红外成像非均匀性校正算法及其FPGA实现研究

发布时间：2018-05-09 19:35

本文选题：非均匀性校正 + FPGA　；参考：《西安电子科技大学》2015年硕士论文

【摘要】：红外焦平面阵列作为凝视型红外成像系统的核心组件,是最具发展潜力的一种红外探测器,但是由于传感器材料和制造工艺、信号处理系统暗电流噪声以及工作环境等因素的影响,探测器各个单元的响应之间存在非均匀性,从而在所输出的图像序列中产生固定图案的噪声。红外成像的非均匀性不仅严重降低了系统的成像质量,而且会对图像的后续处理产生干扰,因此改善红外成像的非均匀性显得十分有必要,目前采用的最有效的方法是基于数字图像处理的非均匀性校正技术。本文的目的在于研究一种基于场景的红外成像非均匀性自适应校正算法,并对其进行硬件实现,从而设计一种红外成像非均匀性实时校正系统。本文研究了基于神经网络的非均匀性校正算法,并分析了算法中存在的缺陷及其产生的原因。在此基础上,对神经网络法做出如下改进:一方面采用引导滤波计算预测图像,在对图像起到平滑作用的同时保留图像中的边缘信息,从而在提高校正效果的同时有效抑制“鬼影”现象的产生;另一方面使用投影法估计场景中的运动情况,并只在场景运动比较充足的情况下对校正参数进行更新,从而避免了场景静止时因参数重复迭代更新造成的图像模糊。针对上述改进算法,本文使用两组具有代表性的红外图像序列进行仿真并观察校正效果,同时结合图像粗糙度、均方误差和信噪比等评价指标对校正结果进行定量分析,最终说明所提出的改进算法相对于传统神经网络法具有明显的优势。另外,本文以FPGA为核心处理器构建硬件平台,充分利用FPGA可编程性强和并行计算的特点,对改进的神经网络非均匀性校正算法进行硬件实现,设计了一种基于FPGA的红外成像非均匀性自适应校正系统,能够对256×256像素的图像序列以每秒25帧的速度进行实时校正。文中根据自顶向下的层次化设计思想详细描述了系统的实现方式,包括FPGA的顶层设计和模块划分,以及各个功能模块之间的数据流向,并详细描述了各个子模块的设计细节。最后,通过仿真和测试验证了系统的功能,并从FPGA资源占用和系统运算速度两方面分析了系统的性能。最终证明,本文给出的硬件系统能够对红外图像中的非均匀性进行实时的自适应校正,并有效防止“鬼影”和模糊现象的产生。
[Abstract]:As the core component of staring infrared imaging system, infrared focal plane array (IRFPA) is one of the most promising infrared detectors. However, due to sensor materials and manufacturing technology, Due to the influence of dark current noise and working environment in the signal processing system, the response of each unit of the detector is non-uniform, which produces the fixed pattern noise in the output image sequence. The nonuniformity of infrared imaging not only seriously reduces the imaging quality of the system, but also interferes with the subsequent processing of the image, so it is necessary to improve the non-uniformity of infrared imaging. The most effective method used at present is non-uniformity correction based on digital image processing. The purpose of this paper is to study a scene based adaptive correction algorithm for nonuniformity of infrared imaging and implement it in hardware, and then design a real-time correction system for nonuniformity of infrared imaging. In this paper, the nonuniformity correction algorithm based on neural network is studied, and the defects in the algorithm and its causes are analyzed. On the basis of this, the neural network method is improved as follows: on the one hand, the guided filter is used to calculate the predicted image, which can smooth the image while preserving the edge information of the image. On the other hand, the projection method is used to estimate the motion of the scene and update the correction parameters only when the scene motion is sufficient. Thus the image blur caused by repeated iterative updating of parameters when the scene is still is avoided. In view of the above improved algorithm, two groups of representative infrared image sequences are used to simulate and observe the correction effect. At the same time, the correction results are quantitatively analyzed by combining the evaluation indexes such as image roughness, mean square error and signal-to-noise ratio. Finally, it is shown that the proposed improved algorithm has obvious advantages over the traditional neural network method. In addition, this paper takes FPGA as the core processor to build the hardware platform, makes full use of the characteristics of FPGA programmable and parallel computing, implements the improved neural network nonuniformity correction algorithm in hardware. An adaptive infrared imaging nonuniformity correction system based on FPGA is designed, which can correct 256x25pixels image sequences at a speed of 25 frames per second. According to the top-down hierarchical design idea, this paper describes the implementation of the system in detail, including the top-level design and module partition of FPGA, as well as the data flow direction between each functional module, and describes the design details of each sub-module in detail. Finally, the function of the system is verified by simulation and test, and the performance of the system is analyzed from the aspects of FPGA resource occupation and system operation speed. Finally, it is proved that the hardware system presented in this paper can correct the nonuniformity of infrared image in real time and effectively prevent "ghost" and blur.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN215;TP391.41;TN791

【参考文献】