Learning in the Deep: Quantifying change in deep-sea benthic environments using 3D image reconstructions

深度学习：使用 3D 图像重建量化深海海底环境的变化

• 批准号: 1950973
• 金额: --
• 依托单位: University of Southampton
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2017
• 资助国家: 英国
• 起止时间: 2017 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-1950973
• 关键词: Learning; Deep; Quantifying; change; deep

Seafloor hydrothermal vents, surface-exposed gas hydrate fields and live coral reefs are known to form local hotspots for biological activity. The complex interactions that take place in these ecosystems makes it difficult to estimate the distribution of biomass in these areas, which in turn limits our ability to predict the potential effects of human activities in these environments. While recent advances in underwater robotics have made it possible to routinely generate multi-hectare scale, high-resolution 3D image reconstructions of the seafloor [1], the dense populations of mega-benthos (with reports of more than 100,000 animals from different species in a single field [2]) makes manual identification of seafloor organisms a major bottleneck in the flow of information from raw in-situ observation to the quantitative estimates needed for scientific interpretation. The aim of this project is to develop computational methods to assist the interpretation of georeferenced seafloor visual imagery, and so enable quantitative studies of the temporal changes the distribution of mega-benthic organisms. This fits within the broader scope of improving our ability to monitor and understand the response of seafloor ecosystems to human influence.

众所周知，海底热液喷口、地表暴露的天然气水合物田和活珊瑚礁是生物活动的局部热点。在这些生态系统中发生的复杂的相互作用使得很难估计这些地区的生物量分布，这反过来又限制了我们预测人类活动在这些环境中潜在影响的能力。尽管水下机器人技术的最新进展使常规生成多公顷规模的海底高分辨率3D图像重建成为可能，但大型底栖动物的密集种群（据报道，在一个领域中有超过10万只来自不同物种的动物）使得人工识别海底生物成为从原始现场观察到科学解释所需的定量估计的信息流的主要瓶颈。该项目的目的是发展计算方法，以协助解释地理参考海底视觉图像，从而能够定量研究巨型底栖生物分布的时间变化。这符合提高我们监测和了解海底生态系统对人类影响的反应的能力这一更广泛的范围。