基于Ghost-YOLOv5s的SAR图像舰船目标检测  

Ship Target Detection in SAR Images Based on Ghost-YOLOv5s

作　　者：张慧敏[1] 黄炜嘉[1] 李锋[1] ZHANG Huimin;HUANG Weijia;LI Feng(Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212100,China)

机构地区：[1]江苏科技大学海洋学院,江苏镇江212100

出　　处：《火力与指挥控制》2024年第4期24-30,共7页Fire Control & Command Control

基　　金：国家自然科学基金资助项目(61701416)。

摘　　要：基于星载合成孔径雷达(synthetic aperture radar,SAR)图像的舰船目标检测中,为了平衡模型大小与检测精度,提出了一种基于Ghost卷积的SAR图像舰船目标检测方法Ghost-YOLOv5s。在YOLOv5s的颈部引入Ghost卷积,以减少模型参数和压缩模型体积;将高效的通道注意力机制(efficient channel attention,ECA)融入到颈部的C3模块里,以突出重要特征,从而保持较高的检测性能;使用SIoU损失函数替换原来的CIoU损失函数,以减少预测框和真实框之间的偏差,提高检测算法精度。实验结果表明,在SSDD遥感数据集上,改进模型与YOLOv5s相比,模型参数量减少了6.28%,模型体积减小了6.21%,检测精度达到了98.21%,实现了模型大小与检测精度的平衡。In the ship target detection based on spaceborne synthetic aperture radar(SAR)images,in order to balance the model size and detection accuracy,a ship target detection method Ghost-YOLOv5s based on SAR images of Ghost convolution is proposed.First,Ghost convolution is introduced into the neck of YOLOv5s to reduce the model parameters and to compress model volume.Then,an efficient channel attention mechanism(ECA)is integrated into the C3 module of the neck to highlight the important features so as to maintain high detection performance.Finally,the SIoU loss function is used to replace the original CIoU loss function,so as to reduce the deviation between the predicted box and the real box and to improve the accuracy of the detection algorithm.The experimental results show that compared with YOLOv5s on SSDD remote sensing data set,the improved model reduces the number of model parameters by 6.28%,the model volume by 6.21%,and the detection accuracy reaches 98.21%,achieving the balance between model size and detection accuracy.

关 键 词：合成孔径雷达 深度学习 Ghost卷积 注意力机制 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]