论文部分内容阅读
洛克希德·桑德斯公司正在研究综合声与雷达数据的融合过程。声和雷达数据是高度互不相关的,它们可提供有关每个目标的互补信息。因此,融合过程使之有可能利用组合分类器的输出产生一种可信度高的目标识别(可靠的ID),分类器的输出包括:声波形、雷达信号特征和航迹形状 而由单个分类器进行可靠的识别是不可能的。通过融合过程可以减少目标位置误差和改善量测与航迹的互连过程。 由于融合过程是高度非线性和自适应的,因此,仿真是必不可少的。用多次重复的具有嵌入式传感器模型的仿真来建立多目标情景,以便激励融合过程。在仿真情景期间,用测量的声和雷达数据产生传感器输出的逼真描述。融合过程对分类和传统的跟踪器使用了神经网站。使用与目标ID有关的数据,以便选择最优的跟踪器参数。 对雷达和声数据的特定实例,验证了使用仿真来设计传感器融合处理的方法。然而,这种方法对于各类型的传感器或数据融合也是有价值的。例如:可在非军事应用中使用仿真,在这种应用中有关复杂系统状态的多源数据可用来控制其他数据和收集并做出最优决策。一般,仿真可用任务级指标,如成本、生存概率和杀伤概率来评定融合过程的好处。仿真对于优化融合过程的任何自适应“子部件”如神经网络来说是必不可少的。
Lockheed Sanders is studying the integration of integrated sound and radar data. Sound and radar data are highly independent and provide complementary information about each goal. Therefore, the fusion process makes it possible to produce a highly reliable target identification (reliable ID) using the output of the combined classifier, which includes the output of the acoustic waveform, radar signal characteristics and track shape, It is impossible to reliably identify the device. Through the fusion process can reduce the target position error and improve the measurement and track the interconnection process. As the fusion process is highly nonlinear and adaptive, simulation is essential. Multi-iterative simulations with embedded sensor models are used to create multi-objective scenarios to stimulate the fusion process. During the simulation scenario, a realistic description of the sensor output is generated using the measured sound and radar data. The fusion process used neural websites for classification and traditional trackers. Use data related to the target ID to choose the best tracker parameters. A specific example of radar and acoustic data validates the design of a sensor fusion process using simulation. However, this method is also valuable for various types of sensors or data fusion. For example: Emulation can be used in non-military applications in which multi-sourced data on complex system states can be used to control other data and collect and make optimal decisions. In general, simulations evaluate the benefits of the fusion process with task-level metrics such as cost, survival probability, and kill probability. Simulation is essential to any adaptive “subcomponent” that optimizes the fusion process, such as neural networks.