Using Machine Learning Methods to Identify and Classify the Regions and Projections of Online Maps. GeoComputation 2019

In this paper, machine learning methods are developed to identify and classify maps from online sources. Given any image, our methods can be used to answer three questions: is it a map, which geographic region is this map about, and what projection is used in the map? A total of 1000 images are collected using Google Images to provide data for non-maps and maps that cover different geographic regions with different projections. Three machine learning methods are tested: multilayer perceptrons (MLP), support vector machine (SVM), and convolutional neural network (CNN). Each method is evaluated by how accurate it can be used to correctly identify and classify a given image. For map identification, all three methods can be used to obtain results with high accuracy, while for map classification based on region and projection, only SVM and CNN perform well.