大量基于深度学习的无监督视频目标分割(Unsupervised video object segmentation,UVOS)算法存在模型参数量与计算量较大的问题,这显著限制了算法在实际中的应用.提出了基于运动引导的视频目标分割网络,在大幅降低模型参数量与计算量的...大量基于深度学习的无监督视频目标分割(Unsupervised video object segmentation,UVOS)算法存在模型参数量与计算量较大的问题,这显著限制了算法在实际中的应用.提出了基于运动引导的视频目标分割网络,在大幅降低模型参数量与计算量的同时,提升视频目标分割性能.整个模型由双流网络、运动引导模块、多尺度渐进融合模块三部分组成.具体地,首先,RGB图像与光流估计输入双流网络提取物体外观特征与运动特征;然后,运动引导模块通过局部注意力提取运动特征中的语义信息,用于引导外观特征学习丰富的语义信息;最后,多尺度渐进融合模块获取双流网络的各个阶段输出的特征,将深层特征渐进地融入浅层特征,最终提升边缘分割效果.在3个标准数据集上进行了大量评测,实验结果表明了该方法的优越性能.展开更多
目的聚二甲基硅氧烷(PDMS)作为微流控芯片研制中常用的高分子材料,其本身的疏水特性是影响微流控芯片整体键合效果的主要障碍。为了在短时间内成功实现PDMS与基底材料的有效键合与封装,设计一种可在普通实验室开展的低成本且高效的PDMS...目的聚二甲基硅氧烷(PDMS)作为微流控芯片研制中常用的高分子材料,其本身的疏水特性是影响微流控芯片整体键合效果的主要障碍。为了在短时间内成功实现PDMS与基底材料的有效键合与封装,设计一种可在普通实验室开展的低成本且高效的PDMS材料改性方法。方法基于紫外臭氧光照改性法对PDMS材料表面进行改性研究,通过正交试验深入研究紫外光照射时间、距离及通氧时间对PDMS表面改性效果的影响,并在50℃水浴环境下通过测量不同时间PDMS基片与盖片(PDMS或玻璃)的键合强度,从而确定最优工艺参数组合。结果相比于传统紫外光照射表面改性法的键合时间(大于50 h),本工艺可在6 h内成功实现PDMS的有效封装,并确定了P—P键合和P—G键合的最优参数组合,两者平均键合强度均大于200 k Pa。结论整个工艺操作简单、成本低,可作为普通实验室开展微流控实验研究的有益补充。展开更多
Carbon nanotube thin film transistor (CNT-TFF) is an emerging technology for future macroelectronics, such as chemical and biological sensors, optical detectors, and the backplane driving circuits for flat panel dis...Carbon nanotube thin film transistor (CNT-TFF) is an emerging technology for future macroelectronics, such as chemical and biological sensors, optical detectors, and the backplane driving circuits for flat panel displays. The mostly reported fabrication method of CNT-TFT is a lift-off based photolithography process. In such fabrication process, photoresist (PR) residue contaminates the interface of tube-metal contact and deteriorates the device performance. In this paper, ultraviolet ozone (UVO) and oxygen plasma treat- ments were employed to remove the PR contamination. Through our well-designed experiments, the UVO treatment is confirmed an effective way of cleaning contamination at the tube-metal interface, while oxygen plasma treatment is too reactive and hard to control, which is not appropriate for CNT-TFTs. It is determined that 2-6 rain UVO treatment is the preferred window, and the best optimized treatment time is 4 rain, which leads to 15% enhancement of device performance.展开更多
文摘大量基于深度学习的无监督视频目标分割(Unsupervised video object segmentation,UVOS)算法存在模型参数量与计算量较大的问题,这显著限制了算法在实际中的应用.提出了基于运动引导的视频目标分割网络,在大幅降低模型参数量与计算量的同时,提升视频目标分割性能.整个模型由双流网络、运动引导模块、多尺度渐进融合模块三部分组成.具体地,首先,RGB图像与光流估计输入双流网络提取物体外观特征与运动特征;然后,运动引导模块通过局部注意力提取运动特征中的语义信息,用于引导外观特征学习丰富的语义信息;最后,多尺度渐进融合模块获取双流网络的各个阶段输出的特征,将深层特征渐进地融入浅层特征,最终提升边缘分割效果.在3个标准数据集上进行了大量评测,实验结果表明了该方法的优越性能.
文摘目的聚二甲基硅氧烷(PDMS)作为微流控芯片研制中常用的高分子材料,其本身的疏水特性是影响微流控芯片整体键合效果的主要障碍。为了在短时间内成功实现PDMS与基底材料的有效键合与封装,设计一种可在普通实验室开展的低成本且高效的PDMS材料改性方法。方法基于紫外臭氧光照改性法对PDMS材料表面进行改性研究,通过正交试验深入研究紫外光照射时间、距离及通氧时间对PDMS表面改性效果的影响,并在50℃水浴环境下通过测量不同时间PDMS基片与盖片(PDMS或玻璃)的键合强度,从而确定最优工艺参数组合。结果相比于传统紫外光照射表面改性法的键合时间(大于50 h),本工艺可在6 h内成功实现PDMS的有效封装,并确定了P—P键合和P—G键合的最优参数组合,两者平均键合强度均大于200 k Pa。结论整个工艺操作简单、成本低,可作为普通实验室开展微流控实验研究的有益补充。
基金supported by the National Key Research and Development Program of China(2016YFA0201902)the National Natural Science Foundation of China(61621061)Beijing Municipal Science&Technology Commission(Z171100002017001)
文摘Carbon nanotube thin film transistor (CNT-TFF) is an emerging technology for future macroelectronics, such as chemical and biological sensors, optical detectors, and the backplane driving circuits for flat panel displays. The mostly reported fabrication method of CNT-TFT is a lift-off based photolithography process. In such fabrication process, photoresist (PR) residue contaminates the interface of tube-metal contact and deteriorates the device performance. In this paper, ultraviolet ozone (UVO) and oxygen plasma treat- ments were employed to remove the PR contamination. Through our well-designed experiments, the UVO treatment is confirmed an effective way of cleaning contamination at the tube-metal interface, while oxygen plasma treatment is too reactive and hard to control, which is not appropriate for CNT-TFTs. It is determined that 2-6 rain UVO treatment is the preferred window, and the best optimized treatment time is 4 rain, which leads to 15% enhancement of device performance.