Counting Buildings from Unmanned Aerial Vehicle Images Using a Deep Learning Based Approach

Abstract

Effective urban planning requires accurate and up-to-date spatial information. Remote sensing has contributed immensely to the efficiency of collecting this information. With remotely sensed high-spatial-resolution images, details such as buildings counted in an area can be extracted; however, traditional methods of extracting this information involve direct counting by humans, which is often demanding in terms of time. Computer vision techniques have shown promising results in handling image-related challenges in recent years. Therefore, this study aimed to adapt deep learning-based algorithms to simplify the counting of buildings from high-spatial-resolution aerial images in a fairly suburban environment. A deep learning algorithm based on convolutional neural networks, You Only Look Once (YOLO), was adapted to detect and count the buildings in the Unmanned Aerial Vehicle (UAV) sensed images. The model achieved high accuracy, with a recall rate of 0.89, an F1 score of 0.89, and an average precision of 91.12% on the validation data. When applied to new testing data, the algorithm successfully identified and counted the number of buildings with an overall accuracy of 71%. The approach presented in this research extracted building counts reliably, quickly, and accurately in a fairly suburban environment. Such information can be applied to tracking urban growth and physical planning.