基于多分类器融合的卫星图像舰船目标识别

Ship Target Recognition from Satellite Images base on Multi-Classifier Fusion

针对遥感图像中舰艇目标识别性能低的问题,提出了加权投票分类器融合方法。首先分析了舰艇的颜色特征与轮廓特征,然后利用SVM、BP神经网络和AdaBoost算法训练三种单分类器,最后采用加权投票方式对单分类器进行融合。采用融合分类器进行舰艇目标识别实验分析,实验结果表明:在google卫星图像舰艇目标识别中,所提方法能够有效提升舰艇目标识别准确率,F-measure可以达到73.54%,相较于SVM提升了2.72%,相较于AdaBoost提升了3.53%,相较于BP神经网络提升了4.28%。

Aiming at the low performance of ship target recognition in remote sensing images, a weighted voting classifier fusion method is proposed. Firstly, the color and contour features of naval vessels are analyzed. Then three single classifiers are trained by SVM algorithm, BP neural network and AdaBoost algorithm. Finally, the single classifier is fused by weighted voting. Fusion classifier is used to carry out experimental analysis of ship target recognition. The experimental results show that the proposed method can effectively improve the accuracy of ship target recognition in Google satellite image. And the F-measure can reach 73.54%, compared with SVM, AdaBoost and BP neural network, increased by 2.72%, 3.53% and 4.28% respectively.