基于迁移学习与多标签平滑策略的图像自动标注

针对图像标注数据集标签分布不平衡问题,提出了基于标签平滑策略的多标签平滑单元（MLSU）。MLSU在网络模型训练过程中自动平滑数据集中的高频标签,使网络适当提升了低频标签的输出值,从而提升了低频标注词的标注性能。为解决图像标注数据集样本数量不足造成网络过拟合的问题,提出了基于迁移学习的卷积神经网络（CNN）模型。首先利用互联网上的大型公共图像数据集对深度网络进行预训练,然后利用目标数据集对网络参数进行微调,构建了一个多标签平滑卷积神经网络模型（CNN-MLSU）。分别在Corel5K和IAPR TC-12图像标注数据集上进行实验,在Corel5K数据集上,CNN-MLSU较卷积神经网络回归方法（CNN-R）的平均准确率与平均召回率分别提升了5个百分点和8个百分点;在IAPR TC-12数据集上,CNN-MLSU较两场K最邻近模型（2PKNN_ML）的平均召回率提升了6个百分点。实验结果表明,基于迁移学习的CNN-MLSU方法能有效地预防网络过拟合,同时提升了低频词的标注效果。     

An approach to automatic learning assessment based on the computational theory of perceptions

E-learning systems output a huge quantity of data on a learning process. However, it takes a lot of specialist human resources to manually process these data and generate an assessment report. Additionally, for formative assessment, the report should state the attainment level of the learning goals defined by the instructor.This paper describes the use of the granular linguistic model of a phenomenon (GLMP) to model the assessment of the learning process and implement the automated generation of an assessment report. GLMP is based on fuzzy logic and the computational theory of perceptions. This technique is useful for implementing complex assessment criteria using inference systems based on linguistic rules. Apart from the grade, the model also generates a detailed natural language progress report on the achieved proficiency level, based exclusively on the objective data gathered from correct and incorrect responses. This is illustrated by applying the model to the assessment of Dijkstra’s algorithm learning using a visual simulation-based graph algorithm learning environment, called GRAPHs.     

Deep learning based enhanced tumor segmentation approach for MR brain images

Automation in medical industry has become one of the necessities in today’s medical scenario. Radiologists/physicians need such automation techniques for accurate diagnosis and treatment planning. Automatic segmentation of tumor portion from Magnetic Resonance (MR) brain images is a challenging task. Several methodologies have been developed with an objective to enhance the segmentation efficiency of the automated system. However, there is always scope for improvement in the segmentation process of medical image analysis. In this work, deep learning-based approach is proposed for brain tumor image segmentation. The proposed method includes the concept of Stationary Wavelet Transform (SWT) and new Growing Convolution Neural Network (GCNN). The significant objective of this work is to enhance the accuracy of the conventional system. A comparative analysis with Support Vector Machine (SVM) and Convolution Neural Network (CNN) is carried out in this work. The experimental results prove that the proposed technique has outperformed SVM and CNN in terms of accuracy, PSNR, MSE and other performance parameters.     

A cost-effective manufacturing process recognition approach based on deep transfer learning for CPS enabled shop-floor

The rapid development of the industrial Internet of Things has promoted manufacturing to develop towards the cyber-physical system, of which highly accurate process recognition plays an important role in achieving proactive monitoring of intelligent manufacturing process. Compared to the traditional handcrafted feature-based method, deep model owns convenience in terms of extracting feature automatically for the recognition. However, training a deep model is time-consuming and also requires large-scale training samples. To solve these problems and obtain high accuracy in the meanwhile, a deep transfer learning-based manufacturing process recognition approach is proposed in this study. A pre-trained model based on a convolutional neural network is used to extract low dimensional features followed by a fine-tuning process to target the specific process recognition task. Experimental verification of two datasets was conducted to demonstrate this cost-effective method. The results showed the proposed method can get better accuracy with less training time and fewer training samples.     

基于深度学习的人物肖像全自动抠图算法

针对抠图任务中人物抠图完整度低、边缘不够精细化等繁琐问题,提出了一种基于深度学习 的人物肖像全自动抠图算法。算法采用三分支网络进行学习,语义分割分支(SSB)学习  图的语义信息,细节 分支(DB)学习  图的细节信息,混合分支(COM)将 2 个分支的学习结果汇总。首先算法的编码网络采用轻量 级卷积神经网络(CNN) MobileNetV2,以加速算法的特征提取过程;其次在 SSB 中加入注意力机制对图像特 征通道重要性进行加权,在 DB 加入空洞空间金字塔池化(ASPP)模块,对图像的不同感受野所提取的特征进 行多尺度融合;然后解码网络的 2 个分支通过跳级连接融合不同阶段编码网络提取到的特征进行解码;最后 将 2 个分支学习的特征融合在一起得到图像的  图。实验结果表明,该算法在公开的数据集上抠图效果优于 所对比的基于深度学习的半自动和全自动抠图算法,在实时流视频抠图的效果优于 Modnet。     

一种HDLC数据转发的全自动硬件引擎设计

讨论了一种HDLC协议数据转发的全自动硬件设计。在SDH产品的ECC接口中采用硬件实现,通过自动控制BD的方式,实现HDLC协议的数据重新打包转发、错误处理、重发帧数据等操作,不必CPU干预。该设计实现了两个ECC口收发全双工、8个DCC口点对点同时收发的高速率、多通道设计。     

基于迁移学习的讽刺检测

准确的讽刺检测对于情感分析等任务至关重要.传统的方法严重依赖于离散的人工制定的特征.现有的研究大多将讽刺检测作为一种标准的监督学习文本分类任务,但是监督学习需要有大量数据,而这些数据的收集和标注都存在困难.由于目标任务有限的数据集可能导致讽刺检测的低性能,为此将讽刺检测作为一种迁移学习任务,将讽刺标记文本的监督学习与外部分析资源的知识转移相结合.通过转移的资源知识来改进神经网络模型,以此提升对目标任务的检测性能.在公开可用的数据集上的实验结果表明,提出的基于迁移学习的讽刺检测模型优于现有较先进的讽刺检测模型.     

基于上下文学习的轻量级自动抠图算法

为解决以往算法无法平衡精度和模型大小的问题,提出一种基于上下文学习的轻量级自动抠图算法.采用上下文特征聚合模块和编解码结构相结合的方式进行网络构建,其中编解码器能够有效进行特征提取,通过恢复空间信息捕获更清晰的对象边界;上下文特征聚合模块能够编码多尺度的上下文信息,保留更多细节纹理特征,提高结果的精度.将深度可分离卷积...     

基于FCM的文本迁移学习算法

传统的机器学习方法是在训练数据和测试数据分布一致的前提下进行的。然而,在一些现实世界中的应用,训练数据和测试数据来自不同的领域。在不考虑数据分布的情况下,传统的机器学习算法可能会失效,针对这一问题,提出一种基于模糊C均值(FCM)的文本迁移学习算法。首先,通过简单分类器对测试样本分类,接着,利用自然邻算法构建样本初始模糊隶属度;然后,利用FCM算法通过迭代更新样本模糊隶属度,修正样本标签;最后,对样本孤立点进行处理,得到最终分类结果。实验结果表明,该算法具有较好的正确率,有效的解决了在训练数据和测试数据分布不一致的情况下的文本分类问题。     

基于元强化学习的自动列车定速控制

本文考虑自动列车在路况变化下的定速控制问题. 由于铁路路况的复杂以及列车动力学的不确定性, 基于模型的控制器难以稳定、快速、精确地进行定速控制. 我们提出了一种无模型控制器, 其只需要很少的列车运行数据即可适应新的路况. 首先, 我们将列车的定速控制问题建模为一系列转移概率未知的静态连续马尔可夫过程. 然后, 我们应用元强化学习去求解该马尔可夫过程, 得到自适应神经网络控制器. 仿真说明该无模型控制器能够高效地进行定速控制, 并能迅速适应新的环境, 同时满足系统约束.     

Computer-aided maintenance management systems selection based on a fuzzy AHP approach

Computerized maintenance management systems (CMMS) are common in today’s industries. CMMS can bring a large number of benefits, which include increased productivity, reduced costs, and effective utilization of the assets in any manufacturing and service producer. The list of CMMS that are available in the market has grown very rapidly during the last years. When purchasing a system, one that suits the specific needs and objectives of the company’s maintenance operations should be preferred. Several selection methods were proposed in literature. Up to now, no article has considered ambiguity and uncertainty factors when selecting effective CMMS. In addition, CMMS selection decisions involve the simultaneous consideration of multiple criteria, including tangible and intangible factors; prioritizing these factors can be a great challenge and a complex task. Therefore, no attempt has been made to incorporate fuzziness into multicriteria decision-making in the area of CMMS selection. This work proposes a fuzzy-based methodology for comparative evaluation of a number of CMMS alternatives. The proposal is based on the well-known multicriteria decision method called Analytical Hierarchy Process (AHP) with triangular numbers. An example is given to illustrate the proposed methodology. Finally, a software prototype for implementing this method was implemented. To illustrate and validate the proposed approach and the software prototype developed some details are presented and discussed.     

基于深度迁移学习的烟雾识别方法

针对传统的基于传感器和图像特征的烟雾识别方法易被外部环境干扰且识别场景单一,从而造成烟雾识别精度较低,而基于深度学习的识别方法对数据量要求较高,对于烟雾数据缺失或数据来源受限的情况模型识别能力较弱的问题,提出一种基于深度迁移学习的烟雾识别方法。将ImageNet数据集作为源数据,利用VGG-16模型进行基于同构数据下的特征迁移。首先,将所有的图像数据进行预处理,对每张图像作随机变换（随机旋转、剪切、翻转等）;其次,引入VGG-16网络,将其卷积层特征进行迁移,并连接预先使用烟雾数据在VGG-16网络中训练过的全连接层;进而构建出基于迁移学习的深度网络,从而训练得到烟雾识别模型。利用公开数据集以及真实场景烟雾图像进行实验验证,实验结果表明,和现有主流烟雾图像识别方法相比,所提方法有较高的烟雾识别率,实验精度达96%以上。     

基于变权重迁移学习的BN参数学习算法

针对小数据集条件下的贝叶斯网络(Bayesian network,BN)参数估计困难问题,提出了一种基于变权重迁移学习(DWTL)的BN参数学习算法。首先,利用MAP和MLE方法学习得到目标域初始参数和各源域参数;然后根据不同源域数据样本贡献的不同计算源权重因子;接着基于目标域样本统计量与小数据集样本阈值的关系设计了目标域初始参数和源域参数的平衡系数;最后,基于上述参数、源权重因子和平衡系数计算得到新的目标参数。在实验研究中,通过对经典BN模型的参数学习问题验证了DWTL算法的有效性;针对小数据集下的轴承故障诊断问题,相较于传统迁移学习(LP)算法,DWTL算法学习精度提高了10%。实验结果表明:所提出的算法能够较好地解决样本数据集在相对稀缺条件下的目标参数建模问题。     

Machine learning approach for automatic diagnosis of Chlorosis in Vigna mungo leaves

Multimedia Tools and Applications - Viral infection in crops is something that may lead to a huge loss in crop yield as there are no known recovery procedures. Also, at the onset of yellowing in a...     

Cross-border e-commerce platform for commodity automatic pricing model based on deep learning

Electronic Commerce Research - With the innovation of information technology and the rise of the Internet economy, cross-border e-commerce has grown up to be an important means and strategy for...     

Guideline for the definition of EMF metamodels using an Entity-Relationship approach

Metamodels are a formalism for defining the abstract syntax of modeling languages. However, designing a suitable metamodel from the features intended for the language is not a trivial task. This paper presents a guideline for defining such metamodels using an Entity-Relationship approach in the Eclipse Modeling Framework. This guideline proposes to begin by determining the structural features of the language, such as types of relationships and elements with attributes. Subsequently, it offers alternative representations for these features aimed at satisfying different requirements, such as changeability or optimized model processing. Two case studies illustrate the use of the guideline and its trade-offs.     

基于迭代学习控制的列车自动运行研究

针对列控系统难以建立精确的动力学模型问题,利用列车运行过程中包含的大量重复信息,选用迭代学习算法对列车动力学模型中的未知参数进行辨识并提出基于迭代学习控制的列车自动运行控制算法。算法核心是利用历史数据生成新的控制量控制列车自动运行。仿真结果表明,经过一定次数的迭代,参数辨识值保持稳定并且列车能够严格跟踪目标曲线行驶,保证列车高精度、高平稳、高安全的运行。     

基于迁移学习的行人检测研究进展

行人检测是计算机视觉的研究热点和难点,近年来基于机器学习的行人检测技术取得了长足的进步,但由于不同场景的数据分布存在差异,已有检测器在新场景下的行人检测性能出现显著下降。为了避免繁琐的人工标注,充分利用原有检测器和标注样本,基于迁移学习的行人检测研究受到越来越多的关注。对其中涉及到的样本获取、迁移学习机制等关键技术进行综述,并从多个角度对现有方法进行分析和比较,最后对该技术的未来进行展望。     

Deep autoencoder based domain adaptation for transfer learning

Multimedia Tools and Applications - The concept of transfer learning has received a great deal of concern and interest throughout the last decade. Selecting an ideal representational framework for...     

基于深度学习的椎间孔狭窄自动多分级研究

椎间孔狭窄症的术前定性分级诊断对临床医生治疗策略的制定和患者健康恢复至关重要,但目前该方面临床上仍然存在很多问题,并且缺乏相关的研究和行之有效的方法用于辅助临床医生诊断。因此,为提高计算机辅助椎间孔狭窄症诊断准确率以及医生工作效率,本文提出一种基于深度学习的椎间孔狭窄图像自动分级算法。从人体矢状切脊柱核磁共振图像中提取脊柱椎间孔图像,并做图像预处理;设计一种监督式深度卷积神经网络模型,用于实现脊柱椎间孔图像数据集的自动多分级;利用迁移学习方法,解决深度学习算法在小样本数据集上的过拟合问题。实验结果表明,本文算法在脊柱椎间孔图像数据集上的分类精确度可达到87.5%以上,且其具有良好的鲁棒性和泛化能力。