基于YOLOv5的遥感图像舰船的检测方法

Detection method of remote sensing image ship based on YOLOv5

利用遥感图像对海面上舰船进行监控已成为目前研究的热点,为了解决传统舰船检测需人工选择特征、耗时耗力、原始YOLO算法对密集分布小目标检测精度不高的缺陷,提出了一种基于YOLOv5的遥感图像舰船检测方法,使用Kaggle平台提供的遥感数据集,在Pytorch框架上训练,损失函数设计为CIOULOSS,目标框的选择使用DIOUNMS算法,使被遮挡、重叠的目标检测效果增强。经实验对比,此目标检测模型对被遮挡、排列密集的舰船的检测精度优于其他模型,其平均检测精度由原始的88.75%提升到91.27%。

The use of remote sensing images to monitor ships on the sea has become a hot spot in current research.In order to solve the defects of traditional ship detection that requires manual feature selection,time-consuming and laborconsuming,and the original YOLO algorithm has low detection accuracy for densely distributed small targets.This paper proposes a remote sensing image ship detection method based on YOLOv5,using the remote sensing data set provided by the Kaggle platform,training on the Pytorch framework,the loss function is designed as CIOU_LOSS,and the selection of the target frame uses the DIOU_NMS algorithm to make the occluded and overlapped the target detection effect is enhanced.After experimental comparison,the detection accuracy of this target detection model for occluded and densely arranged ships is better than other models,and its average detection accuracy is increased from the original 88.75%to 91.27%.