改进的波段选择混合核函数遥感图像分类算法

针对遥感图像多波段不易成像、其图像信息冗余不适合图像分类以及传统LMBP算法迭代次数多且分类不够精确的问题,改进了OIF指数和可分性距离公式,分组并选出遥感图像最佳波段组合,并运用改进的LMBP混合核函数算法进行分类。仿真实验表明,改进算法对各波段信息分析更加全面客观,波段选择更加优化;与传统算法相比,网络训练迭代次数有明显减少,分类精度及Kappa系数分别提高了5%和6.625%,遥感图像分类更有效。

Algorithm of remote sensing image classification improved by bands selection and hybrid kernel functions

As the multi-band of remote sensing image is not easy to imaging, its redundancy image information is not suitable for image classification, what's more, the traditional LMBP algorithm has large iteration number and classification imprecise problems. This paper improved the formula of the OIF index number and separability distance, separated to chose the best band combination, and then used the LMBP algorithm refinement of hybrid kernel function to classify. The simulation results show that the improved method can analyze information of the bands more comprehensive and objective, comparing with the traditional algorithm, the network training iterations are significantly reduced, the classification accuracy and Kappa coefficient can be increased by 5% and 6. 625%, the classification of remote sensing image more effectively.