中国激光, 2019, 46 (10): 1010001, 网络出版: 2019-10-25
基于深度学习的脉冲激光测距回波时刻解算方法 下载: 1328次
Method for Solving Echo Time of Pulse Laser Ranging Based on Deep Learning
遥感 脉冲激光测距 回波时刻解算 深度学习 卷积神经网络 激光雷达 remote sensing pulsed laser ranging echo time solving deep learning convolutional neural network lidar
摘要
为了提高脉冲激光测距回波时刻解算方法的应用场景适应性,将回波时刻解算问题转换为波形分类的问题,采用深度学习的新方法实现回波时刻的解算。通过仿真模拟计算产生0.1 ns时间分辨率的不同距离、信号幅度、波形形状和噪声的样本回波数据,训练一维卷积神经网络模型,在样本测试集上获得了99.85%的分类精度;采用深度学习方法和高斯拟合方法处理同样的机载激光雷达回波数据,墙面线扫数据解算结果相关系数为0.99981,外场飞行试验数据平面拟合残差均在20 mm左右,两种方法回波时刻解算效果相当。结果表明,新方法能够满足机载脉冲激光测距回波时刻解算要求,具备进一步提高解算精度和适应更多应用场景的潜力。
Abstract
To improve the application adaptability of the method used for solving the echo time of pulse laser ranging, this study transforms the echo time solving problem into a waveform classification problem and uses a novel deep learning method to solve the echo time. Further, a one-dimensional convolutional neural network model is trained by simulating the sample echo data containing different distances, signal amplitudes, waveform shapes, and noises with a time resolution of 0.1 ns, and a classification accuracy of 99.85% is obtained using the sample test set. Using the deep learning method and the Gaussian fitting method to process the airborne lidar echo data, the correlation coefficient of the wall surface sweep measurement results is 0.99981. Further, the plane fitting residuals of the field flight test data are approximately 20 mm; the effects of the two methods are observed to be equivalent. The results denote that the proposed method can satisfy the requirements for solving the echo time of airborne pulse laser ranging and can improve the solution accuracy and adapt to several application scenarios.
胡善江, 贺岩, 俞家勇, 吕德亮, 侯春鹤, 陈卫标. 基于深度学习的脉冲激光测距回波时刻解算方法[J]. 中国激光, 2019, 46(10): 1010001. Shanjiang Hu, Yan He, Jiayong Yu, Deliang Lü, Chunhe Hou, Weibiao Chen. Method for Solving Echo Time of Pulse Laser Ranging Based on Deep Learning[J]. Chinese Journal of Lasers, 2019, 46(10): 1010001.