Object Detection Method Based on Aerial Image Instance Segmentation in Poor Optical Conditions for Integration of Data into an Infocommunication System

The article analyses the possibilities to use the unmanned aerial complexes in the system of decision making process in crisis situations that require the object detection at aerial images received by the unmanned aerial complexes under the conditions of atmospheric fog. The Pansharpening method was used for image correction to inject spatial details from panchromatic image to multidimensional image. In order to increase the operational efficiency and accuracy of automotive vehicles detection at aerial images received by the unmanned aerial complexes for more efficient use of received information in the system of decision making support it was selected the Cascade Mask R-CNN model. This model is more suitable for task solution of multiclass classification and small-sized object detection at the image. To improve this model it is suggested using the small-sized anchors making into account the aspect ratio to more classes, function focal loss for model training that along with test time augmentation use enabled to increase mean Average Precision (mAP).