Rotating Target Detection Based on Lightweight Network
收藏中国科学院中国科学技术大学科学数据中心2026-01-10 收录
下载链接：
https://sdc.ustc.edu.cn/dataDetails/ybUaOJYBQwfvTVc55OSG
下载链接
链接失效反馈官方服务：
资源简介：
Current rotating object detection task achieves good results
based on large models. In order to reduce the size of model, we propose a
lightweight network SFC (ShuffleNet combines FPN with CSL) for rotating target detection. SFC first introduces circular smooth label (CSL) to
detect target rotations, which transforms the traditional angle regression
problem into classification problem. Then, the lightweight ShuffleNetV2
is utilized as the backbone to reduce the number of parameters. Shuf-
fleNetV2 is used for feature extraction, and CSL is introduced to address
the periodic problem of angles. Comparative experiments were carried
out on DOTA 1.5 dataset. The experimental results show that the proposed method reduces the parameter by nearly 90% with a slight loss of
accuracy, and increases the inferencing speed by 40% at the same time
提供机构：
中国科学院软件研究所
创建时间：
2023-05-31
