为提高文本数据集的情感识别效率,采用深度学习方法,对互联网文本数据集进行情感分类,分别建立基于朴素贝叶斯(Naive Bayesian Classifier,NBC)、卷积神经网络(Convolutional Neural Network,CNN)和长短期记忆网络(Long Short Term Memo...为提高文本数据集的情感识别效率,采用深度学习方法,对互联网文本数据集进行情感分类,分别建立基于朴素贝叶斯(Naive Bayesian Classifier,NBC)、卷积神经网络(Convolutional Neural Network,CNN)和长短期记忆网络(Long Short Term Memory,LSTM)的情感识别模型。在二分类Focal损失函数的基础上,提出多分类Multi Focal损失函数MFL。基于搜狐新闻数据集和中美贸易战评论数据集的实验表明:使用多分类损失函数MFL的长短期记忆模型明显优于其它模型。展开更多
Based on the vector diffraction theory, the effect of complex phase filters on intensity distribution of a radially polarized multi Gaussian beam in the focal region of high NA lens is theoretically investigated. It i...Based on the vector diffraction theory, the effect of complex phase filters on intensity distribution of a radially polarized multi Gaussian beam in the focal region of high NA lens is theoretically investigated. It is observed that a properly designed multi belt complex phase filter can generate subwavelength novel focal patterns including splitting of focal spots and generation of multiple focal spot segments such as eight, six and four focal spots along the optical axis are obtained. We expect that such an investigation is useful for optical manipulation and material processing, multiple high refractive index particle trapping technologies.展开更多
Objective To compare and evaluate the efficacy of diagnosis and excision for appropriately selected breast multi-focal lesions and solitary lesion by ultrasound-guided vacuum-assisted biopsy(UGVAB).Methods Among 392 a...Objective To compare and evaluate the efficacy of diagnosis and excision for appropriately selected breast multi-focal lesions and solitary lesion by ultrasound-guided vacuum-assisted biopsy(UGVAB).Methods Among 392 appropriately selected patients,187 patients with multi-focal lesions and 205 patients with solitary lesion were treated by the 8-gauge UGVAB from May 2007 to June 2009.All lesions were removed as completely as possible.The patients with benign pathology underwent physical and ultrasound examinations at one week and 6 months after procedure.Results During the procedure,only three patients had vasovagal syncope and twenty others complained of other intraoperative discomfort.An accurate pathological diagnosis was obtained in all lesions.There was no apparent false-negative result among the 696 lesions with benign pathology at a follow-up of 6 months after procedure.The rates of malignant or premalignant pathology,postoperative complications and residual lesions in patients with multi-focal lesions were higher than those in patients with solitary lesion.If each lesion was considered as a subject of study,there was no significant difference between the two groups.Conclusion UGVAB is an effective method for diagnosis and excision of appropriately selected breast multi-focal lesions and can be used routinely.展开更多
为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度...为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。展开更多
文摘为提高文本数据集的情感识别效率,采用深度学习方法,对互联网文本数据集进行情感分类,分别建立基于朴素贝叶斯(Naive Bayesian Classifier,NBC)、卷积神经网络(Convolutional Neural Network,CNN)和长短期记忆网络(Long Short Term Memory,LSTM)的情感识别模型。在二分类Focal损失函数的基础上,提出多分类Multi Focal损失函数MFL。基于搜狐新闻数据集和中美贸易战评论数据集的实验表明:使用多分类损失函数MFL的长短期记忆模型明显优于其它模型。
文摘Based on the vector diffraction theory, the effect of complex phase filters on intensity distribution of a radially polarized multi Gaussian beam in the focal region of high NA lens is theoretically investigated. It is observed that a properly designed multi belt complex phase filter can generate subwavelength novel focal patterns including splitting of focal spots and generation of multiple focal spot segments such as eight, six and four focal spots along the optical axis are obtained. We expect that such an investigation is useful for optical manipulation and material processing, multiple high refractive index particle trapping technologies.
文摘Objective To compare and evaluate the efficacy of diagnosis and excision for appropriately selected breast multi-focal lesions and solitary lesion by ultrasound-guided vacuum-assisted biopsy(UGVAB).Methods Among 392 appropriately selected patients,187 patients with multi-focal lesions and 205 patients with solitary lesion were treated by the 8-gauge UGVAB from May 2007 to June 2009.All lesions were removed as completely as possible.The patients with benign pathology underwent physical and ultrasound examinations at one week and 6 months after procedure.Results During the procedure,only three patients had vasovagal syncope and twenty others complained of other intraoperative discomfort.An accurate pathological diagnosis was obtained in all lesions.There was no apparent false-negative result among the 696 lesions with benign pathology at a follow-up of 6 months after procedure.The rates of malignant or premalignant pathology,postoperative complications and residual lesions in patients with multi-focal lesions were higher than those in patients with solitary lesion.If each lesion was considered as a subject of study,there was no significant difference between the two groups.Conclusion UGVAB is an effective method for diagnosis and excision of appropriately selected breast multi-focal lesions and can be used routinely.
文摘为了提高多视图深度估计结果精度,提出一种基于自适应空间特征增强的多视图深度估计算法。设计了由改进后的特征金字塔网络(feature pyramid network,FPN)和自适应空间特征增强(adaptive space feature enhancement,ASFE)组成的多尺度特征提取模块,获取到具有全局上下文信息和位置信息的多尺度特征图像。通过残差学习网络对深度图进行优化,防止多次卷积操作出现重建边缘模糊的问题。通过分类的思想构建focal loss函数增强网络模型的判断能力。由实验结果可知,该算法在DTU(technical university of denmark)数据集上和CasMVSNet(Cascade MVSNet)算法相比,在整体精度误差、运行时间、显存资源占用上分别降低了14.08%、72.15%、4.62%。在Tanks and Temples数据集整体评价指标Mean上该模型优于其他算法,证明提出的基于自适应空间特征增强的多视图深度估计算法的有效性。
