摘要
舰船目标检测技术因其在海洋监测中具有广泛的应用前景,而成为计算机视觉领域的研究热点之一。但是,实际检测场景存在多尺度舰船目标,并且遥感和合成孔径雷达(Synthetic Aperture Radar,SAR)图像因宽高比较大,以及复杂的背景条件,容易产生漏检、误检。针对这些问题,文中提出了一种改进YOLOV5的遥感和SAR图像舰船目标检测算法。首先,引入3×3的快速空间金字塔池化模块(3×3 Fast Spatial Pyramid Pooling Module,SPPF_t)代替基线网络传统的空间金字塔池化模块(Spatial Pyramid Pooling,SPP);其次,在SPPF_t之后引入卷积块注意力模块(Convolutional Block Attention Module,CBAM);然后,针对小型舰船目标容易造成漏检,提出了膨胀率为2的小尺度特征增强模块(Dilated Small-Scale Feature Enhancement Module with Rate 2,SFEM_t),将增强后的特征图送入检测头进行检测;最后,将GIOU损失函数更换为CIOU损失函数。实验结果表明,改进后的YOLOV5算法在遥感和SAR图像舰船目标检测中均取得了更好的检测效果,具备更高的检测精度。
Ship target detection technology has become one of the research hotspots in the field of computer vision due to its wide application prospects in maritime surveillance.However,practical detection scenarios involve multi-scale ship targets,and remote sensing and SAR images are prone to missed detections and false alarms due to their large aspect ratios and complex background conditions.In this paper,we propose an improved ship target detection algorithm for remote sensing and SAR images based on YOLOV5.Firstly,we introduce a 3×3 fast spatial pyramid pooling module(SPPF_t)to replace the traditional spatial pyramid pooling module(SPP)in the baseline network.Secondly,we incorporate the CBAM attention mechanism after SPPF_t.Then,to address the issue of missed detections caused by small-sized ship targets,we propose a small-scale feature enhancement module(SFEM_t)with a dilation rate of 2,which feeds the enhanced feature maps into the detection head for detection.Finally,we replace the GIOU loss function with the CIOU loss function.Experimental results demonstrate that the proposed improved YOLOV5 algorithm achieves better detection performance and higher accuracy in ship target detection for remote sensing and SAR images.
作者
杨金鹏
黄柏圣
陈小娇
孙喆
YANG Jin-peng;HUANG Bai-sheng;CHEN Xiao-jiao;SUN Zhe(School of Electronics and Information Engineer,Nanjing University of information Science and Technology,Nanjing 210044,China)
出处
《中国电子科学研究院学报》
北大核心
2023年第9期821-829,共9页
Journal of China Academy of Electronics and Information Technology
基金
南京信息工程大学高层次人才基金资助项目(21r036)。
