Small Aircraft Target Detection Using Cascade FP-CNN in Remote Sensing Images

J. Yang,J. Zhi,Y. Zhang,J. Wu,Y. Zhou,B. Zuo
DOI: https://doi.org/10.1049/icp.2021.0797
2020-01-01
Abstract:Aiming at the problem that the remote sensing images of small-scale aircraft targets are poorly detected in deep learning networks, the paper proposes a method of remote sensing image aircraft target detection based on Cascade FP-CNN network. The Cascade FP-CNN network uses feature pyramids to achieve feature fusion of feature maps on different scales, which enhances the target features learned by the network, thereby obtaining high-resolution, high-receptive-field feature maps. In addition, referring to the method for determining the crossover threshold of the Cascade network to set the crossover threshold of the Cascade FP-CNN network, a detection algorithm suitable for small-scale aircraft targets is finally constructed, thereby realizing the accurate recognition of small aircraft targets in remote sensing images. The average accuracy of the algorithm on the UCAS-AOD dataset reaches 96.01%, which is about 2% higher than the existed Faster R-CNN and Cascade R-CNN methods, and the detection capability has been significantly improved.
What problem does this paper attempt to address?