提出了一种基于改进自适应矩估计(adaptive moment estimation,Adam)算法优化器的卷积神经网络(convolutional neural networks,CNN)电子显微镜(电镜)医学图像分类方法。该方法根据卷积神经网络数据迭代的特点,采用具有下降趋势的幂指...提出了一种基于改进自适应矩估计(adaptive moment estimation,Adam)算法优化器的卷积神经网络(convolutional neural networks,CNN)电子显微镜(电镜)医学图像分类方法。该方法根据卷积神经网络数据迭代的特点,采用具有下降趋势的幂指数学习率改进策略,通过添加修正因子,将上一阶段的梯度值与当前梯度值进行对比、调节,通过梯度值衰减来逐次更新学习率的大小,实现优化器学习率的自适应变化,改善CNN网络模型的收敛性能,实现医学电镜图像的分类。实验结果表明,相比经典的Adam优化器分类方法,改进方法能提高电镜医学图像分类算法的精度,最大分类精度可以到达92%,同时减小图像样本在分类时出现的迭代振荡、分类稳定性不足等现象。展开更多
眼底血管的健康状态对于研究各类眼科疾病具有重要的参考意义。为了帮助临床医疗人员对眼底微血管形态结构图像的分析来诊断疾病,文中提出了一种基于编码-解码(Encoder-Decoder)结构的U-net的眼底血管分割方法。首先,在模型训练之前对...眼底血管的健康状态对于研究各类眼科疾病具有重要的参考意义。为了帮助临床医疗人员对眼底微血管形态结构图像的分析来诊断疾病,文中提出了一种基于编码-解码(Encoder-Decoder)结构的U-net的眼底血管分割方法。首先,在模型训练之前对图像进行预处理,然后使用Leaky ReLU激活函数替换U-net ReLU,避免了神经元的死亡问题,同时使用Adam(Adaptive Moment Estimate)优化器代替梯度下降法优化学习策略,最后对血管分割的平均交并比进行计算评估。实验表明,优化后的模型的平均精度可达到93.29%,相比原算法提升了3.26%。展开更多
After introducing a novel 3-DOF high speed and high precision manipulator which combines direct driven planar parallel mechanism and linear actuator, ways of increasing its stiffness are studied through dynamics simul...After introducing a novel 3-DOF high speed and high precision manipulator which combines direct driven planar parallel mechanism and linear actuator, ways of increasing its stiffness are studied through dynamics simulation in ADAMS software environment. Design study is carried out by parametric analysis tools to analyze the approximate sensitivity of the design variables, including the effects of parameters of each beam cross section and relative position of linear actuator on model performance. Conclusions are drawn on the appropriate way of dynamics optimization to get a lightweight and small deformation manipulator. A planar parallel mechanism with different cross section is used to an improved manipulator. Resuits of dynamics simulation of the improved system and another unrefined one are compared. The stiffness of them is almost equal, but the mass of the improved one decreases greatly, which illustrates the wavs efficient.展开更多
文摘提出了一种基于改进自适应矩估计(adaptive moment estimation,Adam)算法优化器的卷积神经网络(convolutional neural networks,CNN)电子显微镜(电镜)医学图像分类方法。该方法根据卷积神经网络数据迭代的特点,采用具有下降趋势的幂指数学习率改进策略,通过添加修正因子,将上一阶段的梯度值与当前梯度值进行对比、调节,通过梯度值衰减来逐次更新学习率的大小,实现优化器学习率的自适应变化,改善CNN网络模型的收敛性能,实现医学电镜图像的分类。实验结果表明,相比经典的Adam优化器分类方法,改进方法能提高电镜医学图像分类算法的精度,最大分类精度可以到达92%,同时减小图像样本在分类时出现的迭代振荡、分类稳定性不足等现象。
文摘眼底血管的健康状态对于研究各类眼科疾病具有重要的参考意义。为了帮助临床医疗人员对眼底微血管形态结构图像的分析来诊断疾病,文中提出了一种基于编码-解码(Encoder-Decoder)结构的U-net的眼底血管分割方法。首先,在模型训练之前对图像进行预处理,然后使用Leaky ReLU激活函数替换U-net ReLU,避免了神经元的死亡问题,同时使用Adam(Adaptive Moment Estimate)优化器代替梯度下降法优化学习策略,最后对血管分割的平均交并比进行计算评估。实验表明,优化后的模型的平均精度可达到93.29%,相比原算法提升了3.26%。
文摘After introducing a novel 3-DOF high speed and high precision manipulator which combines direct driven planar parallel mechanism and linear actuator, ways of increasing its stiffness are studied through dynamics simulation in ADAMS software environment. Design study is carried out by parametric analysis tools to analyze the approximate sensitivity of the design variables, including the effects of parameters of each beam cross section and relative position of linear actuator on model performance. Conclusions are drawn on the appropriate way of dynamics optimization to get a lightweight and small deformation manipulator. A planar parallel mechanism with different cross section is used to an improved manipulator. Resuits of dynamics simulation of the improved system and another unrefined one are compared. The stiffness of them is almost equal, but the mass of the improved one decreases greatly, which illustrates the wavs efficient.
