脊柱侧弯是当今社会中常见的脊柱疾病，在X光图像上快速而准确地定位脊椎骨角点并计算其Cobb角度数是医生诊断脊柱弯曲程度的金指标。针对X光骨科图片中其他器官的遮挡以及复杂背景干扰等问题，提出一种基于嵌入注意力机制和向量损失模块的神经网络模型。所提模型以vertebra-focused landmark detection（VFLD）网络为基础网络，在编码器和解码器之间嵌入旋转注意力机制模块加强网络对于脊椎骨深层、高维特征的提取，抑制其他器官的干扰，同时利用向量相似性的损失函数对网络进行训练。实验结果表明，在MICCAI 2019公开脊椎挑战赛数据集中，所提模型的对称平均绝对百分比误差准确度高达9.31，可以有效提高原模型检测椎骨角点能力。与现有的诸多模型相比，其具有较高的准确率和稳健性。

Transparency and perfect absorption are two contradictory terms; a perfect absorber never permits waves to transmit through. However, this statement only remains true in the linear regime, where the nonlinearity has been omitted and the physical system like the perfect absorber is not affected by the incoming waves. Here we experimentally demonstrate an intriguing self-induced transparency effect in a perfectly absorbing optical microcavity, which perfectly absorbs any incoming waves at the low power level, but allows a portion of waves to be transmitted at the higher power due to the nonlinear coupling between the fundamental and its second harmonic modes. Moreover, the asymmetric scattering nature of the microcavity enables a chiral and unidirectional reflection in one of the input ports, this leads to asymmetric and chiral coherent control of the perfect absorption states through phase varying. More importantly, such chiral behaviors also empower the chiral emission of second-harmonic generation with a high distinct ratio in the transparency state. These results pave the way for controllable transparency in a wide range of fields in optics, microwaves, acoustics, mechanics, and matter waves.

基于立方星辐射计直接检波系统，设计并实现了中心频率分别为89 GHz和150 GHz的两款毫米波零偏置肖特基检波器。检波器设计基于ACST的零偏置肖特基二极管。为了获得稳定、优越的性能以及较好的宽带特性，并易于与前级系统集成，更适于立方星辐射计小型化，在直流接地电路与输出电路处采用扇形线结构，并通过一段可调传输线对其端口阻抗进行优化。本文对肖特基二极管及检波器电路结构进行了分析、建模和优化。测试结果表明，W波段检波器在85 ~ 95 GHz范围内具有2500 V/W的典型灵敏度，在89 GHz测试的线性度为0.9994。D波段检波器在145 ~ 155 GHz范围内的典型灵敏度约为1600 V/W，在150 GHz测试的线性度为0.9992。这些结果验证了改进电路结构在检波器设计上的优越性和在直接检波系统中应用的可行性。

视网膜血管的形态结构具有复杂多变的特点，针对血管图像中的交叉处和延伸处不易分割的问题，提出一种基于多方向滤波的视网膜血管分割算法，该算法进一步提升了血管图像的分割精度。首先，使用直方图均衡化、中值滤波去噪、顶帽变换等方法对绿通道的视网膜血管图像进行增强。然后，对增强后的图像进行多方向的Cake滤波，并对滤波后的结果进行融合，弱化背景中的噪声，增强血管与背景的对比度。最后，采用向量场散度的方法提取阈值并对图像进行分割，得到视网膜血管分割的最终结果。利用公开数据集DRIVE和STARE测试了算法的分割精度，实验结果表明，所提算法能够精确地分割出复杂血管的连接处，灵敏度较高，且执行时间较短。

四足动物行走的运动特征自动识别在动物的仿生学、行为识别、疾病预测，以及个体身份识别等方面都有着非常广阔的应用前景。基于计算机视觉技术，结合深度学习方法，针对四足动物的行走运动，建立了一种四足动物行走步态参数自动提取方法。利用视频帧分解技术对采集的四足动物行走视频进行处理，得到各帧四足动物行走图像；基于改进DeeplabV3+语义分割模型提取运动目标；在此基础之上，结合四足动物行走步态及其时空特性分析，以目标体轮廓中心点-边界距离曲线为基础实现运动角点的检测及匹配。为有效地解决四足动物运动特征参数提取的问题，建立了一种基于四肢运动角点至参考点距离变化曲线的运动特征分析方法。实验结果表明，所提方法能较好地实现四足动物运动角点的检测，对犀牛、水牛、羊驼运动角点检测的最大误差分别为32，27，19 pixel，运动角点匹配关系准确，步态周期、步态频率的计算误差小于2%，步态顺序输出正确，步幅的计算误差最大为2.85%。

Additive manufacturing (AM) technology is considered one of the most promising manufacturing technologies in the aerospace and defense industries. However, AM components are known to have various internal defects, such as powder agglomeration, balling, porosity, internal cracks and thermal/internal stress, which can significantly affect the quality, mechanical properties and safety of final parts. Therefore, defect inspection methods are important for reducing manufactured defects and improving the surface quality and mechanical properties of AM components. This paper describes defect inspection technologies and their applications in AM processes. The architecture of defects in AM processes is reviewed. Traditional defect detection technology and the surface defect detection methods based on deep learning are summarized, and future aspects are suggested.

Due to the rapid development of precision manufacturing technology, much research has been conducted in the field of multisensor measurement and data fusion technology with a goal of enhancing monitoring capabilities in terms of measurement accuracy and information richness, thereby improving the efficiency and precision of manufacturing. In a multisensor system, each sensor independently measures certain parameters. Then, the system uses a relevant signalprocessing algorithm to combine all of the independent measurements into a comprehensive set of measurement results. The purpose of this paper is to describe multisensor measurement and data fusion technology and its applications in precision monitoring systems. The architecture of multisensor measurement systems is reviewed, and some implementations in manufacturing systems are presented. In addition to the multisensor measurement system, related data fusion methods and algorithms are summarized. Further perspectives on multisensor monitoring and data fusion technology are included at the end of this paper.