【作者】 陈鹏；

【导师】 刘雪松；

【作者基本信息】 浙江大学 ， 电子信息技术及仪器， 2022， 硕士

【摘要】 随着人们对海洋资源需求的增长,三维声呐系统凭借广阔的探测范围成为水下探测领域的研究热点,多用于水下场景的图像显示。同时,三维声呐系统的目标识别功能受其背景噪声影响,多依赖于人工判别,自动识别准确率和实时性较低。因此,研发高准确率的三维声呐数据处理及目标识别系统具有较大的工程应用价值。论文针对三维声呐水下目标识别的工程应用需求,研发了三维声呐数据处理及目标识别系统软件。该系统软件包含基础业务功能模块、硬件驱动模块、点云数据预处理模块和点云目标识别模块。基础业务功能模块应用自主设计的数据传输、文件存储和命令交互协议,实现声呐数据的传输、存储和命令交互等功能。硬件驱动模块实现了硬件数据传输、命令透传和代码升级功能。点云数据预处理模块实现了DFT数据接收,通过GPU加速的波束形成算法完成了点云数据转换,并利用优化后的RANSAC算法、统计滤波算法和阈值滤波算法实现了水下目标的前景提取工作。点云目标识别模块利用了优化的KMeans聚类算法加速水下目标的聚类过程,并提取出目标点云的空间尺度特征参数和基于ICP配准算法的模板匹配特征参数完成决策树算法的模型分类器训练,实现了对水下蛙人等目标的实时分类识别功能。此外,论文还研发了数据显示系统软件以实现对点云数据的显示和目标识别结果的标注。系统测试结果表明,论文研发的三维声呐数据处理及目标识别系统软件完成了数据的实时采集、处理、目标识别和显示功能。同时,系统软件的目标识别准确率较高,满足实际应用需求。更多还原

【Abstract】 As the demand for marine resources grows,3D sonar system becomes an research hotspot in the field of underwater detection because of its wide detection range,which is mostly used for the image display of underwater scene.At the same time,the target recognition function of 3D sonar system is affected by its background noise,and mostly depends on manual discrimination,and the automatic recognition accuracy and realtime performance are low.Therefore,the development of 3D sonar data processing and target recognition system high accuracy has important engineering significance.Aiming at the engineering application requirements of 3D sonar underwater target recognition,this thesis develops the software of 3D sonar data processing and target recognition system.The software includes basic business function module,hardware driver module,point cloud data preprocessing module and point cloud target recognition module.The basic business function module applies the self-designed protocols of data transmission,file storage and command interaction to realize the functions of sonar data transmission,storage and command interaction.The hardware driver module realizes the functions of data transmission,command transparent transmission and code upgrade.The point cloud data preprocessing module realizes the reception of DFT data,completes the point cloud data conversion based on GPU accelerated beamforming calculation,and realizes the prospect extraction of underwater targets by using the optimized RANSAC algorithm,statistical filtering algorithm and threshold filtering algorithm.The point cloud target recognition module uses the optimized KMeans clustering algorithm to accelerate the clustering process of underwater targets,then extracts the spatial scale feature parameters of target point cloud and the template matching feature parameters based on ICP registration algorithm,which helps achieve the goal of underwater frogman real-time classification recognition.In addition,the data display system software is developed to display the point cloud data and label the target recognition results.Experimental results show that the 3D sonar data processing and target recognition system software developed in this thesis completes the functions of real-time data acquisition,processing,target recognition and display.At the same time,the target recognition accuracy of the system software is high,which meets the needs of practical application.更多还原