In modern textile industry, Tissue online Automatic Inspection (TAI) is becoming an attractive alternative to Human Vision Inspection (HVI). HVI needs a high level of attention nevertheless leading to low performance ...In modern textile industry, Tissue online Automatic Inspection (TAI) is becoming an attractive alternative to Human Vision Inspection (HVI). HVI needs a high level of attention nevertheless leading to low performance in terms of tissue inspection. Based on the co-occurrence matrix and its statistical features, as an approach for defects textile identification in the digital image, TAI can potentially provide an objective and reliable evaluation on the fabric production quality. The goal of most TAI systems is to detect the presence of faults in textiles and accurately locate the position of the defects. The motivation behind the fabric defects identification is to enable an on-line quality control of the weaving process. In this paper, we proposed a method based on texture analysis and neural networks to identify the textile defects. A feature extractor is designed based on Gray Level Co-occurrence Matrix (GLCM). A neural network is used as a classifier to identify the textile defects. The numerical simulation showed that the error recognition rates were 100% for the training and 100%, 91% for the best and worst testing respectively.展开更多
针对目前第三方监测工作中沉降监测存在的数据传输过程繁琐、数据安全性和真实性难以保障等问题,本文提出开发一款基于Android智能终端的水准测量数据采集与处理软件(Level Data Measuring and Adjustment Software,简称Level DMAS),该...针对目前第三方监测工作中沉降监测存在的数据传输过程繁琐、数据安全性和真实性难以保障等问题,本文提出开发一款基于Android智能终端的水准测量数据采集与处理软件(Level Data Measuring and Adjustment Software,简称Level DMAS),该软件通过蓝牙串口模块实现智能终端与电子水准仪的无线通信,并能对采集到的监测数据进行现场检核与网平差计算,还可记录测站位置信息,最后将整个项目数据加密后上传至第三方监测平台进行科学化管理,从而大大节约人力物力,提高了数据传输效率,最大限度保障了数据的安全性和真实性。展开更多
针对现有方面级情感分类模型存在方面词与上下文交互不充分、分类精度低的问题,提出一种基于多交互特征融合的方面级情感分类方法(ASMFF:Aspect-level Sentiment classification method based on Multi-interaction Feature Fusion)。首...针对现有方面级情感分类模型存在方面词与上下文交互不充分、分类精度低的问题,提出一种基于多交互特征融合的方面级情感分类方法(ASMFF:Aspect-level Sentiment classification method based on Multi-interaction Feature Fusion)。首先,将上下文和方面词分别进行特殊标记,输入BERT(Bidirectional Encoder Representations from Transformers)编码层进行文本特征向量提取。其次,将文本特征向量输入AOA(Attention Over Attention)和IAN(Interactive Attention Networks)网络提取交互注意力特征向量。最后,将得到的两种交互特征向量进行融合学习,通过交叉熵损失函数进行概率计算、损失回传和参数更新。在Laptop、Restaurant和Twitter 3个公开数据集上的实验结果表明,ASMFF模型的分类准确率分别为80.25%、84.38%、75.29%,相比基线模型有显著提升。展开更多
文摘In modern textile industry, Tissue online Automatic Inspection (TAI) is becoming an attractive alternative to Human Vision Inspection (HVI). HVI needs a high level of attention nevertheless leading to low performance in terms of tissue inspection. Based on the co-occurrence matrix and its statistical features, as an approach for defects textile identification in the digital image, TAI can potentially provide an objective and reliable evaluation on the fabric production quality. The goal of most TAI systems is to detect the presence of faults in textiles and accurately locate the position of the defects. The motivation behind the fabric defects identification is to enable an on-line quality control of the weaving process. In this paper, we proposed a method based on texture analysis and neural networks to identify the textile defects. A feature extractor is designed based on Gray Level Co-occurrence Matrix (GLCM). A neural network is used as a classifier to identify the textile defects. The numerical simulation showed that the error recognition rates were 100% for the training and 100%, 91% for the best and worst testing respectively.
文摘针对目前第三方监测工作中沉降监测存在的数据传输过程繁琐、数据安全性和真实性难以保障等问题,本文提出开发一款基于Android智能终端的水准测量数据采集与处理软件(Level Data Measuring and Adjustment Software,简称Level DMAS),该软件通过蓝牙串口模块实现智能终端与电子水准仪的无线通信,并能对采集到的监测数据进行现场检核与网平差计算,还可记录测站位置信息,最后将整个项目数据加密后上传至第三方监测平台进行科学化管理,从而大大节约人力物力,提高了数据传输效率,最大限度保障了数据的安全性和真实性。
文摘针对现有方面级情感分类模型存在方面词与上下文交互不充分、分类精度低的问题,提出一种基于多交互特征融合的方面级情感分类方法(ASMFF:Aspect-level Sentiment classification method based on Multi-interaction Feature Fusion)。首先,将上下文和方面词分别进行特殊标记,输入BERT(Bidirectional Encoder Representations from Transformers)编码层进行文本特征向量提取。其次,将文本特征向量输入AOA(Attention Over Attention)和IAN(Interactive Attention Networks)网络提取交互注意力特征向量。最后,将得到的两种交互特征向量进行融合学习,通过交叉熵损失函数进行概率计算、损失回传和参数更新。在Laptop、Restaurant和Twitter 3个公开数据集上的实验结果表明,ASMFF模型的分类准确率分别为80.25%、84.38%、75.29%,相比基线模型有显著提升。
