基于时空相关的水下移动对象建模技术研究

发布时间：2018-06-07 19:17

本文选题：目标识别 + 水下移动对象　；参考：《哈尔滨工程大学》2016年硕士论文

【摘要】：地理信息系统(Geographic Information System)和水声传感器网络(Underwater Acoustic Sensor Networks)的快速发展使得水下移动对象的跟踪和监测成为现实。然而在水下未知目标识别的研究中,传统方法针对恶劣的水下环境、复杂的水下移动对象运动轨迹和目标分类识别还存在不足。本文主要针对以上不足,对水下移动对象识别过程中的时空数据模型和识别方法进行研究。主要内容如下:1)系统地研究水下移动对象的轨迹特征和属性特征,并详细分析这些特征对水下移动对象识别精度的影响。2)针对水下移动对象轨迹的时空特点,建立水下移动对象模型(Underwater Moving ObjectDatabase,UMOD)。首先,提出了基于球坐标的轨迹分解方法,降低了轨迹的整体复杂程度;其次,UMOD模型采用移动对象时空模型中的动态属性方法,提出了针对水下未知目标的动态阈值位置更新策略,辅助轨迹分解并对误差点进行排除;最后,提出了轨迹拟合策略,对水下移动对象进行全时域轨迹描述,从而系统地得到水下移动对象运动过程中的轨迹识别特征。3)分析和研究水下移动对象的轨迹特征和属性特征,建立水下移动对象特征数据库,同时提出水下移动对象混合特征识别方法(Underwater Moving Objects Mixed Feat-ure Recognition Method,UMOMFRM)。首先,UMOMFRM通过模糊推理的方法排除目标的最小概率分类;其次,采用贝叶斯分类算法计算未知目标对应剩余项目的概率;最后,对这些概率进行概率阈值筛选并进行最大概率比较,若存在最大概率,则其所对应的类别为分类结果,否则该目标所属类别为其他类。4)采用模拟数据集进行对比实验。首先,根据本文提出的UMOD模型对轨迹数据进行处理,验证UMOD模型的准确性。最后,进行UMOMFRM与传统水下目标识别方法的对比试验,比较识别率,以验证UMOMFRM的准确性和有效性。
[Abstract]:With the rapid development of Geographic Information system (GIS) and underwater Acoustic Sensor Networks (underwater Sensor Networks), the tracking and monitoring of underwater moving objects becomes a reality. However, in the research of underwater unknown target recognition, the traditional method is insufficient for the poor underwater environment, the complex underwater moving object moving track and the target classification recognition. In this paper, the spatiotemporal data model and recognition method of underwater moving object recognition are studied. The main contents are as follows: 1) systematically studying the trajectory and attribute characteristics of underwater moving objects, and analyzing in detail the influence of these characteristics on the recognition accuracy of underwater moving objects. 2) aiming at the space-time characteristics of underwater moving objects, An underwater moving object model was established. Firstly, a trajectory decomposition method based on spherical coordinates is proposed to reduce the overall complexity of the trajectory. Secondly, the UMOD model adopts the dynamic attribute method in the moving object space-time model. A dynamic threshold position updating strategy for underwater unknown targets is proposed to assist trajectory decomposition and eliminate error points. Finally, a trajectory fitting strategy is proposed to describe the trajectory of underwater moving object in full time domain. Thus, the trajectory identification features of underwater moving objects are obtained systematically. 3) the trajectory characteristics and attribute characteristics of underwater moving objects are analyzed and studied, and the feature database of underwater moving objects is established. At the same time, a hybrid feature recognition method for underwater moving objects is proposed. First, UMOMFRM eliminates the minimum probability classification of targets by fuzzy reasoning; secondly, Bayesian classification algorithm is used to calculate the probability of unknown targets corresponding to the remaining items; finally, These probabilities are screened by probability threshold and compared with the maximum probabilities. If there is a maximum probability, the corresponding category is the result of classification, otherwise, the target belongs to another class. 4) A comparison experiment is carried out by using simulated data sets. Firstly, the trajectory data are processed according to the proposed UMOD model to verify the accuracy of the UMOD model. Finally, the accuracy and validity of UMOMFRM are verified by comparing UMOMFRM with traditional underwater target recognition methods.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：U675.79;TP212.9;TN929.3

【参考文献】