【作者】 王成安；

【导师】 谈和平；

【作者基本信息】 哈尔滨工业大学 ， 工程热物理， 2008， 硕士

【摘要】 高性能的探测设备需要提供更精细、包含更多数据量的目标信号特征值,为探测弱小目标以及伪装目标提供技术支撑。超光谱信息具有高分辨率特点,这样可以通过处理超光谱信息中目标信号的空间特征和光谱特征,以较高的可信度辨别和区分复杂背景下的弱小信号源。随着数值仿真的迅速发展,虚拟设计作为一门新兴的综合技术学科逐步地应用到航天科技应用领域,可以大大减少研制开销,并提高准确率,缩短开发周期。本文研究了红外波段8~12μm(即833~1250cm -1)下的地物背景的仿真和场景中目标识别问题,主要研究内容如下:1.将红外波段下的地物背景作为研究对象进行仿真计算,获得超光谱遥感图像,这其中包括自然地形以及地物场景的生成,地表的温度模型求解以及大气对辐射能量传输过程影响。2.应用热流法求解辐射传递方程,获得喷焰辐射特性数据,并将计算得到的目标辐射特性数据结果添加到背景中,获得8~12μm波段地物目标的超光谱图片。3.通过计算目标与背景的各波段信号相对比值,初步确定用于目标识别的波段,应用神经网络方法对目标在红外波段进行探测识别,找到了适用于8~12μm目标识别的最佳波段组合以及神经网络方法LS_SVM。4.针对用于红外目标识别的部分连续超光谱图片(80~106)进行相关性分析,计算图片之间的相关信息熵。应用图像融合方法处理相关性强的超谱图片,并对融合前后的图片应用LS_SVM进行识别训练。计算结果表明融合能够提高目标的识别的准确程度,并达到数据压缩的目的。本文利用数值计算的方法实现了远红外波段的超谱图像仿真过程,并在此基础上研究了喷焰目标处于自然背景时的识别问题。本文采用的方法比较真实地反映了实际的过程。从计算结果中可以看到仿真效果很好,说明本文采用的方法是可行的。更多还原

【Abstract】 High performance detection equipment needs target signal characteristic value which is much preciser and contains a larger amount of data in order to detect dim targets and camouflaged targets. Hyperspectral information has the advantage of high spatial and spectral resolution, so this spatial and spectral characteristics can be used to distinguish dim targets in complex background with higher credibility. With the rapid development of numerical simulation, the virtual design will be gradually used in the field of aerospace technology as a new comprehensive subject. It can reduce the research cost greatly, improve the accuracy and shorten the period of development.This dissertation focuses on the simulation of the background and target in far infrared channel (8~12μm ) and the detection of target in background. The main contents of this dissertation are as follows:1.The hyperspectral image is got by the simulation and calculation of background. Calculation process includes the generation of natural terrain and background, the solution of surficial temperature model and the atmosphere influence on radiation transfer.2.The infrared radiation properties of exhaust plume is modeled by heat flux method. Once they are added in the energy field of background, we could get the hyperspectral image (8~12μm ) in which the background contains target.3. The energy ratio of the target and background is calculated and then the band is selected to distinguish target. Dim targets in complex background are distinguished by neural network method and we get the best spectrum band combination and the LS_SVM method to distinguish them.4. The information correlation of the hyperspectral images is quantitated between the band of 80 and 106 by calculating the correlation information entropy. The hyperspectral images with stronger correlation between each other are processed by fusion method. LS_SVM method is used to distinguish dim targets in hyperspectral images before and after fusion. The calculation results show that it can be effective to distinguish target and realize the image compression by using the fusion method.This dissertation realizes the hyperspectral image simulation in the far infrared channel in the numerical method and focuses on target recognition problem when the target is in the natural background. The method developed in this dissertation can present practial process factually. The simulation effect is good and the method developed in the dissertation is feasible.更多还原