东京——日本数据分析初创公司Solafune计划开发并发射一颗搭载人工智能的地球观测卫星，该公司认为，在轨道上对图像进行本地处理能够降低成本，并为卫星数据开辟新的商业应用。
TOKYO -- Japanese data analytics startup Solafune plans to develop and launch an AI-powered earth observation satellite, betting that onboard processing of imagery in orbit can cut costs and open up new commercial uses for satellite data.
这家成立六年的公司利用自身的人工智能技术分析卫星图像和其他开源数据，为非洲各国政府及联合国等客户提供采矿、林业和灾害管理等领域的相关洞察。
The six-year-old company uses its own artificial intelligence to analyze satellite imagery and other open-source data, providing insights to customers, including African governments and the United Nations, in areas such as mining, forestry and disaster management.
Solafune计划将现有的AI能力应用于这颗卫星，并将与东北大学合作开发，预计在本十年末发射。合作伙伴表示，该卫星将面向亚洲及其他地区的环境保护、资源管理和灾害预防等领域。
Solafune plans to apply its existing AI capabilities to the satellite, which it will develop with Tohoku University for launch by the end of the decade. The partners said the satellite would target applications including environmental conservation, resource management and disaster prevention in Asia and elsewhere.
卫星图像通常需要传回地面进行处理，这会消耗通信带宽，并增加数据被分析前的时间和成本。
Satellite imagery is typically transmitted to the ground for processing, consuming communications bandwidth and adding time and cost before the data can be analyzed.
采用人工智能在卫星上处理数据的卫星被称为“边缘AI卫星”，这类卫星可以在将数据传回地面之前，先过滤掉低价值图像，例如被云层遮挡的图片。它们还可能能够自主识别异常或调整成像的优先级。
Satellites that use AI to process data onboard, called Edge AI satellites, can discard low-value imagery, such as pictures obscured by clouds, before transmitting data to the ground. They can also potentially identify anomalies or adjust imaging priorities autonomously.
这有助于降低运营成本，并实现近乎实时的探测与警报，可能对灾害响应、海上监控和国防监视等时间敏感型应用场景带来益处。
This can reduce operating costs and enable near-real-time detection and alerts, potentially benefiting time-sensitive use cases such as disaster response, maritime monitoring and defense surveillance.
Solafune还表示，该卫星还将能够根据电池电量、可用内存等内部因素自主调整观测计划。
Solafune also aims to have the satellite autonomously adjust its observation schedule based on internal factors including battery levels and available memory, it said.
边缘AI卫星是地球观测领域一个尚处起步但不断发展的趋势，英伟达（Nvidia）等公司将其称为“太空计算”。在日本，三菱电机正与该国航天机构JAXA共同测试此类技术。
Edge AI satellites are a nascent but growing trend in earth observation, promoted as "space computing" by companies like Nvidia. In Japan, Mitsubishi Electric is testing such capabilities with the country's space agency JAXA.
索拉富内隶属于日本一家规模虽小但数量不断增多的企业行列，这些企业致力于研发用于收集地球观测数据的卫星。随着日本政府持续加大国防投入，该领域正日益受到重视——此举部分目的便是提升太空能力，以便监测可疑船只动向及开展灾害管理工作。
Solafune joins a small but growing group of Japanese companies building satellites to collect earth observation data. The sector is attracting greater attention from the Japanese government as Tokyo increases defense spending, in part to develop its space-based capabilities for tasks such as monitoring suspicious vessel activity and disaster management.
“要监测特定区域内的变化与异常现象，通常需要部署卫星群，”索拉富内战略规划与业务发展部门负责人汤川贵臣说道。
"Detecting changes and anomalies in a given coverage area often involves using a constellation of satellites," said Takaomi Yukawa, head of strategic planning and business development at Solafune.
“不过我们认为，即便不构建耗资数百亿日元的卫星群，也能在诸多应用场景中实现相当程度的成效。从商业角度来看，这种方式也能显著提升经济效益。”
"However, we believe there may be use cases where a certain level of effectiveness can be achieved without building constellations that cost tens or hundreds of billions of yen. This would also make the economics more viable from a business perspective."