基于PSO-GA优化的TOP-HAT形态学滤波器及其应用  

Optimal Top-Hat Morphological Filters Based on PSO-GA and Application

作　　者：高德远[1] 王建刚[1] 

机构地区：[1]西北工业大学计算机科学与工程学院,陕西西安710072

出　　处：《航空计算技术》2013年第3期119-123,共5页Aeronautical Computing Technique

摘　　要：针对机载红外图像中运动弱小点目标检测的难题,提出了一种基于PSO-GA训练参数的形态学滤波器。以粒子群优化算法(Particle Swarm Optimization,PSO)为主线,按PSO算法中标准的速度和位置更新,遗传算法(Ge-netic Algorithm,GA)采用新的区间离散化编码和自适应的主次式交叉与变异算子,将遗传算法与粒子群优化算法的自动更新特征结合在一起,通过优化搜索全局空间获得形态学滤波器的最优参数,进而确保优化的形态学滤波器具有良好的滤波性及时效性。通过对低信噪比红外图像(SNR约为2)的测试,检测概率可以达到98%以上,与利用神经网络(Neural Network,NN)训练结构元素后的Top-Hat形态学滤波器相比提高了2%～3%。与GA算法相对,训练算法效能提高20%,提高了搜索最佳值的能力。Towards the detection of small moving target in sequential infrared images,a novel method for optimal morphological filtering parameters based PSO-GA is presented in this paper.Particle swarm optimization（PSO） serves as main line,according to the PSO algorithm＇s speed and location update.GA adopts interval discretization code and new crossover and mutation operators called the primary and secondary mood.Selection,crossover,mutation combined the PSO algorithm automatic update features to achieve optimal filtering parameters in a global searching.The experimental results show that the identified probability to the image of SNR 2 can be improved more 2-3% than that training with NN.Compared with GA,PSO-GA improves 20% in training effectiveness and the ability to search the best value.

关 键 词：粒子群优化算法 遗传算法 神经网络 Top-Hat形态学滤波器 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]