【作者】 崔和平；

【导师】 范志刚；

【作者基本信息】 哈尔滨工业大学 ， 光学工程， 2007， 硕士

【摘要】 红外焦平面阵列是目前红外成像系统的主要探测器件。而红外焦平面阵列本身存在的非均匀性响应在很大程度上限制了系统的探测性能。实时、有效的非均匀性校正方法是红外成像系统研制中的关键技术。本文对红外焦平面阵列非均匀性校正方法展开了深入探讨及研究,重点是针对基于场景的非均匀校正方法的研究。文中首先概述了红外焦平面阵列非均匀性的基本原理;分析了红外焦平面阵列成像系统非均匀性产生的机理;对成像系统响应的特点进行了阐述,重点分析了红外焦平面阵列的非线性响应曲线模型,以“S”型曲线响应模型代替线性响应模型,这更逼近探测器单元的真实响应特性。将典型的基于两点温度定标的红外焦平面阵列非均匀性校正算法引入到场景中,提出一种基于场景的最小二乘拟合非均匀性校正算法,该算法可以用任意运动场景直接进行最小二乘拟合校正,而不需要事先用标准参考源进行标定;对非均匀性校正的结果给出了几种定量的科学技术评价指标;从仿真结果中可以看出,该算法能够达到令人满意校正效果。对基于人工神经网络的红外焦平面阵列非均匀性校正算法进行了研究。在此基础上提出了可变学习速度反向传播算法和最优学习步长反向传播算法。可变学习速度反向传播算法可以视具体情况适当改变反向传播的学习速度,从而加快达到稳定。最优学习步长反向传播算法利用黄金分割搜索法选择最佳的学习步长,可以在最短的迭代次数内达到期望目标。由仿真结果可以看出,两种算法都能达到比较理想的校正效果。分析比较了上述三种改进的红外焦平面阵列非均匀性校正方法,从而验证了三种校正方法的有效性及适用性。更多还原

【Abstract】 Now Infrared Focal Plane Array (IRFPA) is the main detector in infrared imaging system. But the nonuniformity response existed in IRFPA restricts the detection performance of the system greatly. Real time and effective of the Nonuniformity Correction algorithm become the key technique of studying infrared imaging system. Discussion and study of nonuniformity response method is deeply developed in this paper in detail. And the keystone is the study of Nonuniformity Correction algorithm based the scene.In this paper firstly the elements of the nonuniformity in the IRFPA is summarize; secondly the generation principle of the nonuniformity in the IRFPA and the characteristics of the infrared imaging system response are analyzed. Analysis is focused on the nonlinear response curve model of infrared focal plane arrays. "S" curve response model replaces the linear response model and makes it more similar to the true response of detector module.The IRFPA Nonuniformity Correction algorithm based on the two-point temperature picketage is used in the scene. The least squares fitting Nonuniformity Correction algorithm based on scene is held out. The least squares fitting correction can directly be carried on using any moving scene by the algorithm. The picketage is not necessary with standard reference source in advance. Some measurable assessment indexes were given to the result of the Nonuniformity Correction. The algorithm’s effectiveness and superiority were verified by simulation.After the research on Nonuniformity Correction in IRFPA base on artificial neural networks, variable learning rate backpropagation and classic learning step-length backpropagation are put forward. Variable learning rate backpropagation could change learning rate of backpropagation to quicken the speed to get stable in peculiar situation. Classic learning step-length backpropagation selects the prime step-length by golden section search method, and could achieve the expect result in the least times of iteration. The effectiveness and superiority of two algorithms were verified by simulation.Simulation experiments were performed. The comparison is done among three above-mentioned algorithms, the effectiveness and adaptability of the three algorithms are validated by the simulation experiments.更多还原