Improved Rooftop Detection in Aerial Images with Machine Learning

摘要: In this paper, we examine the use of machine learning to improve a rooftop detection process, one step in vision system that recognizes buildings overhead imagery. We review problem analyzing aerial images and describe an existing detects such images. briefly four algorithms selected detection. The data sets were highly skewed cost mistakes differed between classes, so used ROC analysis evaluate methods under varying error costs. report three experiments designed illuminate facets applying image task. One investigated with all available determine best performing method. Another focused on within-image learning, which derived training testing from same image. A final experiment addressed between-image came different Results suggest useful generalization occurred when differing location aspect. They demonstrate most conditions, naive Bayes exceeded accuracy other handcrafted classifier, solution currently building system.