基于可见光谱和改进注意力的农作物病害识别

基于可见光谱的农作物病害自动化识别和诊断是一个具有挑战性的研究领域,但现有基于卷积神经网络进行病害识别的研究往往利用深层网络牺牲模型参数量来提高对单一农作物病害识别的准确率,从而造成硬件资源的浪费.为提高农作物病害识别的准确率且避免深层网络的使用,该研究将注意力机制引入农作物病害识别领域,提出了一种基于可见光谱和改进注...     

图像场景识别中深度学习方法综述

场景识别是一种用计算机实现人的视觉功能的技术,它的研究目标是使计算机能够对图像或视频进行处理,自动识别和理解图像和视频中的场景信息。由于场景识别技术拥有广泛的应用前景,因此得到了许多关注。随着大数据时代的来临和深度学习的发展,使用深度学习方法解决场景识别问题已经成为场景识别领域未来的发展方向。文章首先概述介绍了场景识别技术的主要研究内容和发展情况,之后阐述了在图像场景识别中深度学习方法的应用情况,然后介绍了一些在图像场景识别中深度学习方法应用的具体的典型案例,同时给出了这几种方法具体的对比与分析。最后给出了文章的结论,总结了当前图像场景识别中使用深度学习方法的发展情况,并且对未来的发展方向给出了一些展望和建议。     

应用声信号特征识别车辆车型及车速范围

国内外现有研究工作主要针对车型进行识别,该文在此基础上对车速范围的分类识别进行了研究.文中将主成分分析方法和BP神经网络算法结合(PCA-BP)对车型与车速进行识别分析,对比了与BP神经网络算法识别速度的差异.结果表明应用PCA-BP方法的识别效果整体较好,且在识别过程中可以节省计算时长的50％～70％左右,大大提高了...     

基于密集连接时延神经网络的说话人识别算法*

说话人识别技术是一项重要的生物特征识别技术。近年来,使用深度神经网络提取发声特征的说话人识别算法取得了突出成果。时延神经网络作为其中的典型代表之一已被证明具有出色的特征提取能力。为进一步提升识别准确率并节约计算资源,通过对现有的说话人识别算法进行研究,提出一种带有注意力机制的密集连接时延神经网络用于说话人识别。密集连接的网络结构在增强不同网络层之间的信息复用的同时能有效控制模型体积。通道注意力机制和帧注意力机制帮助网络聚焦于更关键的细节特征,使得通过统计池化提取出的说话人特征更具有代表性。实验结果表明,在VoxCeleb1测试数据集上取得了1.40%的等错误率（EER）和0.15的最小检测代价标准（DCF）,证明了在说话人识别任务上的有效性。     

语音情感识别中的特征选择方法

为了解决传统卷积神经网络在识别中文语音时预测错误率较高、泛化性能弱的问题,首先以深度卷积神经网络(DCNN)-连接时序分类(CTC)为研究对象,深入分析了不同卷积层、池化层以及全连接层的组合对其性能的影响;其次,在上述模型的基础上,提出了多路卷积神经网络(MCNN)-连接时序分类(CTC),并联合SENet提出了深度SE-MCNN-CTC声学模型,该模型融合了MCNN与SENet的优势,既能加强卷积神经网络的深层信息的传递、避免梯度问题,又可以对提取的特征图进行自适应重标定。最终实验结果表明:SE-MCNN-CTC相较于DCNN-CTC错误率相对降低13.51%,模型最终的错误率达22.21%;算法改进后的声学模型可以有效地提升泛化性能。     

基于桥接Transformer的小样本优化鸟声识别网络

针对实际鸟类监测环境中,收集鸟声声频数据分布不均匀,导致神经网络训练不充分,分类识别测试准确率低的问题,设计了一种桥接Transformer神经网络模型。该网络首先利用原始鸟声声频信号生成短时傅里叶变换语谱图作为输入特征,之后将语谱图输入到由注意力模块和卷积模块桥接组成的Transformer网络中,完成对语谱图中全局特征和局部特征的信息交互,最后利用单层Transformer编码器实现对每一个批次样本的损失优化,得到最终的分类结果。在Birdsdata和xeno-canto鸟声数据集上进行小样本实验,分别获得了91.34%和82.63%的平均准确率,与其他鸟声识别网络进行了对比实验,验证了该网络的有效性。     

基于GAN的半监督遥感图像场景分类

针对遥感图像背景复杂及有监督场景分类算法无法利用无标签数据的问题,提出一种基于生成对抗网络的半监督遥感图像场景分类方法.首先,引入谱归一化残差块代替传统生成对抗网络中的二维卷积,利用残差块的跳跃连接解决梯度消失问题;其次,引入特征融合思想,将浅层特征与深层特征进行融合,从而减少特征损失;最后,在生成对抗网络的判别器中加...     

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Mel频谱值和深度学习网络的鸟声识别算法*

为了增强网络对鸟鸣声信号的特征学习能力并提高识别精度,提出一种基于深度残差收缩网络和扩张卷积的鸟声识别方法。首先,提取鸟鸣声信号的对数梅尔特征及其一阶和二阶差分系数组成logMel特征集作为网络模型的输入;其次,通过深度残差收缩网络自动学习噪声阈值,减少噪声干扰;然后,引入扩张卷积增大卷积核感受野并利用注意力机制使网络更关注关键帧特征;最后,通过双向长短时记忆网络从学到的局部特征中学习长期依赖关系。以百鸟数据birdsdata鸟声库中的19种中国常见鸟类作为实验对象,识别正确率可以达到96.58%,并对比模型在不同信噪比数据下的识别结果,结果表明该模型在噪声环境下的识别效果优于现有模型。     

基于注意力卷积神经网络的动态手势识别研究北大核心CSCD

针对复杂环境下动态手势识别准确率低的问题,提出了一种基于长短期记忆网络和卷积神经网络的动态手势识别算法。采用长短期记忆网络学习每个滤波器的权重,预测人体外形相关的滤波器组;采用卷积神经网络提取目标手势的轨迹图,创建彩色的轨迹图像;将轨迹图像送入注意力卷积神经网络训练,利用神经网络识别出复杂环境下的手势。实验结果表明,该算法能够准确地检测与跟踪手势的动态变化,并且实现了较好的手势识别准确性。     

An effective recognition method for network attack

With the development of the network, the network security has become the focus problem. How to guarantee the privacy, integrity, and availability of network information also becomes the issue to solve. This paper applies plan recognition method to recognize the network attack. First, a network attack recognition model has been built, which is more clearly to observe and recognize the process of network attack, and it is the foundation for the next work. Secondly, the temporal constraints has been added in the causal network, it can provide the help for alert correlation analysis, analyze the attack planning more effectively, predict the next action, and recognize the invalid planning. Finally, we put forward an effective method for network attack recognition based on the goal graph.     

An effective scaling frequency factor method for scaling of harmonic vibrational frequencies: The use of redundant primitive coordinates

A modified effective scaling frequency factor (ESFF) method that takes advantage of the potential energy distribution (PED) coefficients calculated in the basis of redundant primitive internal coordinates is presented. This approach is simpler and more flexible than that based on the natural internal coordinates. The sets of optimal scaling factors for routine 9- and 11-parameter ESFF calculations based on B3LYP/6-311G∗∗ force fields are derived from Baker’s training set of 30 molecules (660 frequencies). The calculated root-mean-square (RMS) deviations for all frequencies are 11.42 and 11.44 cm⁻¹ for 9- and 11-parameter scaling, respectively. They are somewhat lower than in the case of ordinary ESFF calculations. The new sets of factors seem to be particularly well suited for scaling of frequencies in the middle region of the vibrational spectra (1000-2500 cm⁻¹) - the RMS values in this range are 8.37 for 9-, and 9.56 cm⁻¹ for 11-parameter scaling. These values are to be compared with 9.20 and 10.29 cm⁻¹, respectively, calculated within the natural coordinates based ESFF formalism.     

An interpretable intrusion detection method based on few-shot learning in cloud-ground interconnection

An enterprise’s private cloud may be attacked by attackers when communicating with the public cloud. Although traffic detection methods based on deep learning have been widely used, these methods rely on a large amount of sample data and cannot quickly detect new attacks such as Zero-day Attacks. Moreover, deep learning has a black-box nature and cannot interpret the detection results, which has certain security risks. This paper proposes an interpretable abnormal traffic detection method, which can complete the detection task with only a few malicious traffic samples. Specifically, it uses the covariance matrix to characterize each traffic category and then calculates the similarity between the query traffic and each category according to the covariance metric function to realize the traffic detection based on few-shot learning. After that, the traffic images processed by the random masks are input into the model to obtain the predicted probability of the corresponding traffic category. Finally, the predicted probability is linearly summed with each mask to generate the final saliency map to interpret and analyze the model decision. In this paper, experiments are carried out by simulating only 15 and 25 malicious traffic samples. The results show that the proposed method can obtain good accuracy and recall, and the interpretation analysis shows that the model is reliable and interpretable.     

An effective cluster-based model for robust speech detection and speech recognition in noisy environments

This paper shows an accurate speech detection algorithm for improving the performance of speech recognition systems working in noisy environments. The proposed method is based on a hard decision clustering approach where a set of prototypes is used to characterize the noisy channel. Detecting the presence of speech is enabled by a decision rule formulated in terms of an averaged distance between the observation vector and a cluster-based noise model. The algorithm benefits from using contextual information, a strategy that considers not only a single speech frame but also a neighborhood of data in order to smooth the decision function and improve speech detection robustness. The proposed scheme exhibits reduced computational cost making it adequate for real time applications, i.e., automated speech recognition systems. An exhaustive analysis is conducted on the AURORA 2 and AURORA 3 databases in order to assess the performance of the algorithm and to compare it to existing standard voice activity detection (VAD) methods. The results show significant improvements in detection accuracy and speech recognition rate over standard VADs such as ITU-T G.729, ETSI GSM AMR, and ETSI AFE for distributed speech recognition and a representative set of recently reported VAD algorithms.     

An effective method for reducing speckle noise in digital holography

An effective method for reducing the speckle noise in digital holography is proposed in this paper.Different from the methods based on classical filtering technique,it utilizes the multiple holograms which are generated by rotating the illuminating light continuously.The intensity images reconstructed by a series of holograms generated by rotating the illuminating light possess different speckle patterns.Hence by properly averaging the reconstructed intensity fields,the speckle noises can be reduced greatly.Experimental results show that the proposed method is simple and effective to reduce speckle noise in digital holography.     

An effective method in calibrating nonlinear errors for phase-shifting interferometry

In phase measurement or digital holography for phase-shifting interferometry, the key role is the variation of reference light wave and recover algorithm based on interferograms and reference phase, so the calculation result is directly affected by phase-shift accuracy. However, because of the errors of nonlinear and other random factors, it is difficult to control the actual phase-shifting amount accurately. In this paper, we aim to propose an efficient method for phase-shifting interferometry which does not require accurate initial estimation of phase-shift amounts, only a few pixels with several randomly shifted interferograms are sufficient for accurate extraction of phase information. This method has reduced the dependence of reference phase, and can obtain phase-shifting amount directly without using complex revised algorithm for correcting phase-shifting nonlinear errors.     

An automatic and effective tooth isolation method for dental radiographs

Tooth isolation is a very important step for both computer-aided dental diagnosis and automatic dental identification systems, because it will directly affect the accuracy of feature extraction and, thereby, the final results of both types of systems. This paper presents an effective and fully automatic tooth isolation method for dental X-ray images, which contains up-per-lower jaw separation, single tooth isolation, over-segmentation verification, and under-segmentation detection. The upper-lower jaw separation mechanism is based on a gray-scale integral projection to avoid possible information loss and incorporates with the angle adjustment to handle skewed images. In a single tooth isolation, an adaptive windowing scheme for locating gap valleys is proposed to improve the accuracy. In over-segmentation, an isolation-curve verification scheme is proposed to remove excessive curves; and in under-segmentation, a missing-teeth detection scheme is proposed. The experimental results demonstrate that our method achieves the accuracy rates of 95.63% and 98.71% for the upper and lower jaw images, respectively, from the test database of 60 bitewing dental radiographs, and performs better for images with severe teeth occlusion, excessive dental works, and uneven illumination than that of Nomir and Abdel-Mottaleb’s method. The method without upper-lower jaw separation step also works well for panoramic and periapical images.     

Security concerns on machine learning solutions for 6G networks in mmWave beam prediction

6G – sixth generation – is the latest cellular technology currently under development for wireless communication systems. In recent years, machine learning (ML) algorithms have been applied widely in various fields, such as healthcare, transportation, energy, autonomous cars, and many more. Those algorithms have also been used in communication technologies to improve the system performance in terms of frequency spectrum usage, latency, and security. With the rapid developments of ML techniques, especially deep learning (DL), it is critical to consider the security concern when applying the algorithms. While ML algorithms offer significant advantages for 6G networks, security concerns on artificial intelligence (AI) models are typically ignored by the scientific community so far. However, security is also a vital part of AI algorithms because attackers can poison the AI model itself. This paper proposes a mitigation method for adversarial attacks against proposed 6G ML models for the millimeter-wave (mmWave) beam prediction using adversarial training. The main idea behind generating adversarial attacks against ML models is to produce faulty results by manipulating trained DL models for 6G applications for mmWave beam prediction. We also present a proposed adversarial learning mitigation method’s performance for 6G security in mmWave beam prediction application a fast gradient sign method attack. The results show that the defended model under attack’s mean square errors (i.e., the prediction accuracy) are very close to the undefended model without attack.     

一种高效量子密钥分发系统主动相位补偿方法

针对相位编码量子密钥分发系统相位漂移的实际问题,详细分析了目前解决相位漂移的主要方案,提出了一种"五点法"快速相位漂移参数的扫描方法.该方法只需对五个相位点进行单光子水平的相位扫描,即可得出满足精度要求的相位漂移参数.通过将该方法和其他两种主要相位补偿方法的对比分析,表明该方法可以在更短的扫描时间内有效得到量子密钥分发的相位漂移参数并对相位漂移进行实时补偿.该方法适用于目前常用的相位编码系统,为量子密码系统提供了一种有实际应用价值的主动相位补偿方案.     

水声JANUS信号的分数低阶时频谱迁移学习识别方法

非合作第三方水下标准协议信号识别在水声通信信号识别中具有重要研究意义。针对浅海水声JANUS信号的特征提取因易受脉冲噪声和多径效应等复杂水声环境影响而导致识别率低下的问题,提出一种分数低阶时频谱和ResNet18 （Residual Network 18）相结合的迁移学习识别方法。首先,选取JANUS固定前导作为识别对象,设计分数低阶傅里叶同步压缩变换（FLOFSST）,以分数低阶操作抑制脉冲噪声,以时频重排特性增强时频集中性。其次,将基于ImageNet的ResNet18预训练模型微调,迁移至JANUS信号和常见水声信号时频图集。仿真表明所提算法在信噪比为-10 dB时JANUS信号的识别率为96.15%,能够有效抑制脉冲噪声并减小多径效应影响,比传统算法识别性能好。海试中JANUS信号识别率达90.00%,证明算法识别准确率和网络的泛化性较高。