Advanced Intelligent Computer Vision for Remote Sensing Scene Interpretation

遥感是一门利用航空或卫星获取地球数据的科学，它提供了跨越陆地、海洋和大气的重要信息。作为世界经济中不断增长的一部分，遥感市场预计将在2017年达到近100亿美元。鉴于现在产生的大量遥感数据并在未来不断增加，迫切需要自动化算法来提高吞吐量，减少周转时间，节省资金，并执行例行监测，同时必须培训专家来理解和调查科学问题并领导未来的任务。******我的研究重点是设计和实现使用卫星数据监测地球海洋的自动算法。关键图像由世界级的加拿大卫星RADARSAT-2拍摄，在2018年发射日期之后，由三颗加拿大卫星组成的RADARSAT星座任务（RCM）拍摄。这些卫星的首要任务是监测加拿大境内和周围结冰的水域，加拿大冰局（CIS）是负责制作所需冰图的政府机构。CIS人员每年手动解读10,000个场景，提供巨大（500公里乘500公里）区域的冰类型和范围，以推断冰的厚度、范围和强度，从而促进航运和破冰的决策支持系统。******我开发了人工智能算法，可以像人眼一样有效地读取和解释图像，但我的算法提供的细节远远超过人类操作员。我的算法可以确定一个特定的像素是冰还是水，并可以进一步分解为确定冰的类型（例如，年轻的或古老的冰）。海冰类型和范围是船舶导航的关键信息，船长不想在没有这些信息的情况下进入结冰的水域——我的算法可以将这些信息传递给他们。可以监测湖冰的冻结和融化，以确定湖上的总体冰浓度，这是了解气候变化影响的关键信息。此外，我正在开发能够在雷达图像中检测石油泄漏的算法。一些船只非法向海洋倾倒舱底油，留下的油污痕迹必须通过卫星图像快速识别。我的算法会寻找这些石油泄漏，这样肇事者就能被逮捕。******其他北方国家将有兴趣购买由该算法生成的地图。我在解决这些问题的自动图像解释方面的研究也已经并将继续成功地应用于其他领域的自动化，如医学成像、3d重建、体育分析和视频分析，同时培养学生在计算机视觉方面解决这些先进的智能系统问题，并在行业中获得领导地位。

Remote sensing, the science of aerial or satellite data capture of the earth, provides crucial information across land, oceans, and atmosphere. As a growing part of the world economy, the remote sensing market is expected to reach nearly $10 billion in 2017. Given the vast amount of remote sensing data being produced now and increasing in the future, automated algorithms are urgently needed to increase throughput, reduce turnaround time, save money, and perform routine monitoring while experts must be trained to understand and investigate scientific questions and lead future missions.******My research focuses on the design and implementation of automated algorithms to monitor the earth's oceans using satellite data. The key images are produced by a world-class Canadian satellite, RADARSAT-2 and, after the launch date in 2018, by a group of three Canadian satellites called the RADARSAT Constellation Mission (RCM). The priority of these satellites is the monitoring of ice-infested waters in and around Canada and the Canadian Ice Service (CIS) is the government agency tasked with producing the required ice maps. CIS personnel manually interpret 10,000 scenes annually, providing ice typing and extents over huge (500km by 500km) regions in order to infer ice thickness, extent, and strength to facilitate the decision support systems for shipping and ice breaking.******I develop artificial intelligence algorithms that can read and interpret the imagery as effectively as the human eye, but my algorithms provide far more detail than a human operator. My algorithms can determine whether a particular pixel is ice or water, and can further break this down into determining the ice type (e.g., young or old ice). Sea ice types and extents are crucial information for ship navigation and ship captains do not want to enter ice-infested waters without this information – my algorithms can deliver this information to them. Lake ice can be monitored for freeze up and melt to establish the overall ice concentration on the lake, and this is crucial information for understanding the impact of climate change. Further, I am developing algorithms that are able to detect oil spills in the radar-based imagery. Some ships illegally dump bilge oil into the ocean and this leaves a trail of oil that must be identified by satellite imagery quickly. My algorithms will seek to find these oil spills so that the perpetrators can be apprehended.******Other northern nations would be interested in purchasing maps derived from the proposed algorithm. My research in automated image interpretation to solve these problems has also been and will continue to be successfully applied for automation to other domains such as medical imaging, 3d reconstruction, sports analytics, and video analytics, while simultaneously training students in computer vision to tackle these advanced intelligent systems problems and gain leadership roles in industry.