Effective distributed convolutional neural network architecture for remote sensing images target classification with a pre-training approach

How to recognize targets with similar appearances from remote sensing images(RSIs) effectively and efficiently has become a big challenge. Recently, convolutional neural network(CNN) is preferred in the target classification due to the powerful feature representation ability and better performance. However,the training and testing of CNN mainly rely on single machine.Single machine has its natural limitation and bottleneck in processing RSIs due to limited hardware resources and huge time consuming. Besides, overfitting is a challenge for the CNN model due to the unbalance between RSIs data and the model structure.When a model is complex or the training data is relatively small,overfitting occurs and leads to a poor predictive performance. To address these problems, a distributed CNN architecture for RSIs target classification is proposed, which dramatically increases the training speed of CNN and system scalability. It improves the storage ability and processing efficiency of RSIs. Furthermore,Bayesian regularization approach is utilized in order to initialize the weights of the CNN extractor, which increases the robustness and flexibility of the CNN model. It helps prevent the overfitting and avoid the local optima caused by limited RSI training images or the inappropriate CNN structure. In addition, considering the efficiency of the Na¨?ve Bayes classifier, a distributed Na¨?ve Bayes classifier is designed to reduce the training cost. Compared with other algorithms, the proposed system and method perform the best and increase the recognition accuracy. The results show that the distributed system framework and the proposed algorithms are suitable for RSIs target classification tasks.

A New Childhood Pneumonia Diagnosis Method Based on Fine-Grained Convolutional Neural Network

Pneumonia is part of the main diseases causing the death of children.It is generally diagnosed through chest Xray images.With the development of Deep Learning(DL),the diagnosis of pneumonia based on DL has received extensive attention.However,due to the small difference between pneumonia and normal images,the performance of DL methods could be improved.This research proposes a new fine-grained Convolutional Neural Network(CNN)for children’s pneumonia diagnosis(FG-CPD).Firstly,the fine-grainedCNNclassificationwhich can handle the slight difference in images is investigated.To obtain the raw images from the real-world chest X-ray data,the YOLOv4 algorithm is trained to detect and position the chest part in the raw images.Secondly,a novel attention network is proposed,named SGNet,which integrates the spatial information and channel information of the images to locate the discriminative parts in the chest image for expanding the difference between pneumonia and normal images.Thirdly,the automatic data augmentation method is adopted to increase the diversity of the images and avoid the overfitting of FG-CPD.The FG-CPD has been tested on the public Chest X-ray 2017 dataset,and the results show that it has achieved great effect.Then,the FG-CPD is tested on the real chest X-ray images from children aged 3–12 years ago from Tongji Hospital.The results show that FG-CPD has achieved up to 96.91%accuracy,which can validate the potential of the FG-CPD.

MOVING TARGETS PATTERN RECOGNITION BASED ON THE WAVELET NEURAL NETWORK

Based on pattern recognition theory and neural network technology, moving objects automatic detection and classification method integrating advanced wavelet analysis are discussed in detail. An algorithm of moving targets pattern recognition on the combination of inter-frame difference and wavelet neural network is presented. The experimental results indicate that the designed BP wavelet network using this algorithm can recognize and classify moving targets rapidly and effectively.

Fine-grained classification of grape leaves via a pyramid residual convolution neural network

The value of grape cultivars varies.The use of a mixture of cultivars can negate the benefits of improved cultivars and hamper the protection of genetic resources and the identification of new hybrid cultivars.Classifying cultivars based on their leaves is therefore highly practical.Transplanted grape seedlings take years to bear fruit,but leaves mature in months.Foliar morphology differs among cultivars,so identifying cultivars based on leaves is feasible.Different cultivars,however,can be bred from the same parents,so the leaves of some cultivars can have similar morphologies.In this work,a pyramid residual convolution neural network was developed to classify images of eleven grape cultivars.The model extracts multi-scale feature maps of the leaf images through the convolution layer and enters them into three residual convolution neural networks.Features are fused by adding the value of the convolution kernel feature matrix to enhance the attention on the edge and center regions of the leaves and classify the images.The results indicated that the average accuracy of the model was 92.26%for the proposed leaf dataset.The proposed model is superior to previous models and provides a reliable method for the fine-grained classification and identification of plant cultivars.

Generalization Capabilities of Feedforward Neural Networks for Pattern Recognition

This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.

A Survey on Deep Learning-based Fine-grained Object Classification and Semantic Segmentation

The deep learning technology has shown impressive performance in various vision tasks such as image classification, object detection and semantic segmentation. In particular, recent advances of deep learning techniques bring encouraging performance to fine-grained image classification which aims to distinguish subordinate-level categories, such as bird species or dog breeds. This task is extremely challenging due to high intra-class and low inter-class variance. In this paper, we review four types of deep learning based fine-grained image classification approaches, including the general convolutional neural networks （CNNs）, part detection based, ensemble of networks based and visual attention based fine-grained image classification approaches. Besides, the deep learning based semantic segmentation approaches are also covered in this paper. The region proposal based and fully convolutional networks based approaches for semantic segmentation are introduced respectively.

Network mechanism of effective constituents from the compound Yizhihao against influenza

Influenza caused by influenza virus,seriously threaten human life and health.Drug treatment is one of the effective measurement.However,there are only two classes of drugs,one class is M2 blockers and another is neuraminidase(NA) inhibitors.The recent antiviral surveillance studies reported a global significant increase in M2 blocker resistance among influenza viruses,and the resistant virus strains against NA inhibitor are also reported in clinical treatment.Therefore thediscovery of new medicines with low resistance has become very urgent.As all known,traditional medicines with multi-target features and network mechanism often possess low resistance.Compound Yizhihao,which consists of radix isatidis,folium isatidis,Artemisia rupestris,is one of the famous traditional medicine for influenza treatment in China,however its mechanism of action against influenza is unclear.In this study,the multiple targets related with influenza disease and the known chemical constituents from Compound Yizhihao were collected,and multi-target QSAR(mt-QSAR) classification models were developed by Na?e Bayesian algorithm and verified by various datasets.Then the classification models were applied to predict the effective constituents and their drug targets.Finally,the constituent-target-pathway network was constructed,which revealed the effective constituents and their network mechanism in Compound Yizhihao.This study will lay important basis for the clinical uses for influenza treatment and for the further research and development of the effective constituents.

基于卷积神经网络的逐级唤醒存内计算控制器设计

随着人工智能的发展,边缘端设备对智能图像处理的需求急剧增加。目前边缘端设备主要面临能量受限、吞吐率较低等问题。无人机在侦察过程中,根据任务要求需进行复杂的地形分析、物体识别和环境监测等任务,这些任务常常会给设备带来严重的功耗问题,并且严重影响无人机的飞行时间。针对上述问题,提出一种基于卷积神经网络的逐级唤醒存内计算控制器设计方案,控制器内部能够完成分类型网络的前向推理,可根据分类结果唤醒对应的边缘端设备。控制器的模拟部分采用存算一体计算模式,数字部分采用分块的处理方式,在运行过程中可将空闲模块休眠,以降低系统的整体功耗。此外,控制器内部还集成了级联接口,可以将复杂任务分解为多个层级的子任务,并将其部署到级联的控制器上,从而实现多级唤醒,使系统具备提前输出的潜力。实验采用ResNet-14作为神经网络模型,数据集采用CIFAR-10数据集。实验结果表明,在10 MHz的时钟频率下,基于卷积神经网络的逐级唤醒存内计算控制器可以实现60帧/s的检测帧率,并且分类精度达到84.61%,验证了该架构在能量受限应用场景下的可行性和高效性。

面向嵌入式平台的手势轨迹识别

针对大部分嵌入式平台计算能力较弱、无法实时运行基于神经网络的应用程序的情况,提出了一种基于Anchor-Free轻量化卷积神经网络(CNN)的手势交互方法。该方法缓解了Anchor-Based检测的弊端,通过神经网络实时检测人体手部,并利用动态时间规整(DTW)算法对手势轨迹进行分类。神经网络参数量仅有0.22 M,在自建手部检测数据集上的平均精度均值(mAP)交并比(IoU)=0.50︰0.95可以达到68%。在RK3568嵌入式平台上,每帧推理和后处理时间仅有31 ms,轨迹分类耗时仅有43 ms,CPU使用率仅有34%,满足实时性要求。

基于注意力机制的三维点云目标识别

针对现有基于深度学习方法的三维点云目标识别算法存在多层感知法缺少点间的特征交互、对点云间欧式距离的依赖、未考虑特征通道层面关联性问题,提出一种基于注意力机制的三维点云(PAttenCls)目标识别算法。采用基于点的空间注意力机制,挖掘各点之间的注意力值,实现自适应的云邻域选择;同时采用基于点的通道注意力机制,给特征通道自适应分配权重,实现特征增强。此外,在网络中添加了一个几何均匀化模块,以应对不同局部区域几何结构的不同特征模式。所提算法在ModelNet40数据集上的识别准确率为93.2%,在ScanObjectNN数据集最难子集上的识别准确率为80.9%,并在实测数据上验证了算法的有效性。实验证明了本文所提算法可以更好地提取点云的特征信息,使点云识别结果更加精准。

面向存算架构的神经网络数字系统设计

随着深度学习与神经网络的不断发展,庞大的计算量使得传统的冯·诺依曼架构设备面临“存储墙”等问题,因此“存内计算(Compute-In-Memory,CIM)”成为满足神经网络高时效需求和高运算量要求的主流设计方向。针对高密度数据的高性能计算提供高速且节能的解决方案,设计了一款神经网络加速器。首先,完成了对ResNet14神经网络的量化,依据其结构设计了一种面向存内计算的数字系统。而后,为了增强该系统的多网络适配性,提出了一种兼容性架构构想,使该数字系统可适配ResNet18或其他卷积神经网络的部分卷积层。最后,将该系统加载到FPGA上进行验证。在10 MHz的时钟频率下,以Cifar-10和MNIST数据集进行目标分类任务,分别得到60 FPS下84.17%和98.79%的准确率,具有更小的数据位宽和相近的准确率。

无人车目标识别主干网络技术特点对比分析

目标识别是无人车自动驾驶视觉感知模块的核心技术之一。当前,目标识别主要依靠主干网络提取特征,进而对目标进行分类与回归。通常情况下,无人车嵌入式计算平台的计算与存储能力有限,为了降低主干网络的算力与存储量,提升无人车的计算速度与效率,对目标分类任务的主干网络进行综合比较分析。围绕卷积核、感受野、池化层、全连接层、激活函数等,以cifar10和cifar100为实验数据,从理论分析与数据实践层面,对主干网络算子的选择与网络搭建进行分析对比,总结、归纳特征提取主干网络搭建的主要思路与做法。结果表明,该分析结论对目标分类主干网络在嵌入式无人车系统中的应用具有一定的理论指导与参考价值。

基于集成改进卷积注意力块的SAR图像目标分类算法

在合成孔径雷达(Synthetic Aperture Radar,SAR)图像中,目标的轮廓和细节通常比较复杂。传统的卷积神经网络(Convolutional Neural Network,CNN)只使用单一均值参数进行无差别的特征提取,不能很好地区分SAR特征之间的差异。为了解决此问题,文章提出了1种基于集成改进卷积注意力块(Improved Convolutional Block Attention Module,ICBAM)的SAR图像目标分类算法ICBAM_CNN。首先,该模块通过引入方差参数至传统CBAM模块中,设计了1种改进的CBAM注意力机制,有助于分类识别网络更好地学习SAR图像不同目标卷积层输出与通道注意力之间的差异信息,提升不同SAR目标特征的可分离性;然后,ICBAM设计了1种中心坐标注意力机制,能更好地捕捉SAR图像中目标的中心分布特征,有效抑制杂波对SAR目标分类影像的干扰;最后,为了提高效率,将改进后的ICBAM模块集成到CNN网络中,实现SAR图像目标分类。ICBAM_CNN深度融合了SAR图像目标的多层级特征,提升了SAR目标特征的可分离性,可实现SAR图像目标的高精度、高效率识别分类。通过MSTAR数据集进行实验,结果表明,相比于传统CBAM方法,改进ICBAM方法的精确率提升了2.44%,召回率提升了2.24%,F1-score提升了2.34%。

结合批规范化层的深度学习模型在水中目标识别中的应用

针对深度学习模型在训练水声样本数据时会出现稳定性差进而导致分类识别效果不佳的问题,从网络的局部连接、空间位置排列以及模型设计的角度出发,研究在原有一维序列卷积核与一维序列池化的基础上,引入批规范化层,构建了深度学习网络模型。通过归一化处理,达到加速网络模型的收敛过程以及提高训练过程中的稳定性的目的。最终为验证该模型的有效性,对3类水声目标样本数据进行网络训练和模型验证,证明该模型对水声目标数据分类识别有一定程度的性能提升。

基于图神经网络的盲道无障碍断点识别方法

针对既有盲道存在的问题,基于使用者行为和盲道识别特征,通过分类算法的应用,提出一套可行的盲道无障碍断点识别方法:首先,模拟使用者行为,基于图神经网络实现盲道断点分类,并结合盲道及障碍物检测,构建盲道断点智能识别模型;其次,结合规范和统计学方法确定识别对象并构建数据集,根据有监督的机器学习结果优化模型;最后,通过实际案例论证该方法的可行性,并结合城市大脑、“众包”概念,描述该方法应用前景。实验结果显示,该方法较其余分类算法有更高的准确度,实现机器代替人工对盲道问题高效、准确、系统、全自动的识别。

结合跨层特征融合网络和非局部的识别方法

在以往的方法中,针对于图像识别,主要目标是提取出足够多的具有区分度的局部关键特征。由于细粒度图像分类是对同属于一个大类下面的具体小类别的细分,所以不同的细类之间差别很小,要想提高准确率,更需要找到具有辨别力度的局部区域。论文使用了目标-导航网络,来筛选出前k个最具有判别力的局部区域块,然后针对于筛选出来的k个不同的区域块,利用非局部模块的思想,捕捉不同局部区域之间的联系,更加充分地利用了图像信息,以此来提高精度。与此同时,在残差网络中,使用了卷积注意模块捕捉不同通道注意力特征之间的联系,且在最后的全连接层处改进了网络架构,使用跨层特征融合的方法来代替了简单的级联。考虑到在细粒度识别中,图像的标注需要耗费非常多的人力物力,所以论文中提出的方法,是自监督的。

基于卷积神经网络的ISAR图像目标识别

利用电磁仿真软件构建了多角度、多分辨率、多目标的逆合成孔径雷达(inverse synthetic aperture radar,ISAR)图像数据集,采用AlexNet、VGGNet、ResNet、MobileNet等卷积神经网络对仿真ISAR图像进行了训练和多目标分类识别实验,并对实验结果进行了比较分析。实验结果表明,4种网络模型识别性能接近,目标识别时间在15 ms内,识别准确率接近100%,且具备轻量化识别条件。

一种基于BP神经网络群的自适应分类方法及其应用

本文针对基于BP神经网络的分类系统 ,提出了神经网络群的概念 ,在此基础上给出了一种系统自适应增长算法 ,使得在新的目标类型加入时系统结构能够自适应调整 .验证结果表明 ,该算法可以在增加新的目标类型时简化系统结构的调整过程 ,缩短重新训练网络所需要的时间 。

基于特征的模糊神经网络遥感图像目标分类识别

特征是图像处理中用于辨识目标的最基本属性。提出了利用模糊神经网络方法,针对舰船的几何特征、矩特征和纹理特征进行舰船目标识别处理。首先简单地描述了几何特征、矩特征尤其是Hu矩特征、一阶纹理特征和二阶纹理特征。然后分别对仿真数据、卫星观测数据中的舰船目标,以及自动检测处理获取的舰船目标的几何特征、Hu机特征和纹理特征进行了提取和分析。模糊神经网络方法可以综合模糊集理论和神经网络方法的优势,有效地实现基于特征的图像目标分类识别处理。文章首先描述了一种主从神经元结构的模糊神经网络分类识别方法,然后利用该方法对大型舰船进行分类识别,包括基于单类舰船特征的分类识别和基于多源(时相)数据融合的分类识别。实验结果表明,基于大型舰船的几何特征、矩特征和纹理特征,利用模糊神经网络方法可以实现对大型舰船目标的有效分类识别。通过多源数据融合处理,可以改善分类识别效果。

基于小波变换和神经网络的舰船目标识别和分类

根据舰船轴频电场信号的特点,提出了基于小波变换和神经网络的目标识别和分类方法。利用小波变换计算轴频电场信号的功率谱,提取特征量,根据轴频电场平均累积功率谱对舰船目标和海洋环境进行识别。根据轴频电场信号的功率谱分析,对不同类型舰船目标进行识别。运用BP神经网络,对不同类型舰船目标和海洋环境进行分类。通过对海上试验采集轴频电场数据的仿真计算,证实了该识别和分类方法的有效性。