Optimizing the Snake Model Using Honey-Bee Mating Algorithm for Road Extraction from Very High-Resolution Satellite Images

Abstract

Many geospatial applications rely on the extraction of spatial features, including road networks, from very high-resolution (VHR) satellite images. Researchers have developed many algorithms to achieve this goal, the majority of which are based on image fusion, fuzzy logic, and active contour models. The snake model is among the most widely used methods for road extraction by active contours. In most studies, an initial curve close to available roads is manually defined or based on prior knowledge. These methods also require manual adjustment of the snake model parameters, which is time-consuming. In order to address these limitations, this study proposes an algorithm for extracting roads from VHR satellite images in a semi-urban area that optimizes snake models by Honey-Bee Mating Optimization (HBMO). Based on a support vector machine and some image processing analysis, the presented method can extract an accurate initial curve, as well. According to the results of the experiments, the proposed approach not only eliminates the shortcomings of the snake model but also increases the accuracy of road extraction by 10% in all three study areas compared to the traditional snake method.