改进HSR-FCN的服装图像识别分类算法研究

目前网络上的服装图像数量增长迅猛，对于大量服装图像实现智能分类的需求日益增加。将基于区域的全卷积网络（Region-Based Fully Convolutional Networks，R-FCN）引入到服装图像识别中，针对服装图像分类中网络训练时间长、形变服装图像识别率低的问题，提出一种新颖的改进框架HSR-FCN。新框架将R-FCN中的区域建议网络和HyperNet网络相融合，改变图片特征学习方式，使得HSR-FCN可以在更短的训练时间内达到更高的准确率。在模型中引入了空间转换网络，对输入服装图像和特征图进行了空间变换及对齐，加强了对多角度服装和形变服装的特征学习。实验结果表明，改进后的HSR-FCN模型有效地加强了对形变服装图像的学习，且在训练时间更短的情况下，比原来的网络模型R-FCN平均准确率提高了大约3个百分点，达到96.69%。     

基于张量分解融合RGB-D图像的物体识别

为了充分利用RGB-D图像的深度图像信息，提出了基于张量分解的物体识别方法。首先将RGB-D图像构造成一个四阶张量，然后将该四阶张量分解为一个核心张量和四个因子矩阵，再利用相应的因子矩阵将原张量进行投影，获得融合后的RGB-D数据，最后输入到卷积神经网络中进行识别。RGB-D数据集中三组相似物体的识别结果表明，利用张量分解融合RGB-D图像的物体识别准确率高于未采用张量分解的物体识别准确率，并且单一错分实例的准确率最高可提升99%。     

Parameter estimation for a morphochemical reaction-diffusion model of electrochemical pattern formation

The process of electrodeposition can be described in terms of a reaction-diffusion partial differential equation (PDE) system that models the dynamics of the morphology profile and the chemical composition. Here we fit such a model to the different patterns present in a range of electrodeposited and electrochemically modified alloys using PDE constrained optimization. Experiments with simulated data show how the parameter space of the model can be divided into zones corresponding to the different physical patterns by examining the structure of an appropriate cost function. We then use real data to demonstrate how numerical optimization of the cost function can allow the model to fit the rich variety of patterns arising in experiments. The computational technique developed provides a potential tool for tuning experimental parameters to produce desired patterns.     

Specific emitter identification for satellite communication using probabilistic neural networks

Electromagnetic signal emitted by satellite communication (satcom) transmitters are used to identify specific individual uplink satcom terminals sharing the common transponder in real environment, which is known as specific emitter identification (SEI) that allows for early indications and warning (I&W) of the targets carrying satcom furnishment and furthermore the real time electromagnetic situation awareness in military operations. In this paper, the authors are the first to propose the identification of specific transmitters of satcom by using probabilistic neural networks (PNN) to reach the goal of target recognition. We have been devoted to the examination by exploring the feasibility of utilizing the Hilbert transform to signal preprocessing, applying the discrete wavelet transform to feature extraction, and employing the PNN to perform the classification of stationary signals. There are a total of 1000 sampling time series with binary phase shift keying (BPSK) modulation originated by five types of satcom transmitters in the test. The established PNNs classifier implements the data testing and finally yields satisfactory accuracy at 8 dB(±1 dB) carrier to noise ratio, which indicates the feasibility of our method, and even the keen insight of its application in military.     

基于小波变换的分数阶微分算法在肝脏肿瘤CT图像纹理增强中的应用

图像纹理增强过程中容易丢失平滑区域纹理细节，而分数阶微分增强虽然能够非线性保留平滑区域纹理细节，但对频率分辨率敏感。针对这个问题，提出一种基于小波变换的分数阶微分纹理增强算法，应用于平扫计算机断层扫描（CT）图像的肝脏肿瘤区域的纹理增强。首先，通过小波变换将图像感兴趣区分解成多个子带分量；其次，基于分数阶微分定义构造一个带补偿参数的分数阶微分掩膜；最后，使用该掩膜与每个高频子带分量进行卷积并利用小波逆变换重组图像感兴趣区。实验结果表明，该方法在使用较大分数阶次显著增强肿瘤区域的高频轮廓信息的同时，有效地保留了低频平滑的纹理细节：增强后的肝细胞癌区域与原区域相比，信息熵平均增加36.56%，平均梯度平均增加321.56%，平均绝对差值平均为9.287；增强后的肝血管瘤区域与原区域相比，信息熵平均增加48.77%，平均梯度平均增加511.26%，平均绝对差值平均为14.097。     

结合位图切割和区域合并的彩色图像分割

就经典分水岭图像分割算法中存在的过分割问题，提出一种结合位图切割和区域合并的彩色图像分割算法。对原始彩色图像通过空域梯度算子求其梯度图像，并利用位图切割重建梯度图像；对新梯度图像进行分水岭预分割；对预分割图像基于异质性最小原则进行区域合并，并获得最终分割结果。相比于现有的同类方法，该算法引入位图切割，抑制噪声对分割结果的影响，在边缘模糊处分割准确，得到符合人类视觉的较小分割区域数目，同时在运行效率上提高。     

应用膨胀算法提取目标的直接定位算法

为了解决被动雷达系统中的多发射源定位问题，提出了一种基于多重信号分类(MUSIC)算法和图像膨胀(IE)算法的直接定位方法。该方法结合了谱分析中的MUSIC思想，通过对接收量测协方差矩阵进行特征分析求解目标的位置。首先，在目标个数未知的前提下，利用Akaike信息准则(AIC)来确定模型阶数；然后，推导了基于MUSIC的定位代价函数；之后，利用图像膨胀算法处理得到的代价函数平面；最后，膨胀处理后的输出为目标个数及目标位置的估计值。提出的算法有效地解决了目标检测及提取的问题，能够确定多个目标的位置坐标，为后续的定位性能分析提供可能性，也保证了算法的完整性。进一步地分析了多个临近目标情况下影响目标提取性能的主要因素。     

基于时空的弱点状运动目标检测技术

如何在复杂背景下持续有效地检测目标位置,一直是研究者们需要面对的主要挑战。本文在研究红外点状移动目标特征的基础上,根据目标无纹理,无形状的特性,提出一种改进的形态学目标增强算法,并利用目标连续时空不变性检测目标。首先,建立多尺度的图像金字塔,在每层上采用改进的形态学算法快速、粗糙定位小目标。然后进一步的根据目标在时空上的位置相关性,提出基于目标运动特征分析的精确检测方法。得到精确稳定的检测结果。最后实验结果表明与经典的形态学检测算法及其他算法相比,该技术能更有效地检测弱小目标,具有更高的鲁棒性。     

Multi-innovation auto-constructed least squares identification for 4 DOF ship manoeuvring modelling with full-scale trial data

This research is concerned with the problem of 4 degrees of freedom (DOF) ship manoeuvring identification modelling with the full-scale trial data. To avoid the multi-innovation matrix inversion in the conventional multi-innovation least squares (MILS) algorithm, a new transformed multi-innovation least squares (TMILS) algorithm is first developed by virtue of the coupling identification concept. And much effort is made to guarantee the uniformly ultimate convergence. Furthermore, the auto-constructed TMILS scheme is derived for the ship manoeuvring motion identification by combination with a statistic index. Comparing with the existing results, the proposed scheme has the significant computational advantage and is able to estimate the model structure. The illustrative examples demonstrate the effectiveness of the proposed algorithm, especially including the identification application with full-scale trial data.     

Digital image correlation strain measurement of thick adherend shear test specimen joined with an epoxy film adhesive

Structural bonding and bonded repairs of composite materials become more and more important. Understanding the strain within the bondline leads to suitable bonding design. For new design approaches the strain distribution within the bondline has to be analyzed. Thus, often finite element analysis (FE) are used. However, a huge challenge is the availability of reliable material properties for the adhesives and their validation. Previous work has shown that it is possible to measure the small displacements resulting within thin epoxy film adhesives using high resolution digital image correlation (DIC). In this work a 2D DIC setup with a high resolution consumer camera is used to visualize the strain distribution within the bondline over the length of the joint as well as over the adhesive thickness. Therefore, single lap joints with thick aluminum adherends according to ASTM D 5656 are manufactured and tested. Local 2D DIC strain measurements are performed and analyzed. Two different camera setups are used and compared. The evaluation provides reliable material data and enables a look insight the bondline. The results of the full field strain data measured with DIC are compared with numerical simulations. Thus, material models as well as chosen parameters for the adhesive are validated. Compared to extensometers, giving only point-wise information for fixed measuring points, the DIC allows a virtual point-wise inspection along the complete bondline. Furthermore, it allows measuring close to the bondline to reduce the influence of adherend deformation.