基于YOLOv3-satellite的航天器模型目标检测方法

The method of target detection for spacecraft model based on YOLOv3-satellite

文中提出了一种基于YOLOv3-satellite的航天器模型目标检测方法。改进了原先的Dark-net53网络结构,以提高对航天器数据集检测时的精确度;使用航天器特有的特征信息分析选择出更合适的锚框;通过采用DIoU边框回归函数,改善了计算代价函数时对图片中心不敏感的问题。为验证检测方法的有效性和完备性,在航天器数据集上对YOLOv3-satellite进行了验证对比分析。实验结果表明,基于YOLOv3-satellite的航天器模型目标检测方法,检测正确率达到了93.3%,检测速度达到了30.1 fps,证明该方法具有一定的实践意义和实际价值。

In this paper,the method of target dectection for spacecraft model based on YOLOv3-satellite is proposed.The original Darknet53 network structure is improved to enhance the accuracy of spacecraft data set detection.The spacecraft specific feature information analysis is used to select a more suitable anchor frame.DIoU frame regression function is used to make up for the problem that the cost function is not sensitive to the image center.In order to verify the effectiveness and completeness of the detection method,the comparison and analysis of YOLOv3-satellite are carried out on the spacecraft data set.The experimental results show that the detection accuracy rate of spacecraft model target detection method based on YOLOv3-satellite reaches 93.3%,and the speed of test reaches 30.1 fps.The experimental results show that the method has certain practical significance and practical value.