基于改进的YOLOv3算法的船舶目标检测

Ship Target Detection Based on Improved YOLOv3 Algorithm

针对水面上目标尺度多样、背景复杂等问题,论文提出了一种改进的YOLOv3算法。通过加上自适应特征融合ASFF网络结构,从而能够有效地融合不同尺度的特征,提高检测网络对于多尺寸目标的检测能力;通过使用CIOU损失函数,提高复杂背景下检测框的定位精度;通过利用软非极大值抑制算法Soft-NMS算法,解决了多目标重叠下目标框误删的问题。实验结果表明,论文提出的方法检测精度优于其他算法,而且对小目标的检测也有了一定的提升。

Aiming at the problems of various target scales and complex background on the water surface,this paper proposes an improved YOLOv3 algorithm.By adding the adaptive feature fusion ASFF network structure,it can effectively fuse features of dif-ferent scales and improve the detection ability of the detection network for multi-size targets.By using the CIOU loss function,the positioning accuracy of the detection frame under complex background can be improved.By using the soft non-maximum suppres-sion algorithm Soft-NMS algorithm solves the problem of erroneous deletion of target frames under multi-target overlapping.Experi-mental results show that the detection accuracy of the method proposed in this paper is better than other algorithms,and the detec-tion of small targets is improved to a certain extent.