基于目标分解和加权SVM分类的极化SAR图像舰船检测

Ship Detection in Polarimetric SAR Imagery Based on Target Decomposition and Weighted SVM Classifier

目标分解是极化SAR图像舰船检测的重要方法,但是,在较高分辨率和复杂海况条件下,由相干矩阵分解得到的极化熵参数并不能将舰船目标与海洋等背景完全区分。对极化目标分解理论和地物散射机理的研究和分析表明,极化分解的各个参数从不同角度反映了目标和背景的散射差异,对它们联合使用有助于更准确地在SAR图像中检测目标。而且,各个参数在实际的检测问题中具有不同的重要性。因此,本文构造了包含有多个极化分解参数的特征向量,并根据各分解参数重要性不同,提出一种基于目标分解和加权SVM(support vector machine)分类方法对极化SAR图像中的舰船进行检测。实验结果表明,该方法能够精确地检测舰船目标,并有效地减少虚警。

Target decomposition is an important method for ship detection in polarimetric SAR imagery.Under the condition of relative high resolution and complex sea state,the contrast between ship and sea descends in polarimetric entropy space that deduced from coherence matrix eigenvalue decomposition.The analyses of the polarimetric target decomposition theory and target＇ s scattering mechanism illustrate that parameters come from target decomposition describe the difference between targets and background from different point of view.The combination use of them promotes the detection of target in SAR imagery.However,each parameter has its own diverse significance in the practical detection problem.Therefore,this paper proposes an SVM classification method to detect ships in PolSAR（Polarimetric SAR） imagery.Firstly,the method constructs a feature vector consists of several decomposition parameters;and then,different decomposition parameters are weighted according their essentiality in the SVM classifier;ships are classified from sea background and other false alarms by the classifier in the end.Experiment results illustrate that the method detects ship targets more precisely and reduces false alarms effectively.