基于SOM的高分辨率遥感影像道路网自动提取

本文研究基于SOM(Self-Organizing Feature Map)神经网络学习模型的高分辨率遥感影像道路网自动提取算法。首先利用数学形态学提取遥感图像道路的初始道路区域信息,自动对原始图像进行分区并确定神经元初始权值,用SOM网络学习模型对神经元进行训练学习,经迭代获取道路网中心点位置,最后运用"中心点四邻域跟踪判别法"跟踪连接形成道路中心线。实验表明,该方法在高分辨率遥感影像道路网的提取上有较好的效果,特别在主干道路网的提取上效果更佳,对噪声干扰具有良好的鲁棒性。

Automatic extraction of main roads from high-resolution remote sensing images based on SOM

An efficient method based on SOM neural network was proposed for extracting road networks from the high resolution remote sensing image.Firstly,the road region segmented used methods of mathematic morphology.Then the original image was divided into some of grids,and the original weights of neurons were determined.The road center nodes in the grids were obtained with the SOM algorithm which was inspired from a specialized variation of SOM neural network.Finally,using "Four-Direction Tracking" approach,the road centerlines were tracked automatically.The experiment results showed that this method was capable of rapidly and accurately extracting main road networks in addition to its good robustness to noise.