1,805
Views
8
CrossRef citations to date
0
Altmetric
Articles

Tobler’s First Law in GeoAI: A Spatially Explicit Deep Learning Model for Terrain Feature Detection under Weak Supervision

ORCID Icon, &
Pages 1887-1905 | Received 15 Jul 2020, Accepted 10 Nov 2020, Published online: 23 Apr 2021
 

Abstract

Recent interest in geospatial artificial intelligence (GeoAI) has fostered a wide range of applications using artificial intelligence (AI), especially deep learning for geospatial problem solving. Major challenges, however, such as a lack of training data and ignorance of spatial principles and spatial effects in AI model design remain, significantly hindering the in-depth integration of AI with geospatial research. This article reports our work in developing a cutting-edge deep learning model that enables object detection, especially of natural features, in a weakly supervised manner. Our work has made three innovative contributions: First, we present a novel method of object detection using only weak labels. This is achieved by developing a spatially explicit model according to Tobler’s first law of geography to enable weakly supervised object detection. Second, we integrate the idea of an attention map into the deep learning–based object detection pipeline and develop a multistage training strategy to further boost detection performance. Third, we have successfully applied this model for the automated detection of Mars impact craters, the inspection of which often involved tremendous manual work prior to our solution. Our model is generalizable for detecting both natural and man-made features on the surface of the Earth and other planets. This research has made a major contribution to the enrichment of the theoretical and methodological body of knowledge of GeoAI.

最近对空间人工智能(GeoAI)的兴趣,促进了人工智能(AI)的广泛应用, 特别是用于解决地理空间问题的深度学习。然而, 在人工智能模型的设计中, 存在着缺乏训练数据、忽视空间原则和空间效果等主要挑战, 严重阻碍了人工智能与地理空间研究的深入结合。本文报告了我们对一种先进的深度学习模型的研究工作, 以弱监督方式进行目标检测, 特别是自然特征的检测。我们的研究有三个创新性贡献。首先, 提出了一种仅使用弱标注的目标检测新方法。根据托布勒地理第一定律, 通过空间显式模型, 进行弱监督目标检测。其次, 将注意力映射的思路, 融入到深度学习目标检测流程中, 提出了多阶段训练策略, 进一步提高检测效果。第三, 该模型已经成功地应用于火星撞击坑的自动探测, 而之前撞击坑检测往往需要大量手工操作。我们的模型也可用于探测地球和其它行星表面的自然和人造特征。这项研究对丰富GeoAI理论和方法论作出了重大贡献。

El reciente interés en inteligencia artificial geoespacial (GeoIA) ha impulsado una amplia gama de aplicaciones que usan inteligencia artificial (IA), especialmente en lo relacionado con aprendizaje a fondo para la solución de problemas geoespaciales. Sin embargo, subsisten retos importantes, tales como la falta de datos de entrenamiento y la ignorancia de principios espaciales y de efectos espaciales en el diseño de modelos de IA, obstaculizando de manera significativa la integración a profundidad de la IA con la investigación geoespacial. En este artículo reportamos nuestro trabajo para desarrollar un modelo de punta para aprendizaje a fondo que facilite la detección de objeto, en especial de rasgos naturales, de manera débilmente supervisada. Nuestro trabajo ha aportado tres contribuciones innovadoras: Primero, introdujimos un método novedoso de detección de objeto usando solamente etiquetado débil. Esto se logra desarrollando un modelo espacialmente explícito de acuerdo con la primera ley de la geografía de Tobler para habilitar la detección de objeto débilmente supervisada. Segundo, integramos la idea de un mapa de intención en el canal de detección de objeto basado en aprendizaje a fondo, y desarrollamos una estrategia de entrenamiento a multietapa para alentar aún más el desempeño de la detección. Tercero, hemos aplicado este modelo de manera exitosa para la detección automática en los cráteres de impacto de Marte, cuya inspección implicaba a menudo un tremendo trabajo manual, antes de nuestra solución. Nuestro modelo está generalizado para detectar rasgos tanto naturales como de origen humano en la superficie de la Tierra y de otros planetas. Esta investigación ha hecho una contribución mayor al enriquecimiento del cuerpo teórico y metodológico del conocimiento de la GeoIA.

Acknowledgments

The authors sincerely thank Editor Dr. Ling Bian and the anonymous reviewers for their valuable comments.

Additional information

Funding

This work is in part supported by the National Science Foundation under Grants BCS-1853864, BCS-1455349, OIA-2033521, OIA-1936677, and OIA- 1937908.

Notes on contributors

Wenwen Li

WENWEN LI is an Associate Professor in GIScience in the School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ 85287-5302. E-mail: [email protected]. She has directed the Cyberinfrastructure and Computational Intelligence lab since 2012 at ASU. Her research interests include cyberinfrastructure, geospatial big data, machine learning, and their applications in data-intensive environmental and social sciences.

Chia-Yu Hsu

CHIA-YU HSU is an Associate Scientific Software Engineer in the School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ 85287-5302. E-mail: [email protected]. His research interests are computer vision, deep learning, and object detection.

Maosheng Hu

MAOSHENG HU is a Lecturer in the School of Geography and Information Engineering, China University of Geosciences, Wuhan, China. E-mail: [email protected]. His research interests include multiscale spatial data organization, modeling, and space–time data mining.

Reprints and Corporate Permissions

Please note: Selecting permissions does not provide access to the full text of the article, please see our help page How do I view content?

To request a reprint or corporate permissions for this article, please click on the relevant link below:

Academic Permissions

Please note: Selecting permissions does not provide access to the full text of the article, please see our help page How do I view content?

Obtain permissions instantly via Rightslink by clicking on the button below:

If you are unable to obtain permissions via Rightslink, please complete and submit this Permissions form. For more information, please visit our Permissions help page.