基于多特征决策融合的SAR飞机识别

SAR image recognition based on multi-feature decision fusion for plane

针对高分辨率SAR图像的飞机目标识别问题,提出了一种基于飞机几何特征、PCA特征、Hu不变矩等多特征决策融合的自动目标识别方案。针对飞机样本特点,分别提取飞机的几何长宽特征、PCA特征和Hu不变矩特征,使用三个支持向量机分类器分别对样本的三类特征进行预分类,然后采用基于等级的决策融合方法将预分类结果进行决策融合,输出最终的目标类别。实验过程中,随机选取一定百分比的样本进行训练,获得分类器模型,对全部的样本进行测试识别。通过实验发现,将几何特征、PCA特征和Hu不变矩特征的分类结果进行决策融合后,克服了单一特征决策的不准确性,有效地提高了每一类样本的识别准确率。

In order to recognize the plane target of high resolution synthetic aperture radar （SAR） image, an automatic tar- get recognition scheme based on multi-feature decision fusion of the plane＇s geometric feature, principal component analysis （PCA） feature and Hu moment invariant is presented. Aiming at the characteristics of plane sample, the geometric length-width feature, PCA feature and Hu moment invariant feature are extracted respectively from SAR images. Three support vector ma- chine classifiers are used to presort the three features of the sample respectively. And then the ranking-based decision fusion method is used to decide and fuse the presort results, and output the final target class. The sample with a certain percentage is randomly selected in the experiment process for training. The classifier model was obtained to test and recognize the whole sam- pies. The experimental results show that the decided and fused classification results of geometric feature, PCA feature and Hu moment invariant feature overcame the inaccuracy of using single feature decision, and improved the recognition accuracy of each type sample effectively.