5G毫米波通信用低插损高隔离GaAs PHEMT单刀双掷开关

基于0.15μm GaAs赝配高电子迁移率晶体管(PHEMT)工艺,实现了一款用于5G毫米波通信的低插损高隔离单刀双掷(SPDT)开关芯片。为了降低插损,每个开关支路通过四分之一波长阻抗变换器连接到天线端,并通过优化传输线和器件总栅宽实现了良好的端口匹配;为了提高隔离度,采用了三并联多节枝的分布式架构形成高的输入阻抗状态,实现信号的全反射。芯片面积为2.1 mm×1.1 mm。在片测试结果显示,在24.25~29.5 GHz的5G毫米波频段内该SPDT开关实现了小于1.1 dB的极低插损和大于32 dB的高隔离度,1 dB压缩点输入功率大于26 dBm。     

Siamese visual tracking with multilayer feature fusion and corner distance IoU loss

The tracker based on the Siamese network regards tracking tasks as solving a similarity problem between the target template and search area. Using shallow networks and offline training, these trackers perform well in simple scenarios. However, due to the lack of semantic information, they have difficulty meeting the accuracy requirements of the task when faced with complex backgrounds and other challenging scenarios. In response to this problem, we propose a new model, which uses the improved ResNet-22 network to extract deep features with more semantic information. Multilayer feature fusion is used to obtain a high-quality score map to reduce the influence of interference factors in the complex background on the tracker. In addition, we propose a more powerful Corner Distance IoU (intersection over union) loss function so that the algorithm can better regression to the bounding box. In the experiments, the tracker was extensively evaluated on the object tracking benchmark data sets, OTB2013 and OTB2015, and the visual object tracking data sets, VOT2016 and VOT2017, and achieved competitive performance, proving the effectiveness of this method.     

一种基于YOLOv4-TIA的林业害虫实时检测方法

针对现有基于深度学习的林业昆虫图像检测方法存在检测精度低和检测速度慢的问题,提出一种结合改进PANet结构与三分支注意力机制的目标检测方法YOLOv4-TIA。通过对样本数量较少的昆虫类别进行数据增强,实现样本均衡分布。利用三分支注意力机制改进YOLOv4中的CSPDarkNet53骨干网络,同时通过旋转操作和残差变换建立维度间的依存关系,以提高有效的特征通道权重,在PANet结构上增加将跳跃连接与跨尺度连接相结合的特征融合方式,从而获取更丰富的语义信息和位置信息。在此基础上,采用Focal loss函数优化分类损失,解决正负样本不均衡的问题。实验结果表明,该方法的精确率和召回率分别达到85.9%和91.2%,相比SSD、Faster R-CNN、YOLOv4方法,其在保证检测速度的同时,能够有效提高检测精度,且实现对林业害虫的实时精确监测。     

一种抗背景干扰的多尺度人群计数算法

人群计数技术以估计人群图片或视频中的人数为目标,可以有效预防人群踩踏事故的发生,广泛应用于安防预警、城市规划及大型集会管理等领域。然而,由于人群尺度变化、背景干扰、人群分布不均、遮挡和透视效应等因素的影响,单幅图片的人群计数仍是一项非常具有挑战性的任务。针对人群计数中多尺度变化和背景干扰问题,提出一种抗背景干扰的多尺度人群计数算法。以VGG16网络结构为基础,引入特征金字塔构建多尺度特征融合骨干网络解决人群多尺度变化问题,设计Double-Head-CC结构对融合后的特征图进行前景背景分割和密度图预测以抑制背景干扰。基于密度图的局部相关性和多任务学习,定义多重损失函数和多任务联合损失函数进行网络优化。在ShanghaiTech、UCF-QNRF和JHU-CROWD++数据集上进行训练和评测,实验结果表明,该算法能够很好地预测人群密度分布和人群数量,具有较高的准确性,且鲁棒性强、泛化性能良好。     

基于循环生成对抗网络的含遮挡人脸识别

人脸图像的遮挡会严重影响人脸识别准确率,当前处理带遮挡人脸识别的方法主要有丢弃法和修复法两种。丢弃法因忽略或丢弃大量遮挡区域的有效特征易造成识别准确率不高,而当前大多数修复法需要原图的相关信息,限制了其应用。针对现有含遮挡人脸识别方法存在的问题,提出一种基于循环生成对抗网络的人脸识别算法,通过利用2对生成器和判别器进行循环训练,实现遮挡人脸图像的盲修复,此过程不需要除遮挡区域外的原图信息。在此基础上,采用ResNet-50网络对修复后人脸进行识别,该网络通过跳跃连接消除深层卷积神经网络中的网络退化问题,能够降低网络训练的难度,且不会增加额外的参数和计算量。针对盲修复后人脸特征存在类内差异大和类间差异小的特性,引入一种能够量化类间距离的分类损失函数RegularFace作为识别网络损失函数。实验结果表明,与DCGAN+CNN算法相比,所提算法对不同遮挡类型和遮挡区域的人脸图像识别准确率均有所提高,当线性遮挡面积为40%时,所提算法的识别准确率提高了14.4个百分点。     

Proposal of variable sequential sampling plan having desired operating characteristics indexed by quality loss

The proportion of non-conforming items has been traditionally utilised as an evaluation criterion for quality of items. However, the proportion of non-conforming items is not necessarily useful as a proper evaluation criterion for controlling high-quality manufacturing in recent years. Accordingly, in order to achieve further quality improvement and innovation, more careful quality evaluation has been required newly. Then, a concept of quality loss in the Taguchi methods has been devised as a severe criterion of quality evaluation. Hereby, a variable single sampling plan having desired operating characteristics (OCs) indexed by quality loss has been proposed in the area of statistical quality control. By the way, the most economical sampling inspection in the average sample number (ASN) is the sequential sampling plan based on the Wald’s sequential probability ratio test. Then, from the viewpoint of cost reduction, we discuss a variable sequential sampling plan having desired OC indexed by quality loss with the aim of expansion of the utility of variable sampling plan for quality loss. As the result, the design procedure of the sequential sampling plan for satisfying some required design conditions indexed by quality loss is provided. In addition, the effectiveness of the proposed sequential sampling plan is verified through some numerical examples.     

Optimization of correlated multi-response quality engineering by the upside-down normal loss function

Most of the published literature on robust design is basically concerned with a single response. However, the reality is that common industrial problems usually involve several quality characteristics, which are often correlated. Traditional approaches to multidimensional quality do not offer much information on how much better or worse a process is when finding optimal settings. Köksoy and Fan [Engineering Optimization 44 (8): 935–945] pointed out that the upside-down normal loss function provides a more reasonable risk assessment to the losses of being off-target in product engineering research. However, they only consider the single-response case. This article generalizes their idea to more than one response under possible correlations and co-movement effects of responses on the process loss. The response surface methodology has been adapted, estimating the expected multivariate upside-down normal loss function of a multidimensional system to find the optimal control factor settings of a given problem. The procedure and its merits are illustrated through an example.     

Influence of ZrO2 particles on the optical properties of pc-LEDs

Experimental and modeling progress and results aiming to increase the color uniformity of hemisphere-type pcW-LEDs are proposed. By adding micrometric zirconia particles, the light scattering is enhanced, that induces a decrease of phosphor necessary to obtain a specific CCT. The optical model is able to determine the optical properties (CCT, angular CCT distribution, chromaticity and packaging efficiency) in a hemisphere LED for various amount of YAG-phosphor and ZrO₂-zirconia particles. Based on previous process, the work is in the first step to fit the effective radius and refractive index that will be implemented in the optical model. In the second step, the optical model is compared with the experimental measurement to determine the absorption coefficient and conversion efficiency of a package of silicone resin containing phosphor and zirconia. Finally, the model, confirmed by selected experimental results, allows determining the optical properties of any king of package that lead to a CCT emission ranging from 4500 K to 6500 K. The analysis of these data is interpreted by comparing, for a specific CCT, the phosphor loss, the packaging efficiency and the angular distribution of CCT. The results show that if the effort is targeted on one of the previous parameters, there is always a counterpart on the other ones. Increasing the color uniformity will induce to decrease the packaging efficiency.     

TNT基炸药储存寿命评估研究?

为解决有效评估炸药储存寿命的难题,提出一套基于炸药加速退化试验失重率阈值的储存寿命评估技术。采用基于伪失效寿命的加速寿命试验数据建模与处理方法,对某TNT基炸药加速退化试验数据,使用正态分布和威布尔分布模型,估计了两种模型参数,评定了其储存寿命。研究表明,该评估技术可行,结论可信。