Android平台下的网络机顶盒USB-OTG模块驱动设计

目前,网络机顶盒的市场越来越大,现有的产品功能比较繁杂,为实现功能的可定制化,增强用户与机顶盒间的交互能力,文章中的设计在基于Android系统和Cortex-A7架构主控的网络机顶盒基础上加入了USB-OTG模块。文章介绍了USB子系统的整体框架,并对设备端驱动和主从切换部分等关键技术做了详细描述,最后实验验证了此方案的可行性。     

多尺度YOLO人脸年龄估计方法研究

通过观察人脸估计年龄较为常见，但如何准确预测年龄则是一个难题。为提高人脸图像年龄估计的准确率，提出一种基于YOLO（You Only Look Once）模型的目标检测方法。将多尺度回归思想应用于卷积神经网络（Convolutional Neural Network，CNN），通过多尺度卷积改善模型对小尺寸目标的提取能力，结合特征通道分权重思想，改善特征提取操作中特征信息丢失的问题，构造决策树回归得到年龄估计。这种方法在人脸年龄图像库FG-NET上获得平均绝对误差（MAE）3.43，在GROUP数据集获得区间匹配度（AEM）62.4%。实验结果表明，通过多尺度特征回归以及通道权重分配，可以较为准确地进行人脸信息检测，并由此建立鲁棒性更强的人脸年龄估计模型。     

基于多尺度多任务卷积神经网络的人群计数

在智能监控领域，实现人群计数具有重要价值，针对人群尺度不一、人群密度分布不均及遮挡等问题，提出一种多尺度多任务卷积神经网络（MMCNN）进行人群计数的方法。首先提出一种新颖的自适应人形核生成密度图描述人群信息，消除人群遮挡影响；其次通过构建多尺度卷积神经网络解决人群尺度不一问题，以多任务学习机制同时估计密度图及人群密度等级，解决人群分布不均问题；最后设计一种加权损失函数，提高人群计数准确率。在UCF_CC_50和World Expo'10数据库上进行了评估，验证了自适应人形核的有效性。实验结果表明：所提算法比Sindagi等的方法（SINDAGI V A，PATEL V M.CNN-based cascaded multi-task learning of high-level prior and density estimation for crowd counting.Proceedings of the 2017 14th IEEE International Conference on Advanced Video and Signal Based Surveillance.Piscataway，NJ：IEEE，2017：1-6）在UCF_CC_50数据库上平均绝对误差（MAE）数值和均方误差（MSE）数值分别降低约1.7和45；与Zhang等的方法（ZHANG Y，ZHOU D，CHEN S，et al.Single-image crowd counting via multi-column convolutional neural network.Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition.Washington，DC：IEEE Computer Society，2016：589-597）相比，在World Expo'10数据库上所提算法的MAE值降低约1.5，且在真实公共汽车数据库上仅0~3人的计数误差，表明其实用性较强。     

An improved machine learning technique based on downsized KPCA for Alzheimer's disease classification

Alzheimer's disease (AD), a neurodegenerative disorder, is a very serious illness that cannot be cured, but the early diagnosis allows precautionary measures to be taken. The current used methods to detect Alzheimer's disease are based on tests of cognitive impairment, which does not provide an exact diagnosis before the patient passes a moderate stage of AD. In this article, a novel classifier of brain magnetic resonance images (MRI) based on the new downsized kernel principal component analysis (DKPCA) and multiclass support vector machine (SVM) is proposed. The suggested scheme classifies AD MRIs. First, a multiobjective optimization technique is used to determine the optimal parameter of the kernel function in order to ensure good classification results and to minimize the number of retained principle components simultaneously. The optimal parameter is used to build the optimized DKPCA model. Second, DKPCA is applied to normalized features. Downsized features are then fed to the classifier to output the prediction. To validate the effectiveness of the proposed method, DKPCA was tested using synthetic data to demonstrate its efficiency on dimensionality reduction, then the DKPCA based technique was tested on the OASIS MRI database and the results were satisfactory compared to conventional approaches.     

基于KJADE的列车轴承轨边声学诊断方法研究

为在线诊断运行列车的轴承状态,提出一种基于核特征矩阵联合近似对角化(kernel joint approximate diagonalization of eigen-matrices,简称KJADE)的列车轴承轨边声学故障诊断方法。首先,从校正后的轨边信号中提取原始特征,将其通过非线性映射函数映射到高维特征空间;其次,对特征空间的核矩阵进行四阶累积量的特征分解,获得新融合特征,并采用支持向量机分类器对融合特征进行辨识;最后,对轴承外圈、内圈、滚子故障和正常4种状态下的列车轨边声学信号进行分析。结果表明,该方法可以有效实现对列车轴承轨边声音信号的非线性特征提取,提高了故障的识别率。     

Detection of K in soil using time-resolved laser-induced breakdown spectroscopy based on convolutional neural networks

One of the technical bottlenecks of traditional laser-induced breakdown spectroscopy (LIBS) is the difficulty in quantitative detection caused by the matrix effect. To troubleshoot this problem, this paper investigated a combination of time-resolved LIBS and convolutional neural networks (CNNs) to improve K determination in soil. The time-resolved LIBS contained the information of both wavelength and time dimension. The spectra of wavelength dimension showed the characteristic emission lines of elements, and those of time dimension presented the plasma decay trend. The one-dimensional data of LIBS intensity from the emission line at 766.49 nm were extracted and correlated with the K concentration, showing a poor correlation of R²_c=0.0967, which is caused by the matrix effect of heterogeneous soil. For the wavelength dimension, the two-dimensional data of traditional integrated LIBS were extracted and analyzed by an artificial neural network (ANN), showing R²_v=0.6318 and the root mean square error of validation (RMSEV)=0.6234. For the time dimension, the two-dimensional data of time-decay LIBS were extracted and analyzed by ANN, showing R²_v=0.7366 and RMSEV=0.7855. These higher determination coefficients reveal that both the non-K emission lines of wavelength dimension and the spectral decay of time dimension could assist in quantitative detection of K. However, due to limited calibration samples, the two-dimensional models presented over-fitting. The three-dimensional data of time-resolved LIBS were analyzed by CNNs, which extracted and integrated the information of both the wavelength and time dimension, showing the R²_v=0.9968 and RMSEV=0.0785. CNN analysis of time-resolved LIBS is capable of improving the determination of K in soil.     

Hybrid fully convolutional networks-based skin lesion segmentation and melanoma detection using deep feature

Fully convolutional networks (FCNs) take the input of arbitrary size and produce correspondingly sized output with efficient inference and learning. The automatic diagnosis of melanoma is very essential for reducing the mortality rate by identifying the disease in earlier stages. A two-stage framework is used for implementing the melanoma detection, segmentation of skin lesion, and identification of melanoma lesions. Two FCNs based on VGG-16 and GoogLeNet are incorporated for improving the segmentation accuracy. A hybrid framework is used for incorporating these two FCNs. The classification is done by extracting the feature from segmented lesion by using deep residual network and a hand-crafted feature. Classification is done by support vector machine. The performance analysis of our framework gives a promising accuracy, that is, 0.8892 for classification in ISBI 2016 dataset and 0.853 for ISIC 2017 dataset.     

无人机影像目标检测技术研究进展

无人机影像目标检测技术是无人机影像智能化分析中的关键技术，开展无人机影像目标检测技术的研究有着广阔的应用前景和价值。介绍了无人机影像中目标检测技术的发展历程，简述了无人机影像目标检测技术在森林防火、农业信息采集、电力线路检测、路桥损害检测以及军事侦察等5种特定领域的应用情况，总结归纳了目标检测性能评价时常用的数据集和标准，并对未来无人机影像目标检测技术的发展态势进行了展望。     

基于改进词嵌入表示和卷积神经网络的银行智能派单系统的研究

银行智能派单系统的实现和功能完善,对银行提升客户满意度、提高突发事件处理效率、降低人工处理成本等非常重要。针对现有的基于Word2vec和TextCNN模型的银行智能派单系统进行了改进,针对特征词权重表达性弱,特征词类别及位置区分性弱等问题,提出基于改进TF-IDF加权的Word2vec词嵌入表示和卷积神经网络结合的银行智能派单系统:首先利用Word2vec模型得到输入事件单的词嵌入向量;再针对经典TF-IDF方法不具备类别区分性、位置区分性,也没有考虑极端频率特征词代表性的情况,提出改进型TF-IDF算法,计算每个特征词的权重,得到基于改进TF-IDF加权的Word2vec词嵌入表示;最后在卷积神经网络模型中进行训练,通过迭代训练最终得到分类器,利用分类器可对输入事件单信息自动进行系统类别的判断。实验结果表明改进词嵌入表示的银行智能派单系统分类模型的宏查准率、宏查全率、准确率以及宏F1值都得到进一步的提高。     

Convolutional and generative adversarial neural networks in manufacturing

Manufacturing is undergoing transformation driven by the developments in process technology, information technology, and data science. A future manufacturing enterprise will be highly digital. This will create opportunities for machine learning algorithms to generate predictive models across the enterprise in the spirit of the digital twin concept. Convolutional and generative adversarial neural networks have received some attention of the manufacturing research community. Representative research and applications of the two machine learning concepts in manufacturing are presented. Advantages and limitations of each neural network are discussed. The paper might be helpful in identifying research gaps, inspire machine learning research in new manufacturing domains, contribute to the development of successful neural network architectures, and getting deeper insights into the manufacturing data.