基于边缘检测的神经网络非均匀性校正及硬件实现  

Adaptive Neural Network Non-uniformity Correction Based on Edge Detection and Running on Hardware

作　　者：刘秀[1] 刘咏[2] 金伟其[3] 林招荣[1] 宋立国[1] 

机构地区：[1]北京空间机电研究所,北京100076 [2]中国白城兵器试验中心,吉林白城137001 [3]北京理工大学光电学院,北京100081

出　　处：《光电工程》2014年第2期63-68,共6页Opto-Electronic Engineering

基　　金：国家自然科学基金资助项目(60877060)

摘　　要：红外焦平面探测器的非均匀性校正技术仍然是当前红外热成像系统重点研究的关键技术之一。相对定标类算法,基于场景非均匀性校正根据场景进行非均匀参数更新,不需要使用挡板遮挡视场。本文介绍传统神经网络非均匀性校正算法及加快收敛速度的改进措施,引入边缘检测方法来克服传统神经网络算法的鬼影问题。文中阐述的方法已在以TMS320DM643为处理核心DSP硬件处理平台上实现,取得了较好的校正效果。The Fixed Pattern Noise （FPN） of the infrared focal plane array severely limits the system performance, and the non-uniformity correction algorithm is a key technique of thermal imaging system. The scene-based non-uniformity correction algorithm does not require a shutter to block the field of view, but utilizes the scene information of image sequences to calculate the infrared focal plane array non-uniformity parameters. This paper introduces an improved neural network non-uniformity correction algorithm, which speeds up the convergence rate of the conventional neural network algorithm. The improved algorithm employs the edge detection method to overcome the ghosting artifacts generated by the conventional algorithm. The algorithm has run on a small low power consumption DSP hardware platform with TMS320DM643 as the kernel processor and can do the correction in a simple way with satisfactory results, so the algorithm introduced in this paper is proved to be reasonable and effective.

关 键 词：神经网络 边缘检测 非均匀性校正 

分 类 号：TN911.73[电子电信—通信与信息系统] TN216[电子电信—信息与通信工程]