基于连续数据量化的声纳传感器数据识别方法

为了有效识别声纳信号,提出一种基于连续数据量化的声纳传感器数据识别方法。首先用声学传感器采集数据;其次运用数据离散化方法,有效地将采集到的数据进行连续数据离散化;最后,通过数据挖掘技术(C4.5/C5.0决策树、SVM和Naive-Bayes分类器)对离散后的声纳信号数据进行分类识别。实验首先在UCI数据集上进行Naive-Bayes分类预测来评价提出离散化方法的性能,得到了较好的效果。其后,通过声学传感器收集到的Sonar数据集进行实验。结果表明,新的离散化方法提高了四个分类器的识别精度,表明该声纳传感器数据识别技术是非常有效的。

Sonar Sensor Data Recognition Method Based on Continuous Data Quantization

In order to effectively recognize sonar signals, a sonar sensor data recognition method based on discretization method was presented in this paper. First, this technique was based on the data sampled by acoustic sensors. Second, a novel data discretization method was used to effectively discretize continuous data. Finally, the discretized sonar signals data were classified and recognized by data mining techniques, i.e., C4.5/C5.0 decision tree, SVM and Naive-Bayes classifiers. We first evaluate the performance of the proposed discretization method by building Naive-Bayes classifier on UCI data set. Experimental results demonstrate that the proposed discretization method improves the classification ability of Naive-Bayes classifier. In addition, the new discretization method can improve the accuracy of four classifiers on sonar data set collected by acoustic sensors, and this shows that the sonar sensor data recognition technique was very effective.