Fast-track Geospatial AI Applications with Watson Data Platform and Esri ArcGIS

Combine the strengths of IBM Data Platform and Esri ArcGIS to rapidly build machine learning models with high resolution geospatial data. IBM and Esri offer a powerful suite of tools for building first class geospatial applications. In this post, you’ll learn how to leverage deep learning models with Watson Machine Learning and Watson Visual Recognition in your applications. Watson Machine Learning offers a powerful Python API built for controlling the machine learning lifecycle programmatically. You can build, save, and deploy a variety of models. In this post, we’ll save a retrained TensorFlow image recognition model based on the MobileNet SSD architecture. We’ll also make predictions using a Watson Visual Recognition classifier.

Locate Concentrated Animal Feeding Operations (CAFOs) from Satellite imagery

We deployed a Watson Machine Learning model with TensorFlow trained on satellite images pulled from ArcGIS to recognize a concentrated animal feeding operation (CAFO). The deployed model is accessible via API. Though we develop the model in Python, with Watson Machine Learning, we generate portable code snippets that you can use with cURL, JavaScript, Java, Scala, and more.

Identifying CAFOs in satellite images

The tools involved in this project are:

• Watson Studio — a powerful, scalable interactive cloud computing ecosystem. We’ll use a Python 3.5 runtime in a Small Environment.
• Watson Machine Learning — a suite of machine learning tools for data scientists, developers, and analysts of all levels. We’ll save and deploy a deep learning model with the Python API.
• Esri ArcGIS — a new Python interface provides contextual tools for mapping and spatial reasoning so you can explore data and share location-based insights. We’ll use ArcGISto visualize and work with the data for our model.
• Collaboration makes your team work smarter, together. Take advantage of best-in-class collaboration tools to track, govern, and develop as a team.

Check out the notebook here!

Identifying Broken Insulators at Georgia Power

Use Watson Visual Recognition and Esri ArcGIS to identify broken electrical insulators — structures which help to control the flow of electric current. Georgia Power uses aerial photography to capture images of their insulators. Then, we can visualize the data and pass the images to Watson Visual Recognition which infers whether a given image contains a damaged or broken insulator. Georgia Power has over 17,000 miles of power lines and 160,000 structures to analyze.

Capturing images of insulators

Check out some examples of damaged insulators.

Some of the major ways that insulators can suffer damage

The tools involved in this project are:

• Data Science Experience — a powerful, scalable interactive cloud computing ecosystem. We’ll use a Python 3.5 runtime in a Small Environment.
• Watson Visual Recognition — an IBM service built to quickly and accurately tag, classify, and train visual content using machine learning.
• Esri ArcGIS — a new python interface provides contextual tools for mapping and spatial reasoning so you can explore data and share location-based insights. We’ll use ArcGIS to visualize and work with the data for our model.
• Collaboration — Data Science Experience makes your team work smarter, together. Take advantage of best-in-class collaboration tools to track, govern, and develop as a team.

Check out the video!

Get the code on IBM Data Science Experience.

Thanks to Alberto Nieto and the team at Esri for collaborating on this project!