一种多传感器信息融合的噪声源识别方法研究

Identification of noise sources based on multi-sensors information fusion

水下航行器噪声源识别是一个小样本条件下的模式识别问题。充分利用多个传感器采集的信号是解决小样本问题的有效途径。但是,目前各个传感器在整体评估中所占的权重没有一种合理的评估方法。文章利用直推式置信机(TCM)可以给出分类预测置信的能力,首先提出一种改进的奇异值测量方法,提高计算预测置信的准确性。然后将该置信作为传感器权重的有效表征,提出了一种多传感器信息融合的改进型直推式置信机算法,即TCM-IKNN-M(Transductive Confidence Machine for Improved K-Nearest Neighbors based on Multi_sensors)算法。舱段模型试验表明,文中提出的算法有效地利用了多个传感器的信息,大大提高了识别的正确率。

Identification of underwater vehicle mechanical noise sources can be considered as a pattern recognition problem on small samples.Using the signals of multiple sensors is one of the most effective methods to solve the problem.Currently there is no appropriate method to calculate the weightiness of each sensor during the evaluation.In this paper,the transductive confidence machine（TCM） was used to calculate the confidence of classification prediction,which was an effective representation for the weightiness of each sensor.First,an improved strangeness measuring method was given to increase the accuracy of confidence prediction.Then a new algorithm,named TCM-IKNN-M（Transductive Confidence Machine for Improved K-Nearest Neighbors based on Multi_sensors）,was proposed based on the predicted confidence and improved strangeness measuring method.The results of the experiment conducted on a cabin model show that TCM-IKNN-M algorithm can greatly increase the right rate of identification by fusing the information from multiple sensors.