人工智能深度学习模型在土壤属性数字制图中的应用

Application of AI Deep Learning Model to Digital Mapping of Soil Properties

为提高土壤属性数字制图预测精度,以及随着遥感环境变量数据量的增加、算力的增强和开源深度学习框架的普及,数字土壤制图正在从传统的知识驱动模型向数据驱动的人工智能深度学习模型转变。该文以土壤关键属性有机碳为例,分析归纳土壤有机碳数字制图深度学习模型的理论基础、模型结构、亟待解决的有关环境变量空间上下文信息和多模态数据整合及模型可解释性等问题,旨在促进人工智能深度学习模型在第三次全国土壤普查土壤属性制图中的应用。

In order to increase the prediction accuracy of digital mapping model of soil properties,accompanied by the increase in data quantity especially remote sensing environmental variables,the improvement of computing power,and the popularization of deep learning frameworks,the digital mapping model of soil propertiesis transitioning from traditional knowledge-driven models to data-driven artificial intelligent deep learning models.This article takes the key property of soil organic carbon as an example to analyze and summarize the theoretical basis,model structure,integration of relevant environmental variable spatial context information and multimodality data that urgently needs to be solved,and interpretability of deep learning models.The aim is to promote the application of artificial intelligent deep learning models in soil properties digital mapping of the Third National Soil Survey.