复杂场景SAR图像的船舰目标快速检测研究  

Research on Fast Detection of Complex SAR Ship Targets Based on Improved YOLOv3 and Attention Mechanism

作　　者：曹红 CAO Hong(School of Accounting And Finance,Zhejiang Business College,Hangzhou,China,310000)

机构地区：[1]浙江商业职业技术学院财会金融学院,杭州310000

出　　处：《福建电脑》2024年第7期53-57,共5页Journal of Fujian Computer

基　　金：浙江省教育厅科研项目“基于改进YOLOv3与注意力机制的复杂SAR图像船舰快速检测研究”(No.Y202249939)资助。

摘　　要：复杂场景SAR图像容易受地物和强散射干扰的影响。为提高船舰目标检测算法的效率和精确率,本文提出一种基于改进YOLOv3和注意力机制的检测网络方案。该检测网络主要由目标筛选网络P-FCN和目标精确检测网络S-SSD组成。PFCN是一个轻量型的全卷积网络,用于快速筛选船舰目标。S-SSD是一个改进的YOLOv3网络,通过多层次特征融合系统结合双通道注意力机制CBAM,结合P-FCN对船舰的目标定位实现了对船舰目标的精确检测。实验结果表明,本文算法对于复杂场景SAR图像船舰目标具有较好的检测性能。SAR images in complex scenes are easily affected by terrain and strong scattering interference.To improve the efficiency and accuracy of ship target detection algorithms,this paper proposes a detection network scheme based on improved YOLOv3 and attention mechanism.The detection network mainly consists of the target screening network P-FCN and the target precise detection network S-SSD.P-FCN is a lightweight fully convolutional network used for rapid screening of ship targets.S-SSD is an improved YOLOv3 network that achieves precise detection of ship targets through a multi-level feature fusion system combined with dual channel attention mechanism CBAM and P-FCN for ship target localization.The experimental results show that the algorithm proposed in this paper has good detection performance for ship targets in complex SAR images.

关 键 词：合成孔径雷达图像 船舰目标 快速检测算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]