基于地震动的地面移动目标识别研究

发布时间：2018-02-08 12:35

本文关键词： 目标识别地震动传感器小波包变换经验小波变换支持向量机遗传算法　出处：《中国民航大学》2017年硕士论文　论文类型：学位论文

【摘要】：本文利用单轴加速度传感器来采集地面移动目标的地震动信号,通过对信号的分析处理以及对特征量的提取,设计目标识别算法,从而实现对目标的识别。运用地震动信号对目标进行识别在跑道入侵防范以及安全监测应用例如机场周界、区域保护等方面都有很大的应用价值。本文首先研究了地震波的产生与传播机理,介绍了地震动加速度传感器的工作原理。由于地面运动目标产生的地震动信号易受环境噪声的影响,采用小波包变换对信号先进行去噪处理。然后运用经验小波变换提取信号的IMF分量,与经验模态分解方法相比,该方法分解得到的模态更少而且计算速度更快。再对小波包去噪后的信号进行过零分析并提取去噪后信号的各项时域和频域参数指标;同时提取基于互相关准则选取的各IMF分量的能量指标,最后将这些特征参数进行组合构造特征向量,以此作为多分类支持向量机的输入进行训练和预测。针对传统支持向量机参数难以选择的问题,利用遗传算法参数寻优,找到支持向量机模型的最佳参数:惩罚因子C和核参数g。实验结果表明,采用改进支持向量机模型即GA-SVM模型对特征向量进行训练和测试,能较好地提升分类器性能。而且遗传算法优化支持向量机模型在训练速度和识别效率上优于未改进的交叉验证法支持向量机模型。该模型对地面移动目标如人员、车辆还有飞机能进行正确的识别分类。
[Abstract]:In this paper, the uniaxial acceleration sensor is used to collect the ground motion signal of the moving target on the ground. By analyzing and processing the signal and extracting the characteristic quantity, the target recognition algorithm is designed. In order to realize the recognition of the target, using the ground motion signal to identify the target in the runway intrusion prevention and security monitoring applications such as airport perimeter, Regional protection has great application value. In this paper, the mechanism of seismic wave generation and propagation is studied. The working principle of the ground motion acceleration sensor is introduced. The ground motion signal produced by the ground motion target is easily affected by the environmental noise. The wavelet packet transform is used to Denoise the signal, and then the IMF component of the signal is extracted by the empirical wavelet transform, which is compared with the empirical mode decomposition method. The wavelet packet de-noising signal is analyzed by zero-crossing analysis and the parameters of time domain and frequency domain of de-noised signal are extracted. At the same time, the energy indexes of each IMF component selected based on the cross-correlation criterion are extracted. Finally, the characteristic parameters are combined to construct the eigenvector. It is used as input of multi-classification support vector machine to train and predict. Aiming at the problem that traditional support vector machine parameters are difficult to select, genetic algorithm is used to optimize the parameters. The experimental results show that the improved support vector machine (GA-SVM) model is used to train and test the feature vectors. It can improve the performance of classifier. Moreover, the genetic algorithm optimized support vector machine model is better than the unimproved cross validation support vector machine model in training speed and recognition efficiency. Vehicles and aircraft can be correctly identified and classified.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212;TN911.7

【相似文献】