基于卷积神经网络的机载激光海洋测深波形分类  

Waveform Classification of Airborne Laser Bathymetry Based on Convolutional Neural Network

作　　者：刘超 LIU Chao(Tianjin Surveying&Designing Institute for Water Transport Engineering Co.,Ltd.,Tianjin 300450 China)

机构地区：[1]天津水运工程勘察设计院有限公司,天津300450

出　　处：《自动化技术与应用》2024年第3期48-51,56,共5页Techniques of Automation and Applications

摘　　要：海洋测深的波形分类的结果会影响整体海洋勘测的精度,为降低波形分类误差,提高海陆分界精度,基于卷积神经网络研究机载激光海洋测深波形分类方法。提取海洋测深回波,明确机载激光雷达的测深原理,计算反射回波强度,以计算水深深度;基于卷积神经网络提取候选回波,得到目标函数的响应卷积曲线,获得反卷积的函数估计值,对其进行过滤操作;设计波形分类算法,计算卷积运算输出的波形尺寸,判断信号检测通过的标准复杂度,以获得海洋测深波形的分类方法。在四种波形分类方法的对比中,卷积神经网络算法的分类精度均大于95%,分类精度更高,更优。The result of waveform classification of oceanographic bathymetry affects the accuracy of the overall oceanographic survey.In order to reduce the error of waveform classification and improve the accuracy of sea-land boundary,the method of airborne laser oceanographic bathymetry waveform classification is studied based on convolutional neural network.It extracts ocean sounding echoes,clarifies the sounding principle of airborne lidar,calculates the reflected echo intensity to calculate the water depth;extracts candidate echoes based on convolutional neural network,obtains the response convolution curve of the objective function,and obtains the reverse volume.The function estimation value of the product is used to filter it,the waveform classification algorithm is designed,the waveform size of the output of the convolution operation is calculated,and the standard complexity of the signal detection is judged to obtain the classification method of the ocean sounding waveform.In the comparison of the four waveform classification methods,the classification accuracy of the convolutional neural network algorithm is greater than 95%,and the classification accuracy is higher and superior.

关 键 词：卷积神经网络 机载激光 海洋测深 波形分类 Kappa系数 多航线 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]