摘要
针对BP神经网络在遥感影像分类中存在易陷入局部极值、受初始权阈值影响大和网络训练时间长等问题,提出一种遗传算法(GA)结合粒子群算法(PSO)优化BP神经网络(GA-PSO-BP)的遥感影像分类方法。通过PSO对问题的解空间进行迭代寻优,将粒子群粒子个体转化为GA染色体,利用GA的复制、交叉和变异对种群所有染色体进行寻优。GA-PSO迭代寻优得到的初始权阈值直接赋给BP神经网络,解决其易陷入局部极值的问题,同时提升其训练速率。利用Landsat-8中分辨率和高分二号高分辨率遥感影像进行地物分类。结果表明,相对于最大似然法、支持向量机、传统BP、GA优化BP和PSO优化BP,GA-PSO-BP的分类精度得到有效提高,能与AlexNet卷积神经网络分类精度接近,且简单易操作。
Aiming at the problems of BP neural network in remote sensing image classification,such as easily falling into local extreme value,being greatly influenced by initial weight threshold and long training time of network,this paper proposes a remote sensing image classification method of genetic algorithm(GA)combined with particle swarm optimization(PSO)to optimize BP neural network(GA-PSO-BP).PSO is used to iteratively optimize the solution space of the problem,and the individual particle swarm is transformed into GA chromosome.All the chromosomes of the population are optimized by the replication,crossover and variation of GA.The initial weight threshold obtained by GA-PSO iterative optimization is directly assigned to BP neural network to solve the problem that is easy to fall into local extremum and to improve its training speed.Using Landsat-8 medium resolution and GF-2 high-resolution remote sensing image to classify objects,the results show that compared with maximum likelihood method,support vector machine,traditional BP,GA optimized BP and PSO optimized BP,the classification accuracy of GA-PSO-BP is effectively improved,which is close to the classification accuracy of AlexNet convolutional neural network,and it is simple and easy to operate.
作者
薛明
韦波
杨禄
李景文
姜建武
XUE Ming;WEI Bo;YANG Lu;LI Jingwen;JIANG Jianwu(Guangxi Key Laboratory of Spatial Information and Geomatics,Guilin University of Technology,Guilin,Guangxi 541004,China;College of Geomatics and Geoinformation,Guilin University of Technology,Guilin,Guangxi 541004,China)
出处
《遥感信息》
CSCD
北大核心
2020年第3期110-116,共7页
Remote Sensing Information
基金
国家自然科学基金项目(41461085)
“广西八桂学者”专项经费项目(2019-79)
广西空间信息与测绘重点实验室基金项目(16-380-25-04)
桂林理工大学博士基金项目(1996015)。
关键词
影像分类
BP神经网络
粒子群算法
遗传算法
算法优化
image classification
BP neural network
particle swarm optimization
genetic algorithm
algorithm optimization
