基于GA-LPA算法的船舶图像识别方法研究

Study on ship image recognition based on GA-LPA algorithm

随着海事监管自动化的发展,大量船舶图像需要进行自动标注和追踪,传统的图像信息标注方式已经不能适应海事监管的需求,基于内容的图像信息标注技术在海事监管方面得到越来越多的应用。标签传播算法(label propagation algorithm,LPA)是一种复杂度较低的模式分类方法,适合于处理船舶图像的标注和追踪问题,为此,分别介绍了标签传播算法和遗传算法(genetic algorithm,GA)的原理,分析了LPA算法的参数确定问题,并针对此问题提出了基于GA的LPA算法,建立了GA-LPA算法用于船舶图像识别的方法。通过实例进行计算分析,验证了采用GA-LPA算法进行船舶图像标注和识别的可行性和效率。

With the development of maritime supervision automation,a large number of ship images need automatic tagging and tracking. The traditional image information annotation could not meet the needs of maritime supervision,so image annotation technology based on content have been applied in maritime supervision. The label propagation algorithm（ LPA） is a kind of semi supervised learning algorithm based on graph,suitable for treatment of ship image tagging and tracking problems.The principles of label propagation algorithm and genetic algorithm（ GA） are introduced,and the LPA algorithm＇s parameters are analyzed,and the LPA algorithm based on GA（ named GA-LPA） for ship image recognition is proposed. By the calculation and analysis of the examples,the feasibility and efficiency of the ship image annotation and recognition based on GA-LPA algorithm are verified.