Automated processing and error detection in multibeam sonar data

由于海洋环境的复杂性，多波束声纳数据固有地具有噪声并且容易出错。最近，多波束声纳系统已经在自主或最低限度监督的船舶上实现，这些船舶要么遵循预设的指令来绘制海底地图，要么在平台从一个位置转移到另一个位置时获取数据。自主系统为传统平台提供了几个好处，即提高了安全性，因为它们可以绘制未知区域并将人员从船上移除。然而，来自这些系统的数据没有声纳操作员或数据处理器的持续监督的好处;因此，数据的问题不能立即识别和纠正，必须在后处理中进行补偿。延迟然后传播到数据采集与处理时间的不平衡，并减缓了这些数据收集平台的采用。 该研究计划将通过近实时识别系统错误、环境伪影和噪声来提高自主船舶多波束数据处理的效率。其结果将是一个通知和分类系统，以提醒操作员数据或系统问题，标记深度测量中的噪声，并减少用户与数据的交互。五个目标组成了研究计划。 1.在采集过程中识别和监控实时数据输出，以识别系统错误; 2.自动化环境变量的后处理，以尽量减少环境的不确定性; 3.对所得声纳数据进行不确定性评估; 4.使用三维深度学习建立多波束声纳噪声识别和错误分类方法; 5.分析噪声源和检测到的错误，以改善未来的自主系统; 用于训练和测试的多波束声纳数据集将从两艘船上获得：新玩法大学的调查发射“苍鹭”号，在沿海的不列颠哥伦比亚省工作，和加拿大海岸警卫队的Amundsen号船，在整个加拿大北极地区工作。计算编程，实验设计和现场经验的结合形成了一个沉浸式的HQP培训环境，为不同背景的HQP。计划HQP将介绍最新的三维深度学习算法，多波束声纳数据采集和处理协议，以及在海洋测绘重点工程研究实验室内的包容性培训环境中的统计测试。 改进自主测量平台的应用和将人员从传统的测量船中移除，将提高海底测绘的效率，限制海洋事故，并降低海底测绘的成本，特别是在偏远地区。只有一小部分加拿大和国际沃茨的地图符合现代标准，提高自主船只收集这一关键数据的能力将为加拿大人和世界各地的其他人带来好处。

Multibeam sonar data is inherently noisy and prone to errors due to the complexity of the marine environment. Recently, multibeam sonar systems have been implemented on autonomous, or minimally supervised, vessels which either follow preset instructions to map the seafloor or acquire data as the platform transits from one location to another. Autonomous systems provide several benefits to traditional platforms, namely improved safety, as they can map uncharted areas and remove personnel from the vessel. However, the data from these systems do not have the benefits of constant supervision by a sonar operator or data processor; therefore, problems with the data are not immediately identified and corrected and must be compensated for in post-processing. The delay then propagates to an imbalance in the data acquisition to processing time and has slowed the adoption of these data collection platforms. This proposed research program will improve the efficiency of multibeam data processing for autonomous vessels through near-real-time identification of system blunders, environmental artifacts and noise. The result will be a notification and classification system to alert operators of data or system problems, flag noise in the depth measurements, and reduce user interaction with the data. Five objectives make up the research program. 1. Identify and monitor real-time data outputs during acquisition to identify system errors; 2. Automate post-processing of environmental variables to minimize environmental uncertainty; 3. Perform an uncertainty assessment of resulting sonar data; 4. Establish methods for multibeam sonar noise identification and error classification using three-dimensional deep learning; 5. Analyze noise sources and detected errors to improve future autonomous systems; Multibeam sonar datasets for training and testing will be acquired from two vessels: the University of New Brunswick survey launch Heron, working in coastal British Columbia, and the Canadian Coast Guard Ship Amundsen, working throughout the Canadian Arctic. The combination of computational programming, experimental design, and field experience form an immersive HQP training environment for HQP of diverse backgrounds. Program HQP will be introduced to the latest in three-dimensional deep learning algorithms, multibeam sonar data acquisition and processing protocols, and statistical testing in an inclusive training environment within an ocean mapping focused engineering research laboratory. Improving the application of autonomous survey platforms and removing personnel from traditional survey vessels will improve the efficiency of seafloor mapping, limit marine accidents, and reduce the costs of seafloor mapping, especially in remote areas. Only a small percentage of Canadian and international waters are mapped to modern standards, and improving the ability for autonomous vessels to collect this crucial data will provide benefits to Canadians and others around the world.