基于大数据的遥感参数人工智能反演范式理论形成与工程技术实现  

Theory and Engineering Technology Implementation of Artificial Intelligence Retrieval Paradigm for Parameters of Remote Sensing Based on Big Data

作　　者：毛克彪[1,2,3] 袁紫晋 施建成 武胜利 胡德勇[6] 车进[2] 董立新[5] MAO KeBiao;YUAN ZiJin;SHI JianCheng;WU ShengLi;HU DeYong;CHE Jin;DONG LiXin(State Key Laboratory of Efficient Utilization of Arid and Semi-arid Arable Land in Northern China,Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081;School of Physics and Electronic-Electrical Engineering,Ningxia University,Yinchuan,750021;State Key Laboratory of Remote Sensing Science,Aerospace Information Research Institute,Chinese Academy of Science,Beijing,100094;National Space Science Center,Chinese Academy of Sciences,Beijing 100190;National Satellite Meteorological Center,Beijing 100081;College of Resource Environment&Tourism,Capital Normal University,Beijing 100048)

机构地区：[1]中国农业科学院农业资源与农业区划研究所,北方干旱半干旱耕地高效利用全国重点实验室,北京100081 [2]宁夏大学电子与电气工程学院,银川750021 [3]中国科学院空天信息创新研究院,遥感科学国家重点实验室,北京100094 [4]中国科学院国家空间科学中心,北京100190 [5]国家卫星气象中心,北京100081 [6]首都师范大学资源环境与旅游学院,北京100048

出　　处：《农业大数据学报》2023年第4期1-12,共12页Journal of Agricultural Big Data

基　　金：国家重点研发计划项目“全球粮食和病虫害监测与预警(2023YFB3906202)”。

摘　　要：为了解决人工智能(Artificial Intelligence)应用在地球物理参数反演中的“黑箱”问题,使得人工智能应用具有物理意义和可解释性及普适性,深度学习耦合物理方法和统计方法的理论和技术在各学科领域正在陆续展开。本研究通过梳理作者20余年的相关研究,在前面归纳和演绎得到人工智能地球物理参数反演范式理论和判定条件基础上,分别给出了遥感参数人工智能反演范式和判定条件。目前大家研究普遍遇到一个问题,很多人工智能参数反演理论模拟数据反演精度非常高,但实际应用反演精度不理想,因此深度学习如何耦合物理方法和统计方法成为当前亟须解决的工程与技术难题。我们以被动微波土壤水分和地表温度反演为例进行阐述,分析表明物理模型本身的精度还要很大的提升空间或者模拟数据只代表现实情况中的少部分情况。因此只利用物理模型模拟数据直接进行反演还存在很大的局限,必须补充大量高精度的多源统计观测数据。同时可以通过利用模拟数据对深度学习训练和用实际数据检验物理模型的误差。统计方法是人类最直观的描述,物理方法是对统计方法的归纳演绎总结,但真实世界的信息或能量传输是按量子形式传递,物理模型只是当前人们认识世界的最高形式,大部分模型并没有刻画好真实信息流。深度学习中的不同神经元更适合描述和表达量子信息的传输方式,以微积分量子能量信息流认识真实世界需要提高人类的思维认知方式,这才是最高模式。如何采集满足真实情况(量子信息或能量传输)的数据显得非常重要,当前可以充分利用物理逻辑推理构建物理方法和统计方法,并在范式理论和判定条件框架指导下利用大数据思维模式提高地球物理参数反演精度。通过物理逻辑推理证明输入变量能唯一确定输出变量是形成具有物理意义和�In order to solve the"black box"problem of artificial intelligence application in geophysical parameter retrieval,and make artificial intelligence applications have physical significance,interpretability,and universality,the theory and technology of deep learning coupling physical and statistical methods are gradually being developed in various disciplinary fields.This study summarizes the author's more than 20 years of relevant research,and presents the artificial intelligence inversion paradigms and judgment conditions for remote sensing parameters based on the induction and deduction of the theory and judgment conditions of artificial intelligence geophysical parameter inversion paradigms.At present,a common problem encountered in many studies is that many artificial intelligence parameter retrieval uses theoretical simulation data to achieve high retrieval analysis accuracy,but the actual application retrieval accuracy is not ideal.Therefore,deep learning how to couple physical and statistical methods has become an urgent engineering and technical challenge that needs to be addressed.We will take passive microwave soil moisture and surface temperature retrieval as an example to illustrate that the accuracy of the physical model itself still needs to be greatly improved,or the simulated data only represents a small portion of the actual situation.We believe that there are significant limitations in using only physical models to simulate data for direct retrieval,and high-precision multi-source statistical data must be supplemented.At the same time,we can also improve the physical model by directly using deep learning to simulate data training and testing with actual data to verify the gap between the physical model and the actual situation,determine the errors of the physical model,and thus improve the physical model.Statistical methods are the most intuitive description of human beings,while physical methods summarize and generalize statistical methods.However,information or energy transmission in the real wo

关 键 词：人工智能 深度学习 反演范式 物理逻辑推理 大数据 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]