两种SVM 集成水下目标识别方法的比较

Comparison of Two Underwater Target Recognition Algorithms Based on SVM Integration

分类器集成可有效提高分类器的识别精度和稳定性。采用Bagging算法对SVM分类器进行集成,描述并讨论了B-SVM与CB-SVM两种基于SVM集成的水下目标识别算法,并利用4类实测水下目标样本进行识别实验。实验结果表明:在一定范围内,CB-SVM算法比B-SVM算法能更好地识别测试样本,识别正确率最多提高1.56%;随着抽样数的增加,训练样本集差异性减弱并引入大量无用样本,导致分类器性能下降,同时削弱了CB-SVM算法的优势。

Classifier integration can effectively improve the recognition accuracy and stability of classifier.This paper uses bagging algorithm to integrate Support Vector Machine(SVM)classifiers,describes and discusses two underwater target recognition algorithms based on SVM integration:B-SVM and CB-SVM.Four different classes of measured underwater targets’samples are used in the recognition experiment.The experimental results show that,within a certain range,CB-SVM algorithm can recognize test samples better than B-SVM algorithm.The correct recognition rate can be improved by up to 1.56%.With the increase of sampling number,the difference of training sample set decreases and a large number of useless samples are introduced,which leads to the performance degradation of the classifier and weakens the advantage of CB-SVM algorithm.