Object Detection Method Based on Aerial Image Instance Segmentation by Unmanned Aerial System in the Framework of Decision Making System

摘要

The article analyses the capabilities of unmanned aerial system application in the framework of decision making in the crisis situations that require the object detection on aerial images acquired by the unmanned aerial system. To increase the operational capability and credibility of the automotive vehicles detection at the aerial images acquired by the unmanned aerial systems for more efficient use of acquired information in the framework of decision making support model Mask R-CNN was selected. This model is more appropriate for solving the problem of multiclass classification and object detection of small-size objects on the image. To improve this model, the article recommends using small-size anchors taking into account height-to-width aspect ratio according to greater amount of classes that along with test time augmentation usage enables to augment the mAP.