基于DIGNET网络的数据融合方法

针对数据融合和目标识别的特点,提出了基于DIGNET自组织聚类人工神经网络的数据融合方法。考虑到多传感器系统测量多个参量的特点,用并行的子网络结构代替中间隐层,实现了基于决策层的信息融合目标识别。利用仿真数据对基于DIGNET的数据融合方法进行了实验研究。实验结果表明,该方法具有数据正确分类率高和抗噪能力强等优点,有效地实现了融合识别。将该方法应用于前视红外和可见光双传感器目标跟踪系统的数据融合识别是可行的。     

Super-resolution image restoration algorithm based on orthogonal discrete wavelet transform

By using orthogonal discrete wavelet transform(ODWT)and generalized cross validation(GCV),and combining with Luck-Richardson algorithm based on Poisson-Markovmodel (MPML),several new superresolution image restoration algorithms are proposed.According to simulation experiments for practical images,all the proposed algor ithms could retain image details better than MPML,and be more suitable to low signal-to-noise ratio(SNR)images.The single operation wavelet MPML(SW-MPML)algorithm and MPML algorithm based on single operation wavelet transform(MPML-SW)avoid the iterative operation of self-adaptive parameter in MPML particularly,and improve operating speed and precision.They are instantaneous to super-resolution image restoration process and have extensive application foreground.     

光纤面板成象的理论分析和等效平均MTF的讨论

本文首先从积分抽样理论出发,阐述了六角形排列光纤面板传象的理论机理,然后对六角形排列光纤面板的等效平均MTF进行了分析和讨论·指出采用刀口响应曲线不能简单地转换成传递函数。     

光纤熔接机加热扩芯制作模场适配器的研究

为了改善不同类型光纤熔接时的模场失配,通过调整普通光纤熔接机的熔接参数,对模场直径较小的光纤进行加热扩芯,实现了10/130μm大模场面积双包层光纤和6/125μm单模光纤的低损耗熔接,光纤耦合效率可达到91%,并成功应用于自主研发的小型1064nm光纤激光器中。对利用光纤熔接机加热扩芯制作模场适配器进行了理论分析,并用1064nm光纤激光器测量其实际传输损耗。实验结果表明:采用普通光纤熔接机,适当的调整熔接参数,可以有效地提高大模场面积光纤到单模光纤的耦合效率,为制作模场适配器提供了一种简单实用的方法。     

被动调Q激光器优化的研究

被动调Q激光器以其结构简单、紧凑而获得了广泛的应用。为优化其性能参数,在总结近些年被动调Q基本理论基础上,对于具有能量上转换(Energy-transfer upconversion)的激光工作物质构成的被动调Q激光器,提出了可以将能量上转换引起的损耗归到线性损耗中进行优化设计的方法,并给出了输出镜的反射率和饱和吸收体的小信号透过率两者优化选择的直接关系图。根据被动调Q激光器的特有参数α,在总结提高调Q峰值功率的方法基础上,得出当α接近无穷大时,从被动调Q过渡到主动调Q,实现了被动调Q和主动调Q理论的统一。     

基于微分方程数值解的自由曲面透镜的设计

为满足目标面上均匀照明的需求,设计了一款自由曲面透镜。根据LED光源的发光特点,结合光学成像的特性,非成像光学原理和能量守恒定理,推导出实现光能量在目标面上均匀分布的自由曲面面形的微分方程,采用matlab的ode算法求出面形上的离散坐标点,对离散坐标点拟合后得到透镜模型,通过光学模拟仿真软件对透镜模型进行光线追迹。结果表明,配光角度为80°的透镜,透镜的口径与光源发光面宽度之比大于等于9时,目标面上的照度均匀性大于0.9,光能利用率约为85%。该设计方法可以快速精确地设计出所需的透镜,而且透镜结构紧凑,单颗就能实现均匀照明,有利于LED光源照明系统的小型化。     

Super-resolution image restoration algorithm based on orthogonal discrete wavelet transform

By using orthogonal discrete wavelet transform (ODWT) and generalized cross validation (GCV), and combining with Luck-Richardson algorithm based on Poisson-Markov model (MPML), several new superresolution image restoration algorithms are proposed. According to simulation experiments for practical images, all the proposed algorithms could retain image details better than MPML, and be more suitable to low signal-to-noise ratio (SNR) images. The single operation wavelet MPML (SW-MPML) algorithm and MPML algorithm based on single operation wavelet transform (MPML-SW) avoid the iterative operation of self-adaptive parameter in MPML particularly, and improve operating speed and precision. They are instantaneous to super-resolution image restoration process and have extensive application foreground.     

MAP-based infrared image expansion

Image expansion, as we know, is one of image processing algorithms that acquire new pixels from the original im- age by a certain method, so as to expand the size (pixel numbers) of the image. Image expansion is applicable in many aspects of image processing. To generate pre- cise maps, cartographers must expand small regions of satellite image data. In medical imaging, computerized tomography and X-rays need to be zoomed locally to find out foci of anomalies. Reconnaissance photographs must …     

一种新的自组织聚类网络及其在数据融合目标分类中的应用

介绍了一种新的自组织聚类人工神经网络(DIGNET)模型和网络的非监督学习算法。针对数据融合和目标识别的特点,提出了基于DIGNET的决策层数据融合目标分类方法。利用仿真数据研究了DIGNET和自组织特征映射网络(SOFM)的聚类性能以及基于DIGNET的决策层数据融合结构,实验结果表明DIGNET较SOFM正确分类率高、抗噪能力好;基于DIGNET的决策层数据融合能够有效地实现融合识别。将该数据融合方法应用于前视红外(FLIR)和可见光摄像机目标跟踪系统,结果表明该方法是可行的。