摘要
在SAR图像的船舶检测中,由于SAR图像船舶小、目标多,现有检测方法存在对船舶检测精度低、召回率低等问题。针对上述问题,提出了针对SAR图像的船舶目标检测模型Vessel-YOLO。首先,使用YOLOv8n为基准网络,提出了CASPP上下文空间金字塔池化结构,提高了模型对不同尺度的特征提取能力。其次,通过将所提模型的损失函数改进为Wise-IoU基于动态非单调聚焦机制的边界框损失,提高了模型对不同质量图像的适应程度。通过在标准数据集SAR-Ship-Dataset和SSDD上进行的大量实验,验证了模型的健壮性和可靠性。实验结果表明,相较于YOLOv8n,Vessel-YOLO在两个数据集上mAP_(0.5∶0.95)分别提高了1.8个百分点和2.2个百分点,所提模型以更高精度胜于现有的SAR图像船舶检测模型。
In ship detection of SAR images,the existing detection methods have the problems of low accuracy and low recall rate for ship detection because the ship targets are small and numerous in SAR images.To address the above problems,this paper proposes a ship target detection algorithm,Vessel-YOLO model for SAR images.Firstly,YOLOv8n is taken as the benchmark network,and a CASPP context space pyramid pooling structure is proposed to improve the capability of the model to extract features of different scales.Secondly,by improving the loss function of this model to Wise-IoU bounding box loss based on dynamic non-monotonic focusing mechanism,the model's adaptability to different quality images is improved.The robustness and reliability of the model are verified by extensive experiments on the standard datasets of SAR-Ship-Dataset and SSDD.The experimental results show that,compared with YOLOv8n,Vessel-YOLO improves mAP_(0.5∶0.95)by 1.8 and 2.2 percentage points on the two datasets respectively,and the proposed model with higher accuracy outperforms existing SAR image ship detection models.
作者
宁峰
赵良军
郑莉萍
梁刚
席裕斌
何中良
NING Feng;ZHAO Liangjun;ZHENG Liping;LIANG Gang;XI Yubin;HE Zhongliang(Sichuan University of Science&Engineering,School of Automation and Information Engineering,Yibin 643000,China;Sichuan University of Science&Engineering,School of Computer Science and Engineering,Yibin 643000,China)
出处
《电光与控制》
CSCD
北大核心
2024年第5期60-65,共6页
Electronics Optics & Control
基金
四川省科技计划项目(2023YFS0371)
四川省智慧旅游研究基地项目(ZHZJ22-03)。