Abstract
In the past decade, great progress has been made in general object detection based on deep convolutional neural networks. However, object detection from Unmanned Aerial Vehicles (UAV) imagery received not so much concern. In this paper, a densely connected feature mining network is proposed for high accuracy detection. Specifically, multi-scale predictions are used to enhance the feature representation of the tiny vehicles. Furthermore, a streamlined one-stage detection network is used to achieve satisfactory trade-off between speed and accuracy. Finally, a improved distance metric function is integrated into the priors clustering process, which can lead to a better preliminary location before training. The proposed architecture is evaluated on the highly competitive UAV benchmark (UAVDT). The experimental results show that the proposed dense-darknet network has achieved a competitive performance of 42.03% mAP (mean Average Precision) and good generalization ability on the other UAV benchmarks.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call