基于改进Faster R-CNN的SAR舰船图像检测

SAR Ship Image Detection Based on Improved Faster R-CNN

针对传统的SAR舰船检测算法适应能力和准确率低的问题,提出一种基于改进Faster R-CNN的SAR舰船图像检测算法。改进后的算法以Faster R-CNN为检测框架,利用改进的k-means算法设计更适合舰船目标形状特点的先验锚点框;优化NMS算法以剔除重叠区域的舰船候选框,改善了舰船距离较近导致的漏检问题;同时引入Mask R-CNN算法中的RoI Align单元来消除特征图与原始图像的映射偏差。试验结果表明,改进后的算法相比Faster R-CNN算法平均检测精度提升5.1%,达到86.64%,可以达到船舶数据量庞大情形下的检测要求。

Aiming at the problem of low adaptability and low accuracy of conventional SAR ship detection algorithm,a SAR ship image detection algorithm based on improved Faster R-CNN is proposed.Faster R-CNN is taken as the detection frame in the improved algorithm,and the improved k-means algorithm is used to design prior bounding box that is more suitable for target shape characteristics of ship;NMS algorithm is optimized to eliminate ship candidate bound in overlap areas,which improves the ship undetected problems caused by short beam distance between ships;meanwhile,RoI Align map unit in Mask R-CNN algorithm is introduced to eliminate the mapping deviation between the characteristic pattern and the original image.Experimental results show that compared with the Faster R-CNN algorithm,the average detection accuracy of the improved algorithm increases by 5.1%and reaches 86.64%,which can meet the detection needs in the case of large amounts of ship data.