Automated Processing of Remote Sensing Satellite Imagery using Machine Learning

使用机器学习自动处理遥感卫星图像

• 批准号: 531267-2018
• 负责人: Jeffrey, Ian
• 金额: $ 1.82万
• 依托单位: University of Manitoba
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=654573
• 关键词: Automated; Processing; Remote; Sensing; Satellite

Sightline Innovation, Canada's longest standing Artificial Intelligence (AI) company, has built their business**around providing machine learning and computational intelligence strategies, platforms, and frameworks to**their customers. Sightline's Enterprise Infrastructure for Artificial Intelligence provides an integrated real-time**distributed AI product-as-a-service that can effectively be customized to any client's needs. Currently, Sightline**clients are interested in processing satellite images for a number of use cases including land subsidence**detection, predictive maintenance, civil infrastructure risk assessment, hydrology management, and precision**agriculture. Sightline's client base in this area is attributed to growing access to inexpensive storage, wider**access to satellite data, on-demand computing, and state of the art solutions. Sightline's solution pipeline**enables the diverse array of use cases to be addressed as a service.**As Sightline is actively working with multiple customers at various stages of development for satellite image**processing, two key client-specific technical steps must be addressed on a per-client basis: 1) construction of an**appropriate training set for supervised machine learning specific to the client's application, and 2) selection of**an appropriate neural network classifier that balances classification accuracy and robustness against**computational time/cost for the client's application. Literature reviews have indicated that machine learning**applied to these use cases is under active research and that there is a lack of comparative analysis among**available methods. The proposed research project will develop and implement a class of machine learning**algorithms capable of automating analysis of satellite images for numerous use cases. The tested classifiers will**be made available to Sightline's clients, allowing Sightline to grow their business by enabling them to continue**delivering technologically advanced solutions to client-driven problems.

Sightline Innovation 是加拿大历史最悠久的人工智能 (AI) 公司，其业务**围绕为**客户提供机器学习和计算智能策略、平台和框架而建立。 Sightline 的人工智能企业基础设施提供集成的实时**分布式人工智能产品即服务，可以根据任何客户的需求进行有效定制。目前，Sightline** 客户对处理卫星图像的多种用例感兴趣，包括地面沉降**检测、预测性维护、土木基础设施风险评估、水文管理和精准**农业。 Sightline 在这一领域的客户群归功于不断增长的廉价存储、更广泛的卫星数据访问、按需计算和最先进的解决方案。 Sightline 的解决方案管道**能够将各种用例作为服务进行处理。**由于 Sightline 正在积极与处于不同开发阶段的多个客户合作进行卫星图像**处理，因此必须执行两个关键的客户特定技术步骤根据每个客户的情况进行处理：1) 构建**适当的训练集，用于特定于客户应用的监督机器学习，以及 2) 选择**适当的神经网络分类器，以平衡分类准确性和鲁棒性与**计算之间的关系时间/成本客户的申请。文献综述表明，应用于这些用例的机器学习**正在积极研究中，并且缺乏**可用方法之间的比较分析。拟议的研究项目将开发和实施一类机器学习**算法，能够自动分析众多用例的卫星图像。经过测试的分类器将**提供给 Sightline 的客户，从而使 Sightline 能够继续**为客户驱动的问题提供技术先进的解决方案，从而发展其业务。