Banner Animation

深海海山数字化智能系统 Intelligent Seamount Digital Systems

中国大洋事务管理局联合海洋二所、之江实验室、国家深海基地管理中心、厦门大学共同研发,致力于推动深海科技进步,服务全球海洋可持续发展。

探索系统

Jointly developed by the China Ocean Affairs Administration, the Second Institute of Oceanography, Zhejiang Lab, National Deep Sea Center and Xiamen University dedicated to advancing deep-ocean technology and serving global marine sustainability.

Explore Systems

系统总览 System Overview

ISDS系统旨在利用前沿的人工智能模型和高质量数据集,为深海海山生态系统提供前所未有的数字化体验。通过模拟海山环境的生物多样性与地质特征,我们致力于推动深海科学的边界,支持可持续的海洋资源管理。

The ISDS is designed to provide an unprecedented digital experience of deep-ocean seamount ecosystems using cutting-edge artificial intelligence models and high-quality datasets. By simulating the biodiversity and geological characteristics of seamount environments, we are dedicated to advancing the frontiers of deep-ocean science and supporting sustainable management of marine resources.

System Overview Image

核心能力:

 

  • 深海生物物种识别: 快速准确地检测和分类深海生物,揭示海山独特的生命图谱。
  •  

  • 深海地质特征识别: 深度解析海底图像,智能推断热液、沉积物、岩石等地质构造。

 

我们的目标是提供一个强大的科研平台,服务于海洋生物环境科研工作人员,共同守护这片深海的未知领域。

The system integrates two core capabilities:

 

  • Identification of Biological Species: Rapid and accurate detection/classification of deep-ocean organisms, unveiling unique biodiversity profiles of seamounts.
  •  

  • Identification of Geological Features: Advanced analysis of seabed imagery with intelligent inference of hydrothermal vents, sediments, and rocks.

 

Our goal is delivering a robust scientific research platform for marine biologists and environmental researchers, collectively safeguarding the uncharted realms of the deep ocean.

核心智能模型 ISDS Models

探索驱动OneOcean平台的先进人工智能模型,这些模型专为深海环境的独特挑战而设计。

Explore the advanced artificial intelligence models powering the OneOcean platform, designed for the unique challenges of the deep-ocean environment.

🔬 深海生物物种检测模型 Biological Detection Model for Deep Ocean

基于先进的YOLOv11架构(AGPL-3.0 License),我们的深海生物检测模型专为应对深海图像中生物种类稀有、形态复杂及目标尺寸多变的挑战而设计。

核心特性:

  • 高精度识别106个层级标签的深海生物形态种。
  • 覆盖珊瑚、棘皮动物、甲壳类、鱼类等多种典型生物。
  • 有效处理模糊类别。
  • 提供在线高性能版与开源本地部署版。

Built upon the advanced YOLOv11 architecture (AGPL-3.0 License), our deep-ocean seamount biological detection model is specifically designed to address challenges including rare species, complex morphological variations, and variable size targets in deep-ocean imagery.

Core Features:

  • High-precision recognition of 106 taxonomically labeled deep-ocean biological morphological species.
  • Covering spanning corals, echinoderms, crustaceans, fish, and others.
  • Effective handling of ambiguous categories.
  • Providing an online high-performance test version (currently available) and an open source on-premise deployment version (future release includes model weights & source code).
Biological Detection Model Example
Biological Detection Model Example

🌍 深海地质特征识别模型 Geological Feature Identification Model for Deep Ocean

本试用版本严禁任何形式的公开分享、公开分发或公开传播,违规者将失去测试资格并承担法律责任。采用Qwen2.5-VL-7B-Instruct(Apache 2.0 License)作为基础模型,并结合LoRA参数高效微调技术,我们的地质推理模型能够对深海图像进行深入理解和结构化描述。

模型能力:

  • 强大的图像理解、结构化输出和视觉推理能力。
  • 输出包含图像整体描述及地质类型分类(如热液口、沉积物、岩石等)。
  • 通过在大量图文对数据上进行LoRA微调,适应特定地质分析任务。

Leveraging Qwen2.5-VL-7B-Instruct (Apache 2.0 License) as the foundational model integrated with LoRA parameter-efficient fine-tuning technology, our geological reasoning model achieves in-depth comprehension and structured characterization of deep-ocean imagery.

Model Capabilities:

  • Advanced image understanding, structured data output, and visual reasoning capabilities.
  • Delivers comprehensive scene descriptions and taxonomic classifications of geological formations (such as hydrothermal vents, sediments, rocks, etc.).
  • Task oriented optimization through LoRA fine-tuning to adapt to specific geological analyses workflows.

 

Deep Sea Geological Image Example

开放数据集资源 Dataset Resources

我们致力于提供高质量、精细标注的深海数据集,以推动相关领域的研究与发展。所有数据集均经过严格的质量控制和标准化处理,以确保其科研价值和易用性。

We are committed to providing datasets of high-resolution, precisely annotated deep-ocean image to promote research in marine geoscience and benthic ecology. All datasets are subjected to rigorous quality control and standardization processes to ensure the scientific value and usability of the research.

📊 深海生物标注数据集 Deep Ocean Benthic Organisms Annotation Dataset

为支撑生物检测模型,我们构建了全面的深海生物数据集。该数据集整合了来自多个公开来源及专业采集的图像数据,并进行了精细化标注。

数据集亮点:

  • 包含多种来源的生物图片。
  • 目标分类颗粒度细化至科及以下级别。
  • 数据经过标准化和质量控制,确保专业性与科普性。
  • 下图为已整理的部分底栖生物数据集示例(目录结构)。

We have created a comprehensive deep-ocean biological image dataset that integrates image data from multiple public sources and professional collections by the Second Institute of Oceanography. The dataset has undergone expert-validated annotations and careful reviews.

Dataset highlights:

  • Containing biological imagery from various sources.
  • Taxonomic classifications are refined to family-level granularity.
  • All data undergo standardized processing and rigorous quality control to ensure scientific validity.
  • The accompanying figure demonstrates the curated benthic biological dataset architecture.
Biological Dataset Structure Example
图像数据Image data 精细标注Detailed annotation 多物种Multi-species

📋 地质特征标注数据集 Deep Ocean Geological Feature Dataset

地质特征标注数据集包含了约13000个图文对,这些数据通过对公开数据集中的图片进行详细描述而构建。

数据集构成:

  • 热液口图像及描述 (约1896对)
  • 沉积物图像及描述 (约3902对)
  • 岩石图像及描述 (约5836对)
  • 其他地质特征 (约1366对)

The geologically annotated dataset comprises approximately 13,000 image-text pairs, systematically constructed through detailed and targeted descriptions of images sourced from multiple public sources. Ensuring findability, accessibility, interoperability, and reusability for research.

Dataset composition:

Contains geological imagery from various sources in public domain.

  • Hydrothermal vent: ~1,900
  • Sediment: ~3,900
  • Rocks: ~5,850
  • Others: ~1,400
图文对Structured description 结构化描述Geological reasoning

说明:模型正式发布时将公开训练数据来源。如您认为公开来源中的某些数据侵犯了您的版权并要求移除,请及时联系我们处理。 Notice: Training data sources will be disclosed upon the model's official release. If you believe certain data in the disclosed sources infringe upon your copyright and request their removal, please contact us promptly for resolution.

寻求合作 Cooperation

我们欢迎全世界科研机构、教育者和行业伙伴共同探索深海的奥秘。了解如何访问我们的模型、数据集,或与我们开展合作研究。

We welcome research institutions, educators, and industry partners worldwide to explore the mysteries of the deep ocean. Learn how to access our models and datasets, and explore joint research with us.

邮件联系我们 Contact us via email 我要反馈问题 Feedback
或直接发送邮件至: Or send email directly to:
support.oceanus@zhejianglab.org

在线模型与本地部署版本将通过指定渠道发布,敬请关注。 The online models and local deployment version will be released through designated channels. Please wait.

💬