Workflow for Landmine Detection in Aerial Imagery with YOLO-based Deep Learning

Landmine detection using aerial imagery is relatively new area of remote sensing and computer vision, but due to rapid development of both aerial vehicles and object classification and detection methods it is experiencing its evolution to the stage when it is possible to develop appropriate framework that can include all the stages in single software platform. The components of the system are scientific open-source libraries for object detection based on machine learning and deep learning, and universal geoinformation system extended by the set of configurable plugins for specific tasks. The reliable combination of software components is proposed and workflow is described from detector training to obtaining landmine coordinates. Real life example of successful implementation is provided based on state-of-the-art neural network libraries and geoinformation systems.