基于海岸线候选区域的港口舰船目标检测算法研究

Research on Port Ship Detection Based on Coastline Candidate Area

随着光学遥感卫星分辨率的不断提高,利用高分辨率遥感图像对港口区域舰船目标进行检测能够为各种领域提供非常有价值的信息。由于港口位于海陆交界处,背景较为复杂,陆地区域为舰船目标检测带来了干扰信息。同时,港口停靠舰船尺度大小不一,对舰船检测增加了难度。针对卫星遥感影像中港口舰船背景复杂以及尺度大小不一的问题,提出一种抗干扰、抗尺度变化的舰船检测方法。对遥感影像进行海陆分割,基于海陆分割结果,获取较为准确的海岸线;沿海岸线按照特定尺寸以一定重叠率进行滑窗,生成海岸线候选框;结合特征金字塔网络构建舰船检测模型,完成对港口舰船目标的检测。实验结果表明,提出的算法能够有效去除陆地和海上无关背景信息的干扰,能够检测出不同尺度的舰船目标,大大提高了大幅遥感影像的检测速度,在遥感影像的港口舰船检测中可以获得较高的准确率。

With the continuous improvement of the resolution of optical remote sensing satellites,using high-resolution remote sensing images to detect ship objects in port area can provide very valuable information for various fields.To address the background complexity and scale variety of docked ships in remote sensing images,an anti-interference and anti-scale variety method of ship detection is proposed.Firstly,the remote sensing image is segmented into parts of sea and land,based on which more accurate coastline is obtained,and then sliding window is conducted along the coastline at a certain overlap rate according to specific size to generate coastline candidate box.Finally,combined with the feature pyramid network,the ship detection model is built to complete the detection of port ship objects.The experimental results show that the proposed algorithm can effectively avoid the interference of background information and detect the ship objects of different scales,achieving higher speed and accuracy in ship detection of remote sensing images.