摘要
为解决传统无人机姿态估计过程中出现的精度低、实时性差、容易被干扰、模型复杂、计算量大、处理时间长等问题,研究基于深度学习的四轴无人机飞行姿态估计方法.结合多传感器数据融合技术,使用MARG传感器和光流传感器采集无人机飞行姿态数据,经小波阈值去噪算法和最大—最小标准化算法对采集到的数据进行去噪及归一化预处理后,基于GRU神经网络构建一个新的姿态估计模型.将预处理后传感器的数据输入模型中进行融合,通过模型训练输出重力矢量和地磁矢量,经无人机姿态角转换获取无人机飞行姿态.实验结果表明:基于深度学习的四轴无人机飞行姿态估算结果与实际值接近,且误差小、速度快,同时利用GRU神经网络可以融合多种传感器的数据,无需建立复杂的数学模型,即可实现无人机飞行姿态的高精度测量.
In order to solve the problems of low accuracy,poor real-time performance,easy interference,complex model,high computational complexity,and long processing time in the traditional unmanned aerial vehicle attitude estimation process,a deep learning based four axis unmanned aerial vehicle flight attitude estimation method is studied.By combining multi-sensor data fusion technology,MARG sensors and optical flow sensors are used to collect drone flight attitude data.After denoising and normalizing the collected data using wavelet threshold denoising algorithm and maximum minimum normalization algorithm,a new attitude estimation model is constructed based on GRU neural network.Input the preprocessed sensor data into the model for fusion,output the gravity vector and geomagnetic vector through model training,and obtain the drone flight attitude through drone attitude angle conversion.The experimental results show that the flight attitude estimation results of the four axis unmanned aerial vehicle based on deep learning are close to the actual values,with small errors and fast speed.At the same time,the GRU neural network can fuse data from multiple sensors,and high-precision measurement of unmanned aerial vehicle flight attitude can be achieved without the need to establish complex mathematical models.
作者
潘美琴
PAN Mei-qin(School of Digital Information Engineering,Minjiang Teachers College,Fuzhou Fujian 350018,China)
出处
《广州航海学院学报》
2023年第4期48-54,共7页
Journal of Guangzhou Maritime University
基金
2021年福建省教育厅中青年教育科研项目(JAT210824)。
关键词
深度学习
四轴无人机
飞行姿态估计
光流传感器
GRU神经网络
姿态角
Deep learning
Four axis unmanned aerial vehicle
Flight attitude estimation
Optical flow sensor
GRU neural network
Attitude angle
