Abstract:The target signal of millimeter wave (MMW) radiometer was decomposed to a series of projections in orthogonal subspaces by wavelet transform. The features of MMW radiometer in each orthogonal subspace were analyzed and characterized by neural network. In addition, the results of simulation show that the target recognition method developed by this paper is effective when the SNR is low.

Abstract:A moving object detection method in video sequence was proposed, which is considered a fixed camera model. Firstly, a new method was presented for detection of changes of motion. After the change detection, the recursive higher order mothod was used to extract the change from the Gaussian noise. Compared with the classical higher order statistics method, the recursive higher order statistics method uses the former n images information, so it can obtain better estimates of HOS to reduce the effect of additive noise and detect the small object and perform well on segmetation.

Abstract:提出采用基于迭代算法的分级自适应可变形模型，用于声带振动图像的自动分割．该模型由基本可变形模型和高阶可变形模型构成二级模型，基本可变形模型由２ 条描述声带内壁的样条曲线组成，依据声带振动目标形状的先验知识定义及初始化．高阶模型通过提高每条曲线的阶次，增加控制点数来捕获目标的精细边缘，以得出精确的声带形状分割结果．本文模型充分利用了有关声带形状的先验知识及全局特征，增加了图像分割的正确性及稳定性．

Abstract:With the description of average Holder constant of fractional Brownian motion, and analyzing the different depth of 3D surface between the target and complex background, a novel method for artificial target recognition based on wavelet transformation and Holder constant was proposed. Examples of application of the algorithm to artificial target recognition were given. The anti interference capability was improved by using wavelet transformation. More accurate localization can be attained by using two images in the same region by different scale. This method is suitable for identifying specific targets in a complex background.

Abstract:A new model based on the structure of V1 area in the biological visual system was presented for image searching. The model only uses the known V1 elements, such as orientation selective cells and horizontal intra cortical connections. The knowledge, which will control the dynamic of the neurons, is represented by the chain code of the object's contour. The chain code is presented to the neural network in timing pulse form. By cooperating with the dynamic weights, only the neurons of the contour according with the object contour will reach the active state and stay there. Finally the contour that conforms to the knowledge contour emerges in the visual area, and the object is located.

Abstract:针对前视红外（ＦＬＩＲ）图像的分割，提出采用去中值波滤器进行预处理抑制背景、增强目标，而进利用基于模型的ＦＬＩＲ图像分割（ＭＢＳ）算法完成图像分割，从相容性向量及初始概率计算两方面对ＭＢＳ算法进行了改进，对红外目标图结果证实该方法与ＭＢＳ算法相比，在低对比度、高噪声情况下能得到更为精确的分割结果，同时能极大地降低了背景干扰。

Abstract:Based on the research work of Freeman et al. about the visual physiological character, a mathematical model of complex binocular cell was deduced and thereby an algorithm of disparity computing was realized by using this model. Experiments on random dots stereopair and pentagon stereopair show that this method can be used to estimate disparities rapidly and efficiently.

Abstract:The influence of the spin on the ground state energy of the strong coupling 2 D magnetopolaron was studied by using the linear combination operator method. Numerical calculation, with KI crystal as an example, illustrates that the ground state energy of the strong coupling 2 D magnetopolaron for the electron spin of different direction increases or decreases with increasing magnetic field. The ratio of electron spin energy to the ground state energy of the magnetopolaron increases with increasing magnetic field.

Abstract:采用溶胶－凝胶（Ｓｏｌ－Ｇｅｌ）法在硅衬底上制备了ＰｂＴｉＯ３ 多晶薄膜，利用红外反射光谱测量了薄膜在室温下的声子谱，并首次观察到８ 个红外活性的声子模，分别位于８１、１５５、２０７、３００、３４４、４６０、５００ 和６２１ｃｍ － １，其中位于３００ 和４６０ｃｍ － １的两个声子模以前未曾见报道过．与单晶相比，除了位于１５５ｃｍ － １的声子模向高波数移动之外，其它的声子模均向低波数移动，并讨论了影响声子模移动的因素．

Abstract:Deep level transient spectroscopy (DLTS) technique was successfully applied to characterize the electric properties of p type self organized InAs quantum dots The ground state energy and capture barrier energy of hole of quantum dots were measured for the first time The energy of ground state of 2 5ML InAs quantum dots with respect to the valence band of bulk GaAs was obtained being about 0 09eV, and there was a barrier associated to the change of charge state of quantum dots The capture barrier energy of such dots for hole was about 0 26eV The work is very meaningful for further understanding the intrinsic properties of quantum dots

Abstract:为提高以ＧａＡｓ 为基底的中红外探测器的综合性能，从理论上验证了在ＧａＡｓ 基底上镀ＴｉＯ２ ｓｏｌ－ｇｅｌ纳米颗粒膜能够较大幅度地提高ＧａＡｓ 红外透过率，并在实验中获得了适合于镀膜的稳定ＴｉＯ２ 纳米颗粒溶胶，有效控制镀膜工艺参数，在ＧａＡｓ 基底上成功地镀上了合适的ＴｉＯ２ｓｏｌ－ｇｅｌ膜，实现了增透效果，在２．５～６．０μｍ 波段内，透射比的最大值由未镀膜的５６％ 提高到镀膜后的９４％ ．

Abstract:研究了模板与图像匹配中元定位的方法．分析了在亚像元定位中的重采样和曲面拟合方法，对重采样方法进行如下改进：仅对模板进行ｎ 倍重采样，产生（２ｎ－ １）个子模板，计算每个子模板与图像的归一化相关系数（ＮＣ值），则ＮＣ最大值的子模板所对应的偏移即为要求取的亚像元偏移．并提出了一种导数求取曲面拟合系数的方法；融合重采样和曲面拟合提出了一种新的亚像元定位方法，实验结果表明了此算法的有效性．

Abstract:A novelfuzzy neuralnetw ork (FNN) m odelform ulti-fontcharacterrecognition w as presented, w hich can efficiently process the fuzzy pattern classification problem . This FNN m odelis built by fuzzifying the input layer, output layer and the training algorithm ofa conventionalm ultilayer perceptron (MLP). The sim ulation w ith a lot of m ulti-font character sam ples show s thatthe FNN presented here can geta high recognition rate, and has low sensitivity for different character fonts in com parison w ith classical MLP.Also, this FNN is proved to have a good robustness.