELMyH模放电以来所有历史数据(约26000次放电数据),共识别出1665次H模放电,其中误识别35次,误报率为2.10%。在实际的1634次H模放电中,漏识别4次,漏识别率为0.24%。该误报率和漏报率可以满足ELM实时识别的精度要求。识别算法在实时控制环境下,对单个时间点的平均计算时间为0.46ms,可以满足实时控制的计算速度要求。     

应用深度学习识别法兰螺栓连接状态

针对常规诊断方法对螺栓的连接状态识别效果差、鲁棒性和抗噪性弱等问题,提出了基于深度学习理论的螺栓检测新方法.首先以4种预紧力状态下的法兰螺栓结构产生的声发射信号为研究对象,借助于自适应噪声的完整集成经验模态分解理论以及梅尔频率倒谱系数特征提取方式,实现了声发射信号的自适应消噪和最优模态函数分量组的选取,提取到了可以较好...     

Handwritten digit recognition based on ghost imaging with deep learning

We present a ghost handwritten digit recognition method for the unknown handwritten digits based on ghost imaging(GI) with deep neural network, where a few detection signals from the bucket detector, generated by the cosine transform speckle, are used as the characteristic information and the input of the designed deep neural network(DNN), and the output of the DNN is the classification. The results show that the proposed scheme has a higher recognition accuracy(as high as98% for the simulations, and 91% for the experiments) with a smaller sampling ratio(say 12.76%). With the increase of the sampling ratio, the recognition accuracy is enhanced. Compared with the traditional recognition scheme using the same DNN structure, the proposed scheme has slightly better performance with a lower complexity and non-locality property.The proposed scheme provides a promising way for remote sensing.     

非高斯环境下的深度学习脉冲信号去噪与重构*

侧扫声呐进行沉底小目标探测时,底混响是主要背景干扰。底混响通常是一种非平稳、非高斯的带限噪声,它使得白噪声条件下的滤波器性能受到限制。在混响背景下常利用自回归模型对接收信号进预行白化处理,但对于实际侧扫声呐应用,白化后直接匹配滤波的处理效果不甚理想。针对此问题,在自回归模型预白化的基础上,提出采用一种次最佳检测与多分辨二分奇异值分解相结合的改进方法。该方法首先对接收信号进行分段处理,利用改进Burg算法估计每段数据自回归模型的系数及阶数;然后构造白化滤波器对分段数据预白化,并对白化后的数据进行多分辨二分奇异值分解;最后应用ostu方法对原始声图和处理后的声图进行目标检测。仿真与实验结果表明,该方法明显提高了信混比,改善了侧扫声呐沉底静态小目标的成图质量,有利于后期实现基于图像的目标自动检测。     

连续音素的改进深信度网络的识别算法*

为提高连续语音识别中的音素识别率,提出一种基于改进并行回火训练的受限波尔兹曼机的音素识别算法。首先,利用经过等能量划分后的改进并行回火算法来训练受限玻尔兹曼机,接着将受限玻尔兹曼机堆叠组成一个深信度网络,从而作为深度神经网络预训练的基础模型,然后通过softmax层输出,得到用于音素状态后验概率检测的深度神经网络。接着,利用少量的标签数据,根据反向传播算法对网络权重进行微调。最后,将所得后验概率作为隐马尔科夫的发射概率,然后利用Viterbi解码器实现音素识别。在TIMIT语料库上的实验表明,识别率相比于传统的对比散度类算法提高了约4.5%,在不增加计算量的情况下比原始并行回火算法提高约1%。     

基于双向循环神经网络的汉语语音识别*

当前基于深度神经网络模型中,虽然其隐含层可设置多层,对复杂问题适应能力强,但每层之间的节点连接是相互独立的,这种结构特性导致了在语音序列中无法利用上下文相关信息来提高识别效果,而传统的循环神经网络虽然做出了改进,但是只能对上文信息进行利用。针对以上问题,该文采用可以同时利用语音序列中上下文相关信息的双向循环神经网络模型与深度神经网络模型相结合,并应用于语音识别。构建具有5层隐含层的模型,其中第3层为双向循环神经网络结构,其他层采用深度神经网络结构。实验结果表明:加入了双向循环神经网络结构的模型与其他模型相比,较好地提高了识别正确率;噪声对双向循环神经网络汉语识别有重要影响,尤其是训练集和测试集附加噪声类型不同时,单一的含噪声语音的训练模型无法适应不同噪声类型的语音识别;调整神经网络模型中隐含层神经元数量后,识别正确率并不是一直随着隐含层中神经元数量的增加而增加,神经元数量数目增加到一定程度后正确率出现了降低的趋势。     

Signal area estimation based on deep learning

In many practical application scenarios, radio communication signals are commonly represented as a spectrogram, which represents the signal strength measured at multiple discrete time instants and frequency points within a specific time interval and frequency band, respectively. In the context of spectrum occupancy measurements, the notion of Signal Area (SA) is defined as the rectangular region in the time–frequency domain where a signal is assumed to be present. Signal Area Estimation (SAE) is an important functionality in spectrum-aware wireless systems where spectrum usage monitoring is required. However, the conventional approaches to SAE have a limited estimation accuracy, in particular at low SNR. In this work, a novel technique for SAE is proposed using Deep Learning based on Artificial Neural Network (DL-ANN) for enhanced extraction of SA information from radio spectrograms. The performance of the proposed DL-ANN method is evaluated both with software simulations and hardware experiments, and the results are compared with several conventional methods from the literature, showing significant performance improvements. A key feature of the proposed method is the improvement in the SAE accuracy compared to other existing methods (in particular in the low SNR regime) and the capability to extract the location of the detected SAs automatically. Overall, the proposed technique is a promising solution for the automatic processing of radio spectrograms in spectrum-aware wireless systems.     

UAV path design with connectivity constraint based on deep reinforcement learning

Cellular networks are expected to communicate effectively with unmanned aerial vehicles (UAVs) and support various applications. However, existing cellular networks are primarily designed to cover users on the ground; thus, coverage holes in the sky will exist. In this paper, we investigate the problem of path design for cellular-connected UAVs, taking into account the interruption performance throughout the UAV mission to minimize the completion time. Two types of connectivity constraints requirements are assumed to be available. The first is defined as the maximum continuous time interval that the UAV loses connection with base stations (BSs) below a predefined threshold. For the second, we consider the sum outage of UAV is limited during the entire UAV mission. The UAV is tasked with flying from a starting location to a final destination while minimization the mission time, satisfying the two constraints, separately. The formulated path design problem which involves continues variables and a dynamic radio environment, is not convex and thus is extremely difficult to solve directly. To tackle this challenge, a deep reinforcement learning (DRL) based trajectory design algorithm is proposed, where the Dueling Double Deep Q Network(Dueling DDQN) with multi-steps learning method is applied. Simulation results demonstrate the effectiveness of the proposed DRL algorithm and achieve a trade-off between the trajectory length of the UAV and connection quality.     

Sound quality prediction of vehicle interior noise using deep belief networks

The sound quality of vehicle interior noise strongly influences passengers’ psychological and physiological perceptions. To predict the sound quality of interior noise, a vehicle road test with four compact cars has been conducted. All recorded interior noise signals have been denoised via a discrete wavelet transform (DWT) denoising procedure and subsequently evaluated subjectively through the anchor semantic differential (ASD) test by a jury. In addition, a novel prediction method, namely, regression-based deep belief networks (DBNs), which substitute the support vector regression (SVR) layer for the linear softmax classification layer at the top of the general DBN’s structure, has been proposed to predict the interior sound quality. The parameter selection of the DBN model has been compared and studied using a grid search. In addition, four conventional machine-learning-based methods have been introduced to enable a comparison of the performance with the newly developed DBNs. Furthermore, the feature fusion ability of DBNs has been studied by varying the amount of information that the dataset offers. The results show the following: (1) The accuracy and robustness of the proposed DBN-based sound quality prediction approach are better than those of the 4 other referenced methods. (2) The multiple-feature fusing process can strongly affect the prediction performance. (3) Finally, the unsupervised pre-training process of the DBNs can enhance the information fusing ability. Finally, the newly proposed regression-based DBN approach may be extended to address other vehicle noises in the future.