Research Article
Ship Detection in SAR Images Based on an Improved Detector with Rotational Boxes
@INPROCEEDINGS{10.1007/978-3-030-89814-4_61, author={Xiaowei Ding and Changbo Hou and Yongjian Xu}, title={Ship Detection in SAR Images Based on an Improved Detector with Rotational Boxes}, proceedings={Mobile Multimedia Communications. 14th EAI International Conference, Mobimedia 2021, Virtual Event, July 23-25, 2021, Proceedings}, proceedings_a={MOBIMEDIA}, year={2021}, month={11}, keywords={SAR image Ship target detection Convolutional neural network}, doi={10.1007/978-3-030-89814-4_61} }- Xiaowei Ding
Changbo Hou
Yongjian Xu
Year: 2021
Ship Detection in SAR Images Based on an Improved Detector with Rotational Boxes
MOBIMEDIA
Springer
DOI: 10.1007/978-3-030-89814-4_61
Abstract
In the SAR ship data under complex backgrounds, especially in the coastal area, the horizontal bounding box detection algorithm makes a large number of coastal noise interference targets feature extraction and bounding box regression. In addition, the horizontal bounding box cannot well reflect the characteristics of large aspect ratio of ships. Therefore, this paper proposes an improved YOLOv3 detection algorithm based on the rotational bounding box, which increases the encoding method of the angle parameter, and generates the prediction bounding box at a fixed angle interval. Different angle intervals will have different effects. Focus loss function is used to solve the problem of positive and negative sample balance and difficult sample feature learning. The experimental results show that the average precision of the R-YOLOv3 algorithm based on the rotational bounding box on the SAR ship data set is 87.3%, which is a 13.5% gain compared with the classic YOLOv3, which reflects the high precision of the ship targets.
