Road Geometric Features Extraction based on Self-Organizing Map (SOM) Neural Network

Abstract

Geometry is the key parameter when extracting road from high-resolution remote sensing imagery. We propose a method for road geometry parameter s extraction from high spatial resolution remote sensing imagery automatically based on self-organizing map (SOM) neural network algorithm. SOM is a no-tutor clustering segmentation method and the algorithm is the foundation of later road automatic extraction. Our approach may adjust cluster number and cluster center of the image through analyzing the point density distribution of self-organizing feature map neural network competition layer, which is good for flexible processing on the image excessive segmentation problem and succeed in accurate segmentation object. Then we can extract geometric features of the terrain target. The results are demonstrate d that the algorithm proposed is both accurate and effective.