ABSTRACT

Unmanned Aerial Vehicles (UAVs), in recent years, are developing rapidly. However, when fly into private residences or public areas without authorizations, UAVs pose latent threats to personal privacy and public security. UAVs localization is a significant part of an anti-UAV system. In this paper, a remolded acoustic energy decay model preserved relative in acoustic energy attenuation inverse of distance square is used to generate training data. Multilayer perceptron(MLP) is the model to train these data and predicts accurate relative 3D space coordinates. Four different UAV flight trajectories are simulated. We also test robustness against noise with different levels. Simulation experiment results show that the deviation is less than 1.48 m in specific distances and noise levels, even with higher noise levels the deviation can still be accepted. The problem of limited detection range is overcome by the use of wireless sensor networks (WSNs) with more sensors. Long and short-term memory (LSTM) is investigated, but it doesn’t outperform MLP in accuracy and processing time.

摘要

近年来，无人飞行器（UAV）发展迅速。但是，未经许可而飞入私人住宅或公共区域时，无人机会对个人隐私和公共安全构成潜在威胁。无人机定位是反无人机系统的重要组成部分。在本文中，使用保留的与距离平方成反比的声能衰减逆相关的重塑声能衰减模型来生成训练数据。多层感知器（MLP）是训练这些数据并预测准确的相对3D空间坐标的模型。模拟了四种不同的无人机飞行轨迹。我们还测试了针对不同级别噪声的鲁棒性。仿真实验结果表明，在特定距离和噪声水平下，该偏差小于1.48 m，即使噪声水平较高，该偏差仍可以接受。通过使用具有更多传感器的无线传感器网络（WSN）可以克服检测范围受限的问题。研究了长期和短期记忆（LSTM），但在准确性和处理时间方面均不胜过MLP。