结合文本监督和多粒度特征的姿态情绪识别

通过分析人体3D骨架的连续运动规律,建模姿态情绪识别方法,在人机交互等领域具有广泛的应用前景。针对姿态特征易混淆导致的骨架局部情绪性特征难捕获问题,提出了一种结合文本监督和多粒度特征的姿态情绪识别方法。通过多粒度特征融合网络模块,对情绪识别问题中人体的动态姿态进行建模。同时,采用文本化描述的标签对融合情绪特征进行监督训练,引导网络提取骨架各个部位的情绪性特征。在MPI情绪身体表达数据集上对该方法进行评估,试验结果表明,该方法准确率达80.50%,验证了其有效性。

基于DSA-BCNN的毫米波图像隐藏物检测

自动检测毫米波图像中被检人员是否携带隐藏物,是实现智能毫米波安检系统的重要技术之一。针对隐藏物在毫米波图像中的特征局部性和低辨识性问题,提出一种动态自注意力的双线性卷积神经网络,能够仅以图像级标签训练,实现图像中隐藏物的存在检测。引入自注意力机制以引导网络对隐藏物区域进行特征提取,增强网络刻画全局信息的能力;双线性池化构建的二阶特征丰富了网络对隐藏物与非检测区域的细微差异表征。实验结果验证了所提出方法对隐藏物检测的有效性,在各项评价指标上均高于其他基于卷积神经网络的方法,准确率达到93.6%。

6-31 Progress of Cryogenic System in 2014

During the year of 2014, the work of cryogenic group was mainly focused on operation of refrigerator system,test of TCM1, design and machining of HCM6. At the end of 2014 refrigerator system has achieved closed-loopoperation and recycle of helium gas, which did great contribution to superconducting cavity and superconductingmagnet test. Since the refrigerator was put into use(Fig. 1), it has operated continuously and reliably for more than5 110 h, reached helium lique ed capacity 270 L/h at 850 W.

6-32 Design and Operation of Cryogenic Vertical Testing System for Superconducting Cavity

The superconducting accelerating segment in injectorⅡof Accelerator Driven Sub-critical System(ADS) uses superconductingcavities whose frequency are 162.5 MHz. The vertical performance of superconducting cavity devicesmust be tested at liquid helium temperature(2, 4.2 K) before being assembled in cryostat. To meet the runninggoal of the 0  25 MeV linac system, it is necessary to build a set of cryogenic testing system for superconductingcavity, which is used to test the performance of the superconducting cavity and whether cavity meet the designgoal.

Progress of Cryomodule System in 2017

ADS Injector-II linear accelerator is mainly composed of ECR ion source,LEBT,RFQ,MEBT,a superconducting accelerating segment and HEBT.The superconducting accelerating segment is composed of two HWR010 cavity cryomodules,a taper cavity cryomodule and a spoke cavity cryomodule(CM1+CM2+CM3+CM4),which has accelerated the proton beam to 25 MeV in Jun.2017.

Testing Research on Cryogenic Surface Adsorption

In superconducting linear accelerator(LINAC),the performance and stability can be adversely impacted by gas adsorbed on cryogenic surfaces.Adsorption capacity and peculiarity of adsorption process on cryogenic niobium material surface are measured in order to improve the performance and stability of superconductive accelerator.The results and analysis(Fig.1-4)in this paper indicate:firstly,the saturated vapor pressure of hydrogen and helium is much higher than the maximum allowable value for superconductive LINAC operation,secondly,at cryogenic temperature,such as 4 K,it is the gas adsorption amount that determines the pressure,thirdly,pressure will not arise until single monolayer accomplished,it may take a long time.

6-5 Operation of HWRs for C-ADS Injector

There are two types of HWR cavities have been developed in IMP for C-ADS Injector.HWR010s have been used to accelerate proton from 2.1 MeV to 10 MeV.There are 12 HWR010s assembled in two CMs.The taper type HWR015s have been used to accelerate proton from 10MeV to 18.5 MeV.There are 5 HWRs assembled in the third CM.Proton beam has been accelerated to 26 MeV by HWR010,HWR015 and Spoke021 in June 2017.The beam has been accelerated to 18.5 MeV by HWR010 and HWR015.The CMs in the tunnel are shown in Fig.1.

基于DenseNet的散射成像景深拓展研究

散射现象广泛存在于自然界中。透过散射介质的大景深成像在计算成像领域具有重要的意义和应用价值。近年来,随着深度学习在计算成像领域的广泛应用,散射成像系统中的景深问题亟待进一步研究和拓展。以DenseNet为基础,结合UNet框架,建立了一个具有良好迁移性和景深拓展能力的深度卷积神经网络模型——DUNet。通过使用透过不同目数毛玻璃的散斑图像对网络进行训练,使成像景深拓展至距焦面50 mm处。初步的小鼠脑片实验结果表明,DUNet模型将有望应用于深层组织断层扫描。