水下运动目标被动声纳信号识别研究

发布时间：2018-07-10 09:25

本文选题：水下运动目标识别 + 特征提取　；参考：《昆明理工大学》2014年硕士论文

【摘要】：水下运动目标识别技术一直是水声信号处理技术中的重要研究内容。水下运动目标识别技术的研究不仅对国防建设有着非常重要的价值,对于商用和民用领域也有着重要的应用价值,因此受到世界各国的广泛关注,是当今世界各国相关领域的研究热点。本文主要针对被动声纳信号识别技术、围绕信号特征提取方法和分类器的设计进行研究,并通过实验仿真,对本文提出的分类方法进行了验证。特征提取和选择是目标识别过程中的首要环节,提取有效而稳定可靠的特征是保证分类识别准确率的前提。本文首先深入探讨了小波变换和小波包变换的基本概念和原理,引入小波包分析技术提取了水下运动目标被动声纳信号的能量特征；然后,探索了经验模式分解算法的原理,将绝对均值引入到水下运动目标被动声纳信号的特征提取中来,提出了在经验模式分解的基础上对分解后得到的固有模态函数的绝对均值进行计算的特征提取方法,并且从数据融合的角度出发,结合特征融合技术,构造了一种新的分类特征向量。分类器的设计是目标识别过程中的第二环节,分类器设计的好坏和适应性将直接影响整个系统最终的识别性能。本文详细讨论了模糊C均值(FCM)算法的原理及神经网络的相关理论,并就FCM算法和广义回归神经网络(GRNN)的优缺点进行了分析：基于FCM算法的性能在很大程度上依赖于随机的初始聚类中心,本身是无监督算法,容易陷入局部极值；而广义回归神经网络(GRNN)隐含层结点中的作用函数采用高斯函数,具有全局逼近能力。基于此,提出了基于FCM和GRNN的一种结合算法,形成一种新分类器,对前述的特征向量进行分类识别。实验结果表明,基于FCM和GRNN的结合算法的目标正确识别率相对较高,达到了研究目的。
[Abstract]:Underwater moving target recognition technology has been an important research content in underwater acoustic signal processing technology. The research of underwater moving target recognition technology not only has very important value for national defense construction, but also has important application value for commercial and civil fields. Nowadays, it is a hot spot in the relevant fields of the world. This paper mainly focuses on the passive sonar signal recognition technology, focusing on the signal feature extraction method and the design of the classifier, and through experimental simulation, the proposed classification method is verified. Feature extraction and selection is the first step in the process of target recognition, and the extraction of effective and stable features is the prerequisite to ensure the accuracy of classification recognition. In this paper, the basic concepts and principles of wavelet transform and wavelet packet transform are discussed, the energy characteristics of passive sonar signal of underwater moving target are extracted by wavelet packet analysis technique, and the principle of empirical mode decomposition algorithm is explored. The absolute mean value is introduced into the feature extraction of passive sonar signal of underwater moving target. Based on the empirical mode decomposition, a feature extraction method for calculating the absolute mean value of the decomposed inherent mode function is proposed. From the point of view of data fusion, a new classification feature vector is constructed based on feature fusion technology. The design of classifier is the second step in the process of target recognition. The design and adaptability of classifier will directly affect the final recognition performance of the whole system. In this paper, the principle of fuzzy C-means (FCM) algorithm and the related theory of neural network are discussed in detail. The advantages and disadvantages of FCM algorithm and generalized regression neural network (GRNN) are analyzed. The performance of FCM algorithm depends on the random initial clustering center to a great extent, and it is an unsupervised algorithm, which is easy to fall into local extremum. The generalized regression neural network (GRNN) uses Gao Si function for the function in the hidden layer node, so it has the ability of global approximation. Based on this, a combined algorithm based on FCM and GRNN is proposed to form a new classifier to recognize the feature vectors mentioned above. The experimental results show that the combined algorithm based on FCM and GRNN has a relatively high recognition rate and achieves the purpose of the research.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.3

【参考文献】