Transfer-Learning-Based Approach for the Diagnosis of Lung Diseases from Chest X-ray Images

Using chest X-ray images is one of the least expensive and easiest ways to diagnose patients who suffer from lung diseases such as pneumonia and bronchitis. Inspired by existing work, a deep learning model is proposed to classify chest X-ray images into 14 lung-related pathological conditions. However, small datasets are not sufficient to train the deep learning model. Two methods were used to tackle this: (1) transfer learning based on two pretrained neural networks, DenseNet and ResNet, was employed; (2) data were preprocessed, including checking data leakage, handling class imbalance, and performing data augmentation, before feeding the neural network. The proposed model was evaluated according to the classification accuracy and receiver operating characteristic (ROC) curves, as well as visualized by class activation maps. DenseNet121 and ResNet50 were used in the simulations, and the results showed that the model trained by DenseNet121 had better accuracy than that trained by ResNet50.     

由傅里叶功率谱恢复物函数的研究

相位恢复工作中,针对物函数是实函数的情形,Fienup算法取得了较大的成功.本文给出一新算法,其收敛程度比Fienup算法更优.以天文学中双星情形为例,比较了两种不同算法的结果.     

Numerical demonstration of three-dimensional terahertz metamaterials based on the causality principle

A parameter retrieval algorithm based on the causality principle and Kramers-Kronig （KK） relations is employed to calculate the effective parameters of three-dimensional （3D） metamaterials. Using KK relations, the branch selecting problem, which is the challenge of effective parameter retrieval method, can be removed. To reveal the validity of the proposed algorithm, the constitutive refractive index of a homogeneous polymide cube is extracted. The result is in excellent agreement with the intrinsic refractive index of the polymide. Finally, the two terahertz metamaterials with 3D structures are designed and their effective parameters are then retrieved using the proposed algorithm. Numerical simulations are performed using the fuiI-wave electromagnetic solver, CST Microwave Studio.     

基于级联相位恢复算法的光学图像加密

在虚拟光学数据加密理论模型的基础上,提出了一种光学图像加密的可视化密码构造算法。该加密算法基于自由空间传播的光学系统,利用级联迭代角谱相位恢复算法把待加密图像分别编码到两块相位模板之中,从而实现图像的加密。该加密技术不但可通过同时调整两块相位模板的相位分布的搜索策略来扩大搜索空间,提高安全强度,而且扩大了系统密钥空间,使系统获得更高的安全性,且能通过简单的数值运算或光学实验装置得到质量非常高的解密图像,还从理论上分析了该算法的时间复杂度。计算机模拟结果表明,该加密算法的收敛速度快,能迅速找到非常好的近似解,解密图像质量高且系统安全性良好。     

Densely Connected Neural Networks for Nonlinear Regression

Densely connected convolutional networks (DenseNet) behave well in image processing. However, for regression tasks, convolutional DenseNet may lose essential information from independent input features. To tackle this issue, we propose a novel DenseNet regression model where convolution and pooling layers are replaced by fully connected layers and the original concatenation shortcuts are maintained to reuse the feature. To investigate the effects of depth and input dimensions of the proposed model, careful validations are performed by extensive numerical simulation. The results give an optimal depth (19) and recommend a limited input dimension (under 200). Furthermore, compared with the baseline models, including support vector regression, decision tree regression, and residual regression, our proposed model with the optimal depth performs best. Ultimately, DenseNet regression is applied to predict relative humidity, and the outcome shows a high correlation with observations, which indicates that our model could advance environmental data science.     

瑞利散射多普勒激光雷达风场反演方法

介绍了瑞利散射多普勒激光雷达的风场探测原理和系统结构. 给出了Fabry-Perot(FP)标准具透过率曲线的校准方法. 指出对透过率采用Lorentz或Voigt拟合会产生较大误差,特别是采用Lorentz拟合最大将引起8%的误差. 提出了采用非线性最小二乘法拟合标准具的透过率函数,该方法可以有效消除拟合误差,提高风速测量精度. 考虑到温度不确定度在风场反演过程中的主导影响,提出了同时反演风速和大气温度的非线性迭代算法. 风场反演仿真试验结果表明:在不考虑米散射信号的影响下,该反演算法与传统的反演方法相 关键词： 激光雷达 瑞利散射 多普勒 风     

Double-image self-encoding and hiding based on phase-truncated Fourier transforms and phase retrieval

We propose a method to encrypt two covert images into an overt image based on phase-truncated Fourier transforms and phase retrieval. In this method, the two original images are self-encoded in the manner that one of the two images is directly separated into two phase masks (PMs) and used as keys for encryption, and then multiplied by a PM which is generated by using phase retrieval algorithm. At last, the whole encryption process is completed by a Fourier transform operation. In the decryption process, the image without a separation and the two PMs used as keys for encryption are all treated as encoded data. The cryptosystem is asymmetric which means the keys for encryption are different from those for decryption. Numerical simulations are presented to show the viability and good performance of the proposed method.     

高帧频信标光定位与跟踪实验研究

信标光斑准确定位是完成空间光通信跟踪和捕获时的一个重要方法。应用改进的Otsu分割方法,对采集序列图像进行实时处理,提取潜在的信标光斑,对后续图像处理引入感兴趣区域的读入技术,在感兴趣区域内对信标光图像进行处理,应用卡尔曼预测模型实现了信标光跟踪,并将跟踪误差控制在精跟踪视场之内。实验结果表明,改进的Otsu具有更佳的抗噪能力和分割效果,并且实时性和定位精度均获得了满意的结果。     

用SHG-FROG方法测量超短光脉冲的振幅和相位

介绍了二次谐波振荡频率分辨光学快门(SHGFROG)测量超短脉冲的实验系统和二维相位恢复算法,这种方法可以实现对超短脉冲振幅和相位的完全测量.利用矩阵方法数值模拟了几种常见超短脉冲的SHGFROG迹线,并用主元素广义投影算法(PCGPA)从这些无噪音的SHGFROG迹线中恢复了脉冲的振幅和相位,误差接近收敛的标准.研究结果表明,这种测量方法结构简单,算法可靠,是测量超短脉冲的理想方法之一.     

连续多阈值神经元反馈神经网络盲检测光基带信号

空间分集接收可补偿信道衰落,提出了一种基于幅相联合激励法的连续多阈值神经元反馈神经网络(RNNCMVN)的光基带信号直接盲检测方法。针对多进制相移键控(MPSK)信号的特点,设计了两种连续相位多阈值激励函数形式,并简要讨论了该两类激励函数参数的选择;分别推演基于幅相联合激励法的RNNCMVN光基带信号盲检测方法工作于同步和异步两种模式下的新能量函数及其相关证明。同时针对正交调幅(QAM)信号的特点,分别设计出连续振幅和相位多阈值激励函数形式,最后通过仿真验证了该方法的有效性。     

Optical Learning Neural Network Using Photorefractive Waveguides

A model of an optical neural network with learning ability is proposed. We numerically evaluate the learning ability of the proposed network by using parameters determined by experiments. Adaptive connections between artificial neurons are implemented using photorefractive (PR) waveguides that can be optically modified by guided beams. The network consists of three layers and has bipolar weights within the limited range. The bipolar weight is encoded as the difference between optical power transmittances of signal beams in two channels of the PR waveguides. The adaptivity of the transmittance of PR waveguide is experimentally evaluated and is incorporated into the proposed network simulated in a computer. The proposed network is trained by a simplified local learning algorithm. Numerical results showed that the proposed three-layered network with six hidden neurons can solve the exclusive-or problem.     

一种基于瞳面相位差的波前传感器相位恢复

根据瞳面相位差波前传感器的特点,提出把模式分解和随机并行梯度下降算法结合起来实现波前相位的恢复.以32单元变形镜的初始面形和Roddier提出的相位生成方法随机产生的一帧相屏为研究对象,分析所提出的相位恢复算法的性能.结果表明,两种情况下的畸变波前都得到了很好的恢复.当像差仅含有低阶成分时,使用模式分解法就可获得令人满...     

基于光谱曲线形态的高光谱影像检索方法研究

随着传感器技术、数据通讯技术的飞速发展,利用各种机载和星载传感器,己经获取到各种不同的海量遥感影像数据。巨大的数据量带来了数据存储和管理的问题,如何实现从海量影像数据中检索出我们所需要的信息显得十分迫切。影像检索最早由Chang于1980年提出,是对传统信息检索的扩展。针对海量遥感影像高效检索的需求和高光谱遥感影像波段数目多的特点,分析了影像检索中的影像距离函数和相似性度量问题,基于经典的曲线简化Douglas-Peucke算法(简称DP算法)提取光谱曲线的形态特征,利用“提取特征”的思想,提出了基于DP算法的光谱曲线和影像检索(简称DPSR)方法,将光谱形态特征应用于影像检索当中。DPSR利用光谱曲线上的特征点,减小了计算量,实现了有效地匹配和检索,适合高光谱遥感影像的光谱检索。文章选择了OMISI高光谱数据的四种易混分地类进行了相似性度量的对比实验。通过与常规的分析方法光谱角匹配(SAM)、光谱信息散度(SID)的对比可以看到,DPSR在较少计算量的情况下能保持较高的计算精度,提供了一种新的影像光谱高效检索方法。此外,文章还提出了尚待进一步研究的问题。     

Two-stage deep learning-based hybrid precoder design for very large scale massive MIMO systems

Wireless networking is approaching a new era, which necessitates new frequency ranges and novel strategies. With recent circuit growth, communications over the Terahertz (THz) band is proving to be a viable option because of the tremendous bandwidth and low cost. On the other hand, THz signals suffer from significant direction loss, necessitating the use of precoding. In this paper, Deep Learning (DL) based precoding techniques for upcoming 6G networks were examined, along with their complexities. Based on the signal-to-noise ratio (SNR) and spectral efficiency (SE), the proposed DL-based precoding scheme is compared to traditional model-based precoding schemes. The proposed DL-based precoding technique is ideal for 6G networks, according to simulation results. Furthermore, the proposed DL-based precoding technique has lower computational complexity, making it suitable for parallel processing and high-speed data transmission.     

Fast phase retrieval in slightly off-axis digital holography

In this study, three efficient algorithms are proposed for fast phase retrieval in slightly off-axis digital holography using spectrum cropping, spatial multiplexing, and complex encoding. In the first algorithm, the real spectral order of the subtracted hologram is filtered and cropped, and the number of pixels is decreased in the subsequent retrieval operations. In the second algorithm, two sequential subtracted holograms are digitally phase shifted and spatial multiplexed into one synthetic hologram, and thus only one inverse Fourier transformation is then required. In the third algorithm, two sequential subtracted holograms are encoded separately into the real part and the imaginary part of a complex hologram. Two cross-correlations can be used to reconstruct the phase, thereby improving the utilization of the spectrum. The three new algorithms speed up our previously proposed retrieval method with the assistance of specimen-free holograms. Our experiments demonstrated the validity and improved time requirements of the proposed methods.     

光学加工过程中高次非球面的三坐标测量数据处理

三坐标轮廓检测是大口径高次非球面确定性加工过程中的主要面形测量手段。由于原始三坐标数据包含较大的检测误差,无法直接应用于加工过程,本文提出了一组数据处理算法对误差进行全面去除。首先,对获取的检测数据采用基于球心曲面重建的测头半径补偿算法进行测头半径误差补偿,然后对补偿后数据进行坐标系旋转平移误差去除,最后对提取的检测面形残差进行基于KNN的残差噪点过滤。其中,提出的基于球心曲面重建的测头半径补偿算法通过引入一个高精度的测头球心包络面拟合模型,来计算各检测点的测头半径补偿向量,仿真实验证明:算法补偿精度达到RMS<4 nm;提出的基于KNN的残差噪点过滤算法,通过采用插值方法提高样本空间密度和优化噪声度量值的计算,提高了噪点的识别敏感度并实现了噪点的自动化去除。最终根据整个误差清理算法构建了检测点云处理软件,应用实践表明其有效提高了镜面加工过程中检测点云的数据处理精度和效率。     

采用加权函数修正的差分光学吸收光谱反演环境大气中的CO2垂直柱浓度

采用加权函数修正的差分光学吸收光谱技术(weighting function modified differential optical absorption spectroscopy, WFM-DOAS)测量环境大气中的CO₂垂直柱浓度. 以直射太阳光测量光谱为例, 演示了WFM-DOAS算法的实现过程. 将环境大气分为50层, 借助辐射传输模拟软件SCIATRAN对各个测量光谱进行了模拟计算, 获得了最小二乘法拟合所需的目标气体CO₂及干扰气体H₂O、CH₄ 的柱权重函数和太阳归一化光谱常量. 采用WFM-DOAS方法对一整天的直射太阳光测量光谱进行了反演, 得到的反演误差均小于3%. 最后比较了两种不同DOAS算法对同一条测量光谱的反演结果, 验证了WFM-DOAS算法在红外被动气体遥感的优越性. 关键词： 环境污染监测 光学测量技术 红外光谱 温室气体     

前向迭代多随机相位光学图像加密方法的研究

基于光学4f级联系统,采用前向迭代算法和通过产生多个随机相位板对图像进行了加密和解密。给出了该方法的原理和迭代实现的过程。对该方法在收敛速度、解密图像质量、相位板数目和相位量化阶数对解密图像质量的影响以及保密和安全性能等方面进行了分析和研究。模拟结果表明,该方法收敛速度快,解密图像质量好,可以降低相位板相位量化阶数有限所引起的误差对解密图像质量的影响,图像的安全和保密性能也得到了很大的提高。     

C波段机载合成孔径雷达海面风场反演新方法

针对基于散射计地球物理模型函数的机载合成孔径雷达(SAR)海面风场反演中存在的风向获取依赖于图像风条纹或数值预报、 散射计数据和浮标等背景场资料, 风向与SAR图像时空分辨率不匹配, 进而影响机载SAR海面风场反演精度等问题, 本文根据机载SAR对海探测特点, 研究一种适用于C波段机载SAR的海面风场反演新方法. 利用SAR图像距离向不同入射角的后向散射系数, 依据地球物理模型构造最小代价函数, 通过代价函数的求解直接从机载SAR数据同时反演出海面风速和风向. 利用论文提出的海面风场反演方法分别对仿真SAR数据和实测C波段机载SAR数据进行风向、 风速的反演误差分析及试验验证.研究结果表明, 该方法适用于机载SAR海面风场反演, 可不依赖背景风向直接反演出精度较高的风速和风向; 雷达后向散射系数误差是决定海面风速、风向反演精度的关键因素, 辐射定标精度越高则反演误差越小; 海面风速反演误差随着风速的提高而增大, 当海面风速大于18 m/s时, 风速反演误差显著增加, 而海面风向的反演误差与风速无明显关系. 关键词： 机载合成孔径雷达 海面风场 多入射角     

Main melody extraction from polyphonic music based on modified Euclidean algorithm

Extracting main melody from polyphonic music is one of the most appealing and challenging tasks in music information retrieval (MIR). In this paper, a new melody extraction method based on a modified Euclidean algorithm (MEA) is proposed. Firstly, the instantaneous frequency is adopted to gain better frequency discrimination, and the frame-wise pitch candidates are estimated based on the modified Euclidean algorithm. Next, the candidate trajectories are formed using these potential candidates, and padded by the candidate one octave above or below if there is a gap at some isolated frames. Finally, the melodic contours are extracted using the melody smoothness and salience principle. The proposed modified Euclidean algorithm can deal with diverse coprime harmonic combinations, and work well at low computational cost and memory requirement. The experimental results show that the proposed method can extract main melody extraction effectively with few pitch candidates.