基于Matlab的GA-BPNN遥感图像分类

Application of BP Neural Network Classification with Optimization of Genetic Algorithm for Remote Sensing Imagery Based on Matlab

在高原山地等地类复杂地区,传统遥感分类方法和标准BP神经网络分类方法存在一定的局限性,提出了基于Matlab的遗传算法优化的BP人工神经网络遥感图像分类方法。以Matlab神经网络和遗传算法工具箱为平台,在对数据源进行主成分分析特征选择的基础上,用量化共轭梯度法改进标准BP算法,采用GA优化BP网络的隐层神经元数目和初始权重,并以香格里拉县ETM+遥感图像为例,在DEM地形数据辅助下,训练网络使其收敛,仿真输出。结果表明,该方法分类总精度为84.52%,Kappa系数为0.8317,比最大似然法分类精度提高了9.08个百分点,验证了GA优化的BP网络遥感图像分类的可行性和有效性。

The accuracy of classification was very low for remote sensing imagery of mountain areas in the highlands with traditional method,and it was not satisfactory with the method of standard Back-Propagation neural networks in practical applications too.The new method was presented,with neural networks and genetic algorithm toolbox of the Matlab for the platform,using conjugate gradient method to improve the standard BP algorithm,using GA to optimize the BP network to identify the number of hidden layer neurons and the initial weights.For example,the ETM ＋ remote sensing image of Shangri-La County was classified in this method.The results show that the Kappa coefficient is 0.8317,the overall classification accuracy is 84.52%,and it is improved by 9.08 percentage points,compared with the maximum likelihood classification method.And it shows that it is feasible and effective to classify the remote sensing imagery by the BP network based on the optimization of genetic algorithm.