【作者】 杨淳清；

【导师】 李兵；

【作者基本信息】 西华大学 ， 计算机应用技术， 2011， 硕士

【摘要】 红外焦平面阵列(IRFPA)非均匀性的存在,极大的限制了成像系统的性能,因此实现红外焦平面阵列非均匀性自适应校正是高级红外探测系统追求的重要目标。基于场景校正方法的优越性的存在,本文采用了基于场景的BP人工神经网络校正算法,由于经典BP校正算法存在参数难以选取的缺点,本文选用了一种改进的BP校正算法—归一化BP人工神经网络算法作为系统的校正算法。但随着校正系统算法复杂度和规模的增加,设计难度大大提高,需要对系统设计采用先进的设计方法,因此本文从数字系统设计的角度来实现此系统,采用“描述一综合”的的系统级建模方法,对基于BP神经网络的IRFPA非均匀性校正系统进行自顶向下的有层次的建模。首先对抽象出的系统对象类进行描述;基于对象类的聚合,建立系统的层次模型;然后采用Verilog语言来描述此模型,并在模拟环境进行仿真。仿真结果证明了模型的正确性和方法的可行性。本文在结合已有的研究成果和理论基础上就IRFPA的非均匀性校正开展了如下研究工作:1、归一化BP人工神经网络算法研究。2、由于归一化BP人工神经网络校正系统具有离散性,事件驱动的特点,结合高级Petri网理论采用面向对象计时双流Petri网,并对进行了定义和描述;从而建立了建模方法,为归一化BP人工神经网络校正奠定了理论基础。3、采用了基于电路模拟的嵌入式系统模型分析验证技术。着重研究了Petrie网模型的电路描述技术,采用硬件描述语言仿真Petri网模型运行、分析验证系统功能的方法和技术,为本系统的模型提供了一种简明实用的功能验证方法。4、结合归一化BP人工神经网络校正算法和面向对象计时双流Petri网,为IRFPA非均匀性校正系统建立了校正模型,并采用Verilog语言描述此模型以及在模拟环境进行仿真,实验证明模型的正确性和方法的可行性。更多还原

【Abstract】 The heterogeneity of infrared focal plane arrays (IRFPA)greatly limits the performance of the imaging system. Therefore, the adjustment and correction of IRFPA is the important object of the advanced infrared detection system. Because of the superiority of the scene-based correction, it takes the scene-based BP artificial neural network correction algorithm. As it is difficult to select the parameters through the classical BP correction algorithm, and in order to realize the SOC system, it chooses a more advanced BP correction—the normalized BP artificial neural network correction—as the systematic correction.As the complexity and scale of correction algorithm increases, the difficulty of design also greatly increases. We need to take an advanced method to design systems and so in this paper, it realizes the system through the digital system design. It takes the“description—combination”systematic modeling method to conduct the top-down hierarchical modeling on the BP neural network–based IRFPA heterogeneity correction system.At first, it describes the abstracted system object classes. On the basis of the aggregation of the object classes , it builds the systematic hierarchical model; And then, it uses Verilog to describe that model and to simulate it in a simulated environment. The results proves the correctness of the model and the feasibility of the method.Based on the existing research achievements and theories, it conducts the following research projects regarding the heterogeneity correction of IRFPA:1. Research of the normalized BP artificial neural network algorithms.2. Due to the discrete and event-driven characteristic of the normalized BP neural network correction system, it combines with the advanced Petri net theory, takes the object-oriented time counting Petri net, defines and describes , and then establishes the method of modeling. This establishes the theoretical foundation of the normalized BP artificial neural network correction.3. It takes the technology of verification and analysis of the embedded system which is based on circuit simulation, and particularly studies the circuit description technology of Petri net model. It also comes up with the method and technology of using hardware description language to simulate the operation of Petri net model and to analyze and verify the function of system. This provides a simple and practical functional verification method for model.4. Combining the normalized BP artificial neural network correction algorithm with OO-TDPN, it builds model for the IRFPA heterogeneity correction system. It also uses Verilog to describe this model and the simulation of it in a simulated environment. Experiments prove the correctness of the model and the feasibility of the method.更多还原