An Antijamming and Lightweight Ship Detector Designed for Spaceborne Optical Images
Huanqian Yan, Bo Hu Li, Hong Zhang, Xingxing Wei · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 2022
Ship detection in spaceborne optical images is a challenging task because ships have various orientations and scales, especially complex backgrounds, i.e., ships are easily obscured by various jamming. Moreover, most accurate object detectors have enormous computation and parameter numbers, which are unsuitable for resource-bounded spaceborne platforms that contain restrictive memory access and computation. In this paper, in order to cope with the influence of complex backgrounds and jamming on detection, and improve the practicality of detection algorithms, a new satellite optical image dataset and a novel ship detector are proposed. We have collected a new dataset from the satellite, which contains images of different time periods, different illuminations, and different levels of jamming. The proposed dataset is different from the widely used public remote sensing datasets, it is more practical and challenging. The proposed ship detector can deal with various images well and is robust to various complex backgrounds. Specifically, a feature refining module is designed to extract features effectively, which can improve detection performance significantly. An anti-jamming module is proposed to highlight the features of objects in the whole feature map. In contrast to mainstream ship detectors, the proposed method is effective and lightweight. It can also predict objects with oriented bounding boxes. Extensive experiments demonstrate that the proposed detector is efficient and robust to various complex backgrounds, and the new dataset CBSD is more suitable for application scenarios and is quite challenging.