基于Faster R-CNN的航母舰面多尺度目标检测算法

Multi-scale object detection algorithm for aircraft carrier surface based on Faster R-CNN

针对航母舰面复杂的多尺度目标检测环境,且现有算法对牵引车、人员等小目标检测性能不佳的问题,提出一种改进快速区域卷积神经网络(faster region convolutional neural netuorks, Faster R-CNN)的舰面多尺度目标检测算法。基于多尺度特征层提取了不同尺度的区域建议网络,提高了算法对不同尺度目标尤其是对小目标的检测性能。基于K-means聚类算法生成了适合于舰面目标数据集的先验框尺寸,进一步提升了算法的性能。实验表明,所提算法有效地提升了不同尺度目标的检测性能,尤其是对小目标的检测效果,并对所提算法进行了消融实验,最后与不同算法的性能进行了对比,所提算法检测准确率取得了最优水平。

Aiming at the complex multi-scale object detection environment on the aircraft carrier surface, and the poor performance of existing algorithms for small objects such as tractors and personnel, an improved faster region convolutional neural netuorks(Faster R-CNN) aircraft carrier surface multi-scale object detection algorithm is proposed. Based on the multi-scale feature layer, the region proposal network of different scales is extracted, which improves the detection performance of the algorithm for objects of different scales, especially small objects. Based on the K-means clustering algorithm, a prior box suitable for the aircraft carrier surface object dataset is generated, which further improves the performance of the algorithm. Experiments show that the proposed algorithm effectively improves the detection performance for different scales of objects, especially the detection effect of small objects, and conducts ablation experiments on the improved algorithm. Finally, the performance of different algorithms is compared, which shows that the detection accuracy of the proposed algorithm achieveds the optimal level.