Q-SAT-GEN - Hybrid generative modelling for satellite image denoising and infilling

Q-SAT-GEN - 卫星图像去噪和填充的混合生成模型

基本信息

批准号：
10086555
负责人：
金额：
$ 13.58万
依托单位：
ORCA COMPUTING LTD.
依托单位国家：
英国
项目类别：
Small Business Research Initiative
财政年份：
2023
资助国家：
英国
起止时间：
2023 至无数据
项目状态：
已结题

来源：
https://gtr.ukri.org/projects?ref=10086555
关键词：
SAT GEN Hybrid generative modelling

项目摘要

Satellite data provides opportunities for governments and businesses around the world in scientific, socio-economic management and commercial applications. Access to timely, reliable, and actionable information is increasingly critical to a growing number of organizations and decision makers who rely on earth observation data.Gaining useful insights from satellite images can be difficult due to sources of noise such as sensor processing errors, clouds or atmospheric perturbations, the low spatial resolution of typical satellite images, and revisit times on the order of a few hours to days. Holes or gaps in of missing pixel information introduces uncertainty and affects decision making capabilities of stakeholders who rely on accurate information for near-real-time monitoring and inference purposes.Traditionally, classical generative AI methods such as GANs or diffusion models have been used to address these issues. However, these methods require significant computational resources and suffer from issues such as mode collapse in GANs.ORCA Computing will deliver a hybrid quantum/classical generative algorithm for satellite image processing both to reduce the computational resources required to train models and to improve their performance. This algorithm uses ORCA's PT-Series quantum processor and unique software stack. This solution will be beneficial for satellite monitoring purposes in areas such as climate and weather monitoring, defence, the environment and agriculture.

卫星数据为世界各地的政府和企业提供了科学，社会经济管理和商业应用的机会。对于越来越多的组织和依赖地球观察数据的组织和决策者而言，访问及时，可靠和可行的信息越来越重要。由于噪音的来源，诸如传感器处理错误，云或大气扰动之类的噪音源，从卫星图像中获得有用的见解可能很困难，这是典型的卫星图像的低空间分辨率，以及几个小时的时间订购时间。缺少像素信息的孔或空白引入了不确定性，并影响利益相关者的决策能力，这些利益相关者依靠准确的信息来实现近实时监测和推理目的。在传统上，使用经典的生成AI方法，例如gans或扩散模型来解决这些问题。但是，这些方法需要大量的计算资源，并且遭受诸如gans.orca计算中模式崩溃之类的问题，将为卫星图像处理提供一种混合量子/经典生成算法，以减少培训模型所需的计算资源并提高其性能。该算法使用Orca的PT系列量子处理器和唯一的软件堆栈。该解决方案将有助于在气候监测，防御，环境和农业等领域的卫星监测目的。