Fast recognition using convolutional neural network for the coal particle density range based on images captured under multiple light sources

A method based on multiple images captured under different light sources at different incident angles was developed to recognize the coal density range in this study.The innovation is that two new images were constructed based on images captured under four single light sources.Reconstruction image 1 was constructed by fusing greyscale versions of the original images into one image,and Reconstruction image2 was constructed based on the differences between the images captured under the different light sources.Subsequently,the four original images and two reconstructed images were input into the convolutional neural network AlexNet to recognize the density range in three cases:-1.5(clean coal) and+1.5 g/cm^(3)(non-clean coal);-1.8(non-gangue) and+1.8 g/cm^(3)(gangue);-1.5(clean coal),1.5-1.8(middlings),and+1.8 g/cm^(3)(gangue).The results show the following:(1) The reconstructed images,especially Reconstruction image 2,can effectively improve the recognition accuracy for the coal density range compared with images captured under single light source.(2) The recognition accuracies for gangue and non-gangue,clean coal and non-clean coal,and clean coal,middlings,and gangue reached88.44%,86.72% and 77.08%,respectively.(3) The recognition accuracy increases as the density moves further away from the boundary density.

Advanced Feature Fusion Algorithm Based on Multiple Convolutional Neural Network for Scene Recognition

Scene recognition is a popular open problem in the computer vision field.Among lots of methods proposed in recent years,Convolutional Neural Network(CNN)based approaches achieve the best performance in scene recognition.We propose in this paper an advanced feature fusion algorithm using Multiple Convolutional Neural Network(Multi-CNN)for scene recognition.Unlike existing works that usually use individual convolutional neural network,a fusion of multiple different convolutional neural networks is applied for scene recognition.Firstly,we split training images in two directions and apply to three deep CNN model,and then extract features from the last full-connected(FC)layer and probabilistic layer on each model.Finally,feature vectors are fused with different fusion strategies in groups forwarded into SoftMax classifier.Our proposed algorithm is evaluated on three scene datasets for scene recognition.The experimental results demonstrate the effectiveness of proposed algorithm compared with other state-of-art approaches.

Uplink NOMA signal transmission with convolutional neural networks approach

Non-orthogonal multiple access(NOMA), featuring high spectrum efficiency, massive connectivity and low latency, holds immense potential to be a novel multi-access technique in fifth-generation(5G) communication. Successive interference cancellation(SIC) is proved to be an effective method to detect the NOMA signal by ordering the power of received signals and then decoding them. However, the error accumulation effect referred to as error propagation is an inevitable problem. In this paper,we propose a convolutional neural networks(CNNs) approach to restore the desired signal impaired by the multiple input multiple output(MIMO) channel. Especially in the uplink NOMA scenario,the proposed method can decode multiple users' information in a cluster instantaneously without any traditional communication signal processing steps. Simulation experiments are conducted in the Rayleigh channel and the results demonstrate that the error performance of the proposed learning system outperforms that of the classic SIC detection. Consequently, deep learning has disruptive potential to replace the conventional signal detection method.

Non-Intrusive Load Identification Model Based on 3D Spatial Feature and Convolutional Neural Network

Load identification method is one of the major technical difficulties of non-intrusive composite monitoring. Binary V-I trajectory image can reflect the original V-I trajectory characteristics to a large extent, so it is widely used in load identification. However, using single binary V-I trajectory feature for load identification has certain limitations. In order to improve the accuracy of load identification, the power feature is added on the basis of the binary V-I trajectory feature in this paper. We change the initial binary V-I trajectory into a new 3D feature by mapping the power feature to the third dimension. In order to reduce the impact of imbalance samples on load identification, the SVM SMOTE algorithm is used to balance the samples. Based on the deep learning method, the convolutional neural network model is used to extract the newly produced 3D feature to achieve load identification in this paper. The results indicate the new 3D feature has better observability and the proposed model has higher identification performance compared with other classification models on the public data set PLAID.

Hybrid Deep Learning-Based Adaptive Multiple Access Schemes Underwater Wireless Networks

Achieving sound communication systems in Under Water Acoustic(UWA)environment remains challenging for researchers.The communication scheme is complex since these acoustic channels exhibit uneven characteristics such as long propagation delay and irregular Doppler shifts.The development of machine and deep learning algorithms has reduced the burden of achieving reli-able and good communication schemes in the underwater acoustic environment.This paper proposes a novel intelligent selection method between the different modulation schemes such as Code Division Multiple Access(CDMA),Time Divi-sion Multiple Access(TDMA),and Orthogonal Frequency Division Multiplexing(OFDM)techniques using the hybrid combination of the convolutional neural net-works(CNN)and ensemble single feedforward layers(SFL).The convolutional neural networks are used for channel feature extraction,and boosted ensembled feedforward layers are used for modulation selection based on the CNN outputs.The extensive experimentation is carried out and compared with other hybrid learning models and conventional methods.Simulation results demonstrate that the performance of the proposed hybrid learning model has achieved nearly 98%accuracy and a 30%increase in BER performance which outperformed the other learning models in achieving the communication schemes under dynamic underwater environments.

MSSTNet:Multi-scale facial videos pulse extraction network based on separable spatiotemporal convolution and dimension separable attention

Background The use of remote photoplethysmography(rPPG)to estimate blood volume pulse in a noncontact manner has been an active research topic in recent years.Existing methods are primarily based on a singlescale region of interest(ROI).However,some noise signals that are not easily separated in a single-scale space can be easily separated in a multi-scale space.Also,existing spatiotemporal networks mainly focus on local spatiotemporal information and do not emphasize temporal information,which is crucial in pulse extraction problems,resulting in insufficient spatiotemporal feature modelling.Methods Here,we propose a multi-scale facial video pulse extraction network based on separable spatiotemporal convolution(SSTC)and dimension separable attention(DSAT).First,to solve the problem of a single-scale ROI,we constructed a multi-scale feature space for initial signal separation.Second,SSTC and DSAT were designed for efficient spatiotemporal correlation modeling,which increased the information interaction between the long-span time and space dimensions;this placed more emphasis on temporal features.Results The signal-to-noise ratio(SNR)of the proposed network reached 9.58dB on the PURE dataset and 6.77dB on the UBFC-rPPG dataset,outperforming state-of-the-art algorithms.Conclusions The results showed that fusing multi-scale signals yielded better results than methods based on only single-scale signals.The proposed SSTC and dimension-separable attention mechanism will contribute to more accurate pulse signal extraction.

Convolutional neural network adaptation and optimization method in SIMT computing mode

For studying and optimizing the performance of general-purpose computing on graphics processing units(GPGPU)based on single instruction multiple threads(SIMT)processor about the neural network application,this work contributes a self-developed SIMT processor named Pomelo and correlated assembly program.The parallel mechanism of SIMT computing mode and self-developed Pomelo processor is briefly introduced.A common convolutional neural network(CNN)is built to verify the compatibility and functionality of the Pomelo processor.CNN computing flow with task level and hardware level optimization is adopted on the Pomelo processor.A specific algorithm for organizing a Z-shaped memory structure is developed,which addresses reducing memory access in mass data computing tasks.Performing the above-combined adaptation and optimization strategy,the experimental result demonstrates that reducing memory access in SIMT computing mode plays a crucial role in improving performance.A 6.52 times performance is achieved on the 4 processing elements case.

基于PCA-VMD-CNN的输电线路覆冰重量预测模型

为防止覆冰灾害危及电路安全,提出1种输电线路覆冰重量预测模型。首先对多个气象因素进行主成分分析提取气象因素中的有效信息,再对覆冰历史数据进行变分模态分解,获得具有不同特性的本征模态分量;然后基于卷积神经网络,对具有不同时间尺度(周期性、波动性不同)的各个分量进行训练及预测,并将每个分量的预测结果相加。研究结果表明:通过对某覆冰区域的输电线路监测数据进行实验仿真,研究所提出的覆冰重量预测模型有更高精度。

Application of LSTM and CONV1D LSTM Network in Stock Forecasting Model

Predicting the direction of the stock market has always been a huge challenge.Also,the way of forecasting the stock market reduces the risk in the financial market,thus ensuring that brokers can make normal returns.Despite the complexities of the stock market,the challenge has been increasingly addressed by experts in a variety of disciplines,including economics,statistics,and computer science.The introduction of machine learning,in-depth understanding of the prospects of the financial market,thus doing many experiments to predict the future so that the stock price trend has different degrees of success.In this paper,we propose a method to predict stocks from different industries and markets,as well as trend prediction using traditional machine learning algorithms such as linear regression,polynomial regression and learning techniques in time series prediction using two forms of special types of recursive neural networks:long and short time memory(LSTM)and spoken short-term memory.

基于遥感多参数和CNN-Transformer的冬小麦单产估测

为了提高冬小麦单产估测精度,改善估产模型存在的高产低估和低产高估等现象,以陕西省关中平原为研究区域,选取旬尺度条件植被温度指数(VTCI)、叶面积指数(LAI)和光合有效辐射吸收比率(FPAR)为遥感特征参数,结合卷积神经网络(CNN)局部特征提取能力和基于自注意力机制的Transformer网络的全局信息提取能力,构建CNN-Transformer深度学习模型,用于估测关中平原冬小麦产量。与Transformer模型(R^(2)为0.64,RMSE为465.40 kg/hm^(2),MAPE为8.04%)相比,CNN-Transformer模型具有更高的冬小麦单产估测精度(R^(2)为0.70,RMSE为420.39 kg/hm^(2),MAPE为7.65%),能够从遥感多参数中提取更多与产量相关的信息,且对于Transformer模型存在的高产低估和低产高估现象均有所改善。基于5折交叉验证法和留一法进一步验证了CNN-Transformer模型的鲁棒性和泛化能力。此外,基于CNN-Transformer模型捕获冬小麦生长过程的累积效应,分析逐步累积旬尺度输入参数对产量估测的影响,评估模型对于冬小麦不同生长阶段的累积过程的表征能力。结果表明,模型能有效捕捉冬小麦生长的关键时期,3月下旬至5月上旬是冬小麦生长的关键时期。

基于双分支并联的特征融合电能质量扰动分类方法

为了提高对电能质量扰动信号(power quality disturbance signal,PQDs)在受到噪声和异常数据干扰时的分类准确率,提出了一种双分支并联特征融合网络的PQDs分类方法。首先,采用一维残差神经网络和一维卷积神经网络两个分支进行特征提取。然后,通过特征融合模块将这些特征融合在一起。最终,通过分类模块对PQDs进行准确分类。相对于串联神经网络,所提方法融合特征向量,增强了特征的区分度,同时适用于并行计算,进一步提高了识别速度。仿真结果表明,所提方法在叠加信噪比为13 dB、15 dB和18 dB的PQDs分类任务中,识别率均超过95%,此外,该方法对异常数据的分类效果也具有一定的鲁棒性。

基于卷积神经网络的多工况多传感滚动轴承实时监控方法

针对工业环境中广泛在多工况下多滚动轴承实时状态监测的需求和部署环境受限的挑战,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)的面向多传感器滚动轴承运行状态监控方法。该方法将两个不同工况下的一维时间序列数据集以均方根(Root Mean Square,RMS)指标标注,并通过将一维时间序列多传感器数据重构为二维空间张量的形式输入卷积神经网络训练。最后利用层融合和16比特量化优化,将网络部署到FPGA上,用以解决CNN的计算开销。实验结果表明,在结合了两种不同工况的数据集下,网络测试推理准确度依然高达99.24%,比多层感知机实现高10.48%,比多层感知机结合支持向量机的实现高2.91%,该算法对于新加入的数据集也有较强的鲁棒性,经过重训练,新加入的数据集准确率可以达到99.17%。基于FPGA部署优化的网络的峰值能效为76.217GPOS/W,为CPU实现的33.09倍,GPU实现的5.39倍。其中,16比特精度部署的网络测试精度相较32比特精度实现仅降低0.001%。

融合CNN与Transformer的MRI脑肿瘤图像分割

为解决卷积神经网络(convolutional neural network,CNN)在学习全局上下文信息和边缘细节方面受到很大限制的问题,提出一种同时学习局语义信息和局部空间细节的级联神经网络用于脑肿瘤医学图像分割。首先将输入体素分别送入CNN和Transformer分支,在编码阶段结束后,采用一种双分支融合模块将2个分支学习到的特征有效地结合起来以实现全局信息与局部信息的融合。双分支融合模块利用哈达玛积对双分支特征之间的细粒度交互进行建模,同时使用多重注意力机制充分提取特征图通道和空间信息并抑制无效的噪声信息。在BraTS竞赛官网评估了本文方法,在BraTS2019验证集上增强型肿瘤区、全肿瘤区和肿瘤核心区的Dice分数分别为77.92%,89.20%和81.20%。相较于其他先进的三维医学图像分割方法,本文方法表现出了更好的分割性能,为临床医生做出准确的脑肿瘤细胞评估和治疗方案提供了可靠依据。

基于时序生成对抗网络的居民用户非侵入式负荷分解

现有的非侵入式负荷分解算法往往需要大量电器设备级的负荷数据才能保证分解精度,但由于用户对隐私性的考虑以及安装成本过高等问题,很难获取这些数据。因此,构建一种能深度挖掘电力负荷数据时序特性和电器相关性的时序生成对抗网络。利用降维网络对所有电器有功功率序列组成的高维向量进行降维以降低计算的复杂度,通过复原网络将结果还原为高维向量。基于电器运行状态和深度学习的非侵入式分解方法,运用卷积神经网络-双向门控循环单元构建状态复杂电器的负荷分解回归模型,对状态简单电器利用深度神经网络构建负荷识别分类模型。通过对比其他数据生成方法,以及改变典型公开数据集中生成数据比例所得的负荷分解结果验证了所提方法的有效性。

基于深度学习的视频异常检测研究综述

视频异常检测涉及概率统计、机器学习和深度学习等方法.文中旨在综合作者课题组研究成果和其它前沿科研工作,聚焦于基于深度学习的视频异常检测方法,全面探讨该领域的背景、挑战与解决方案.综合领域内的大多数相关论文,对其进行系统分析,以期为学者提供现阶段研究进展的基础认知.对基于深度学习的视频异常检测方法进行分类、分析,总结各类方法的网络模型选择,详细介绍常用数据集和性能评价指标,以性能对比突显各类方法的优势,并对视频异常检测领域的未来研究方向和应用场景进行深入探讨和展望.

基于多光谱遥感和CNN的玉米地上生物量估算模型

目前玉米地上生物量(Aboveground biomass,AGB)的预测方法集中在使用从无人机图像中提取光学植被指数,通过线性模型或机器学习算法与AGB建立关系,原始图像信息损失严重,玉米生长后期的饱和效应会严重降低模型精度。针对此问题,本文收集了玉米拔节期、吐丝期和乳熟期的无人机图像和地面数据。分析了不同生育期玉米干地上生物量、鲜地上生物量与8个植被指数(Vegetation indexes,VIs)之间的相关性。分别以最优植被指数作为输入建立多层感知机(Multilayer perceptron,MLP)模型、以无人机多光谱图像作为输入建立卷积神经网络(Convolutional neural network,CNN)模型来估算玉米干地上生物量、鲜地上生物量。结果表明,基于MLP的玉米干地上生物量估算模型随着玉米生育期推进,模型的精度急剧下降,3个生长期MLP模型验证集R^(2)分别为0.65、0.23、0.32,RMSE分别为0.27、2.15、5.03 t/hm^(2)。CNN模型能够较好地克服光谱饱和问题,具有良好的精度和适用性,3个生育期验证集R^(2)分别提高27.69%、191.30%、171.88%,RMSE分别降低22.22%、38.14%、45.53%。基于MLP的玉米鲜地上生物量估算模型在玉米生长后期模型的精度同样较低,吐丝期、乳熟期验证集的R^(2)分别为0.27、0.37,RMSE分别为11.57、14.98 t/hm^(2)。CNN模型2个生育期验证集的R^(2)分别提高159.26%、129.73%,RMSE分别降低26.62%、54.01%。使用原始多光谱图像作为输入的CNN模型取得了最好的估计结果,可为玉米不同生育期的监测研究、精准管理提供指导。

脑电情感识别中多上下文向量优化的卷积递归神经网络

目前的脑电(EEG)情感识别模型忽略了不同时段情感状态的差异性,未能强化关键的情感信息。针对上述问题,提出一种多上下文向量优化的卷积递归神经网络(CR-MCV)。首先构造脑电信号的特征矩阵序列,通过卷积神经网络(CNN)学习多通道脑电的空间特征;然后利用基于多头注意力的递归神经网络生成多上下文向量进行高层抽象特征提取;最后利用全连接层进行情感分类。在DEAP(Database for Emotion Analysis using Physiological signals)数据集上进行实验,CR-MCV在唤醒和效价维度上分类准确率分别为88.09%和89.30%。实验结果表明,CR-MCV在利用电极空间位置信息和不同时段情感状态显著性特征基础上,能够自适应地分配特征的注意力并强化情感状态显著性信息。

基于深度学习的三维肿瘤及器官分割

针对三维医学图像中由于肿瘤或器官的形状、尺度差异较大导致分割精度较低的问题,提出一种端到端的三维全卷积分割模型。首先,设计空洞立方集成模块在不同分辨率阶段实现多尺度集成,增强复杂边界上的识别能力;其次,引入跨阶段上下文融合模块融合浅层和深层特征,促进收敛并更准确地定位目标对象;最后,解码器对来自编码器的特征进行拼接以实现分割。在脑肿瘤分割数据集上,平均Dice相似性系数值达到85.37%;在腹部器官分割数据集上,平均Dice相似性系数值达到83.99%。实验结果表明所提模型在三维肿瘤和器官的分割上具有较高精度。

基于高维多目标序贯三支决策的恶意代码检测模型

针对传统基于二支决策的恶意代码检测方法在面对动态环境中的复杂海量数据时,没有考虑在信息不充足条件下进行决策产生影响的问题,本文提出了一种基于卷积神经网络的序贯三支决策恶意代码检测模型。通过卷积神经网络对样本数据进行特征提取并构建多粒度特征集,引入序贯三支决策理论对恶意代码进行检测。为改善检测模型整体性能,避免阈值选取的主观性,本文在上述模型的基础上,同时考虑模型的综合分类性能、决策效率和决策风险代价建立高维多目标序贯三支决策模型,并采用高维多目标优化算法对模型进行求解。仿真结果表明,模型在保证检测性能的同时,有效地提升了决策效率,降低了决策时产生风险代价,更好地拟合了真实动态检测环境。

基于GADF-CWT-GCNN的滚动轴承故障诊断方法研究

针对滚动轴承故障诊断在小样本环境下引起的模型泛化能力差、诊断精度低的问题,提出一种基于格拉姆角分场(GADF)和连续小波变化(continuous wavelet transform,CWT)与并行二维组归一化卷积神经网络(parallel convolutional neural network,P2D-GCNN)的滚动轴承故障诊断方法。对采集的数据进行预处理,采用格拉姆角场和连续小波变换将一维振动信号转换成二维图像作为模型输入,再选用数据增强技术扩充样本子图,满足网络输入要求,并将其导入搭建的组归一化卷积神经网络中进行诊断检测。结果表明:文中数据处理方法与搭建模型在小样本环境下泛化能力远高于SVM和1D-CNN等其他网络模型。为进一步验证模型在小样本数据下的识别能力,取数据集的70%,40%和20%样本量进行多次实验,所对应的训练准确率及测试准确率分为99.38%,99.02%,99.47%,98.29%,99.05%,97.08%。结果证明,文中模型在小样本环境下对轴承故障诊断具有很高的准确率。