基于自适应遗传算法的超平面分类及遥感应用

智能分类算法是遥感影像分类研究的热点,遗传算法作为一种智能全局优化技术在遥感影像分类中具有良好应用前景.针对现有多光谱遥感影像分类方法的不足,提出了基于自适应遗传算法的超平面分类方法(hyper plane-adaptive genetic algorithm,HP-AGA)并应用于遥感影像分类,该方法利用神经网络中的神经元激活函数Sigmoid函数,对遗传算法中交叉率、变异率进行非线性自适应性调整,不再需要反复训练遗传参数,同时利用快速全局寻优特点,确定分类超平面的各个位置参数,从而获取最佳分类超平面集进行分类.多光谱遥感影像分类方法的应用实验表明,基于自适应遗传算法的超平面遥感分类方法能更快、更稳定地收敛到全局最优解,具有更好的效率及鲁棒性,并能取得优于简单遗传超平面分类算法及传统分类方法的分类精度.

Hyper plane-adaptive GA classification and its application in remotely-sensed images

Intelligent classification has been a hot research topic in remote sensing image processing. And genetic algorithm (GA), as an intelligent global optimizing method, has a good application prospect in the remote sensing image classification. To improve the shortcomings of classification of multi-spectral remote sensing image, the paper proposed an improved model of hyperplane using adaptive genetic algorithm (HP-AGA), which adjusts the crossover probability and mutation probability adaptively and nonlinearly by using neurons in neural network to activate function-Sigmoid function, no longer needs to train the parameters of GA repeatedly. Based on the characteristics of global and fast optimizing, the position parameters of classification hyperplanes can be determined, thus an optimum classification hyperplanes set can be obtained to get a better classification. The experiment of classification shows that HP-AGA can search the global optimization faster and more stably. It is an effective and robust classification modeland, and has a better classification precision compared with the normal hyperplane classification, support vector machine classification method and the maximum likelihood classification.