提出一种基于概率潜在语义分析(Probabilistic Latent Semantic Analysis, PLSA)的检测算法, 首先通过PLSA将目标表述为潜在成分的概率组合, 然后利用统计模式识别方法对获取的潜在成分概率进行判别, 从而完成最终的检测。其中, 生成的潜在成分反映了目标与特征之间相互出现的频率关系, 并以潜在成分在目标中概率差异的形式对上述不对应现象给出了直观描述。实验结果表明, 所提出的算法对多种复杂情况下的光学图像舰船检测具有很好的准确性和鲁棒性。

Ship detection is one of the important areas in remote sensing applications. However, many ship detection approaches often face a difficult dilemma between low detection rate and high false rate, because of the un-matching between object and its features caused by the complicated characteristics of remote sensing images. Therefore, this paper proposes a novel detection algorithm based on Probabilistic Latent Semantic Analysis (PLSA). It firstly describes the object in terms of the probability combination of latent aspects generated by PLSA, then discriminates the latent aspects model of object by statistics recognition method to obtain the final detection result. The generated latent aspects model represents the joint probability of objects and their features, and gives an explanation for the above un-matching problem by the probability distribution of latent aspects. The performance of the proposed algorithm is demonstrated through the ship detection in various optical remote sensing images, and substantiated using quantitative criteria.