基于深度学习的手写书法字体识别算法

为解决因手写书法作品种类繁多而识别困难的问题,降低人们观赏书法的门槛,本文提出了基于深度学习的手写书法字体识别算法.识别过程中首先使用投影法等图像处理方法对书法作品图像中的汉字进行定位和分割,然后分别利用GoogLeNet Inception-v3模型和ResNet-50残差网络进行书体风格识别和字形识别.实验结果表明...     

Solution-Processed and Room-Temperature Spin Light-Emitting Diode Based on Quantum Dots/Chiral Metal-Organic Framework Heterostructure

Spin optoelectronics is an indispensable key for the future development of spintronics. In conventional spin light emitting diodes (LEDs), spin-polarized carrier pairs are injected electrically into the light emitting layer and create circularly polarized light (CPL). Generally, spin-polarized carriers are accomplished using ferromagnetic contacts or applying an external magnetic field, which will produce several drawbacks, including low temperature operation, low spin-polarized carriers injection efficiency, etc. To circumvent the existing shortcomings, here, an alternative approach is proposed and achieves spin-polarized LEDs at room temperature based on quantum dots (QDs)/chiral metal-organic framework heterojunction without using ferromagnetic contacts or magnetic fields. The spin-polarized injected layer composed of self-assembled monolayer (SAM)/Chiral-MOF ([Sr(9,10-adc)(DMAc)2]n)) film, which produces spin-polarized holes with spin orientation, determining the polarization and strength of circularly polarized electroluminescence (CP-EL). The spin-QLED emits CP-EL at a rate of 12.24% efficiency, which provides an excellent alternative to generate new functionality for conventional QLEDs. The approach is anticipated to be very useful, enabling to offer a general methodology for generating not yet realized spin optoelectronic devices.     

基于中心移动的轨迹离群点检测

AIS数据是指通过AIS系统获取的船舶运动轨迹信息, 对其进行挖掘可以获得船舶的运动模式、航行路线、停靠地点等信息. 但其在采集过程中产生的离群点会对聚类等任务造成负面影响, 因此对AIS数据挖掘之前需要进行离群点检测. 然而, 当AIS轨迹数据中存在大量离群点时, 会导致大多数离群点检测算法的准确率显著下降. 为了解决这个问题, 本文提出了一种基于中心移动的轨迹离群点检测算法(center shift outlier detection, CSOD). 通过迫使数据点向其K近邻集合的中心移动, 使每个数据点更加接近典型数据, 从而有效地消除了离群点对聚类的影响. 为了验证本文算法的有效性, 使用浙江海域AIS渔船轨迹数据集, 将本文提出的CSOD算法与一些经典的离群点检测算法进行了对比实验. 实验结果表明, CSOD算法整体上性能更加优越.