基于卷积神经网络特征提取的拉曼光谱分类研究

拉曼光谱物质定性识别已被广泛的应用于化工、安防、缉毒等行业和研究领域,但是传统的拉曼光谱分析技术依赖于光谱数据库,通过光谱特征提取进行识别。特征提取是拉曼识别的关键处理步骤,通常利用主成分分析,因子分析等方法进行特征提取,而后通过KNN,SVM和随机森林等方法进行光谱特征定性识别,当拉曼数据库不存在待定性物质时,易造成待检测物质的错误分类。针对此问题,提出一种基于卷积神经网络的对数据库缺少物质光谱识别方法。在实验过程中,采用九类,200余种精神类药品拉曼光谱作为测试对象,通过搭建卷积神经网络自动特征提取并利用Softmax分类器将200余种物质,按照Amphetamine, cathinone, cannabinoids等九种类别进行定性分析。通过与传统机器学习方法如K近邻,支持向量机等方法进行比较,基于卷积神经网络的模型识别准确性有显著提高,该方法可为拉曼光谱数据库的光谱识别检索提供一种新的识别方法。     

3D打印贝壳仿生复合材料的拉伸力学行为

采用硬质和软质双组分材料,通过调控两种基体材料的装配夹角,采用光固化3D打印技术制备了不同装配方式的仿贝壳珍珠层复合材料,开展了准静态拉伸实验,结合扫描电镜观察,分析了其拉伸力学性能、断裂及能量耗散机理。研究结果表明,保持胞元边长不变,随着面内装配角度增加,仿贝壳珍珠层复合材料的强度呈线性增加趋势,断裂应变呈线性减小的趋势;随着面外装配角度增大,断裂应变呈线性减小趋势,而强度在面外装配角小于45°时呈增强趋势,超过45°时趋于稳定;面外装配角度为45°时,材料的强度达到最大值。试样在断裂前主要通过硬质材料的拔出、软/硬相界面处微裂纹的生成及微裂纹在扩展过程中的合并和偏转等方式耗散能量。     

氧化光栅型垂直腔面发射激光器的研究

通过分析矩形出光孔径和亚波长金属光栅结构,发现大孔径高功率垂直腔面发射激光器(VCSEL)的偏振控制的难点在于横模非常复杂。因此提出一种新型的氧化光栅型VCSEL结构,不仅能够很好地在有源区内引入各项异性的增益结构,并且最大的优势还在于能够完美地控制大孔径VCSEL的横模。通过有限元软件对器件有源区的电流分布进行了模拟,发现当光栅脊的宽度为1.8 μm时,载流子在光栅两端聚集的现象基本上可以消除,而且其电流密度分布差可以达到很高。     

Alpha Decay of the Neutron-deficient Isotopes 215,216U

The most neutron-deficient isotopes 215;216U were produced in the complete-fusion reaction 180W(40Ar, 4-5n)215,216U. Evaporation residues recoiled from the target were separated in-flight from the primary beam by the gas-filled recoil separator SHANS and subsequently identified on the basis of correlated -decay chains. Two -decaying states were identified in 216U, one for the ground state and the other for the isomeric state with 8+(h9=2f7=2) configuration. The -decay properties for 215;216U and the systematics of 8+ isomeric state in N =124,126 isotones were investigated.     

Structure of ^6He in the frame of a cluster model

A microscopic cluster model with a fully correlated Gaussian basis is developed. In the model, the stochastic variational method is used in order to calculate the ground-energy and the mean-square radius conveniently. Based on this model, the ground-energy level and radius of the neutron halo nucleus, ⁶He, are calculated as a α+n+n three-cluster model. The results are in good agreement with the experimental data.     

激光功率和能量计量技术的现状与展望

功率和能量是激光器的两个基本参数,激光功率和能量测量一直是激光参数计量中最基础的测量工作.总结了美国标准技术研究院(NIST)、英国国家物理实验室(NPL)以及中国国家计量院(NIM)在激光功率能量计量方面的技术现状,重点介绍了国防科工委光学计量一级站激光功率能量计量方面已具备的条件,详细说明了所采用的测量原理以及所达到的测量范围和测量不确定度等技术指标.提出了激光功率和能量计量的发展趋势和发展方向.     

Quantum information splitting of a two-qubit Bell state using a four-qubit entangled state

A scheme is proposed for quantum information splitting of a two-qubit Bell state by using a four-qubit entangled state as a quantum channel. In the scenario, it is supposed that there are three legitimate parties, say Alice, Bob and Charlie. Alice is the sender of quantum information. Bob and Charlie are two agents. Alice first performs GHZ state measurement and tells Bob and Charlie the measurement results via a classical channel. It is impossible for Bob to reconstruct the original state with local operations unless help is obtained from Charlie. If Charlie allows Bob to reconstruct the original state information, he needs to perform a single-qubit measurement and tell Bob the measurement result. Using the measurement results from Alice and Charlie, Bob can reconstruct the original state. We also consider the problem of security attacks. This protocol is considered to be secure.     

基于热图像的种蛋气室变化监测算法

种蛋气室的大小是监测种蛋孵化过程的重要指标之一。根据种蛋的热力学结构,种蛋在孵化过程中,包裹气室部分蛋壳会与其他部分蛋壳产生温差,从而可通过热红外图像进行观察。针对在种蛋孵化过程中,人工照蛋检测气室效率低的问题,探索设计了一种基于热图像的种蛋气室变化俯视监测算法。监测种蛋气室热图像的算法主要包括种蛋目标检测,种蛋图像分割和种蛋气室面积计算3个部分,其中种蛋的目标检测采用Faster-RCNN算法实现;种蛋图像分割采用BP神经网络算法实现;种蛋气室面积是在种蛋图像分割的基础上进行计算。使用孵化5天及以上的种蛋作为研究对象,并拍取种蛋的热图像进行试验。试验结果表明：种蛋热图像的目标检测的平均精度（mAP）为99.85%,拥有较好的检测效果。使用BP网络对种蛋进行图像分割。BP神经网络经过调参后,其网络最佳的结构为三层隐藏层,每个隐藏层拥有1 000个神经元,最优初始学习率为0.000 1,最优最大迭代次数为500。以F1-measure作为分割效果的评价指标,BP神经网络的图像分割总体结果为87.02%,Otsu算法的总体结果为65.25%。其中只有一个蛋的情况下,BP神经网络的分割结果为87.17%,Otsu算法的结果为68.86%。存在其他种蛋的干扰条件下,BP神经网络的分割结果为86.94%,Otsu算法的结果为61.64%,BP神经网络的分割效果优于Otsu分割算法,BP神经网络拥有更强的抗干扰能力。最后提取了孵化5～19 d种蛋的气室变化,通过观察种蛋气室大小曲线来监测种蛋的孵化情况,可看出随着天数的增加,气室有着明显变大的趋势。人工测量法与热红外测量法比较结果说明两者相关性为0.934 3,拥有较好的相关性。基于热图像的种蛋气室变化监测算法可在实际生产中实现种蛋的识别与气室大小的快速监测,为实现监测种蛋孵化的自动化提供了技术参考。     

On Extension of Optimal Entanglement Concentration of GHZ States in Quantum-Dot and Micro-Cavity Coupled System

We extend an optimal entanglement distillation of the triplet Greenberger-Horne-Zeilinger （GHZ） state via entanglement concentrating in the three-partite partially electron-spin-entangled systems. Two entanglement concen- tration protocols are similarly designed in detail with the post-selection in quantum-dot （QD） and micro-cavity coupled systems. The proposed protocol can be repeated several rounds to achieve an optimal success probability with an as- sistance of the ancillary QD, where only the single photon needs to pass through the micro-cavity for each round. It increases the total success probability of the distillation even if the implemented cavity is imperfect in practice during the whole process.     

A MOCVD-Growth Multi-Wavelength Laser Monolithically Integrated on InP

We present an arrayed waveguide grating multi-wavelength laser （MWL）. The device is operated with five wave- length channels of 194 GHz spacing around a central wavelength of 1.57μm. A side mode suppression ratio of better than 35 dB for all channels is demonstrated. A very attractive feature of the MWL is that it has been realized by a novel one step regrowth approach to achieve a high quality active and passive interface.