Remote sensing image dense target detection based on rotating frame
Zhixin Wang, Shaojun Wan, Xiaoying Ma · Journal of Physics Conference Series · 2021
Abstract Detecting densely arranged and arbitrarily oriented targets on optical remote sensing images is a challenge, and there is much room for improvement in existing algorithms. In this paper, an end-to-end two-stage rotating target detection model based on rotating frames is proposed. This model adds FPN + PAN feature fusion structure after the backbone network to obtain enhanced features that fuse the feature information of each layer, and secondly, fine-grained rotation detection is accomplished by introducing Oriented Region Proposal Network and Oriented Region of Interest Pooling layer. By comparing the performance of this paper’s algorithm with the current mainstream rotation detection algorithm on DOTA remote sensing dataset, this paper’s algorithm can solve the difficult problem of detecting remote sensing targets with tight arrangement and arbitrary orientation to a certain extent.