Deep learning enables us to identify patterns and insights from data that were once difficult to detect with traditional methods. The Esri education team recently conducted a webinar on the topic of “Unlocking the Power of Deep Learning Applications Using ArcGIS. This blog provides links to the resources.
Multi levels approach to Geospatial Deep Learning in ArcGIS
ArcGIS has tools to help with every step of the deep learning workflow including data preparation and exploratory data analysis, training deep learning models, deploying them for inferencing, and disseminating results using web layers and maps.
Geospatial deep learning capabilities are available at multiple levels to accommodate a wide range of audiences.
The core of all geospatial deep learning in ArcGIS lies within the arcgis.learn module of Python API, at the very centermost circle. This module serves as a foundation for geospatial AI across imagery, 3D, tabular, text, and timeseries data.
While the Python API is designed for data scientists, scripters, and advanced users, the intermediate circle features GeoAI tools and apps that are primarily designed for geospatial analysts, or image analysts, which are user centered experiences.
Finally, the outermost circle is built around ready-to-use pretrained models, aimed at making geospatial deep learning more accessible to general users.
For the deployment, deep learning in ArcGIS can be performed in ArcGIS Online, ArcGIS Pro, ArcGIS Enterprise, and ArcGIS Notebooks, providing options for users with different proficiency levels.
Performing Deep Learning using Ready-to-use pretrained models
We understand that the most time-consuming part of deep learning is collecting samples, labeling, and training the model. To make the application of deep learning more accessible, particularly for new users or those seeking expedited outcomes, Esri has made available a vast collection of geospatial deep learning models on the ArcGIS Living Atlas of the World. As of the date of this blog, there are currently 61 pretrained deep learning models available, with more expected in the future.
Through this storymap, you can explore a range of examples displaying how pretrained models have been implemented to extract features of interest including building footprints, parking lots, land cover classification, waterbody extraction, and more.
The list of pretrained models are available to explore and download from with the ArcGIS Living Atlas of the World.
Every pretrained model includes metadata and guide instructions. They can be utilized in ArcGIS Online Image Analysis tools, as well as in ArcGIS Pro and ArcGIS Enterprise.
For more information on these pre-built geospatial deep learning models, please refer to the following resources:
It is also worth mentioning that the list includes the latest Segment Anything Model (SAM) from Meta:
Perform End-to-End Deep Learning in ArcGIS Pro
ArcGIS Pro is a powerful desktop application with the capability to perform deep learning analysis. ArcGIS Pro Image Analyst extension comes with a series of geoprocessing tools to perform end-to-end deep learning analysis from sample collection, export training data, to model training and inference. Once a model has been successfully trained and evaluated, it can be applied for various applications such as object detection, terrain analysis, and image classification. The results can be published to ArcGIS Online for sharing.
One important note is that deep learning capability is not inherently included in ArcGIS Pro. To gain access to this capability, it is necessary to first download the deep learning libraries installers for ArcGIS and install it into ArcGIS Pro. The installer includes a broad collection of deep learning components such as PyTorch, TensorFlow, Fast.ai and scikit-learn.
Check out the following resources to learn more about performing deep learning analysis using tools in ArcGIS Pro:
Performing deep learning analysis using Python in ArcGIS
The arcgis.learn is the core Python library of all geospatial deep learning within ArcGIS that provides deep learning capabilities within the ArcGIS desktop and web environments. This library simplifies the process of training diverse deep learning models, enabling users to not only train these models but also prepare data, manage that data, and infer at scale. The library is seamlessly integrated with ArcGIS and open science Python libraries.
Check out the following resources to learn more about performing deep learning analysis utilizing arcgis.learn python library:
There are several best practices to follow when performing geospatial deep learning which include deciding the ideal training dataset and size, knowing if the model is good, imagery and hardware considerations, etc. The following blogs and documentation provide some resources on this topic:
Overall, ArcGIS has powerful deep learning capabilities that are available at multiple levels to accommodate a wide range of users. Contact me at email@example.com if you need more information or would like to discuss your geospatial deep learning project.