Making imagery more useful and beautiful with automated methods

Our friends at Planet Labs recently published a blog post on the manual methods that can be used to compensate for the ways that atmospheric conditions can affect the usefulness and visual consistency of satellite imagery. As the author points out, all earth observing satellites have to deal with the same challenges that come with peering through 100 kilometers of atmosphere. Haze obscures features on the ground, cloud shadows darken the ground, and atmospheric scattering and absorption effects change the apparent color of the ground. Any or all of these conditions may be present in any given scene, and they’re often not uniformly distributed.

Manually adjusting a single image is a good starting point, but that approach doesn’t scale for constellations collecting upwards of 70 terabytes of imagery per day, as the DigitalGlobe constellation does. Further, adjusting the colors of individual images destroys the precise spectral information that is required for many large-scale machine learning and automated feature extraction techniques.

Fortunately, DigitalGlobe has developed proprietary technology that solves these problems. It’s called AComp, short for atmospheric compensation. It’s a fully automated process that can keep up with our daily collection volume. It measures and compensates for the atmospheric variability from pixel to pixel within a scene, and it can eliminate haze while preserving the full spectral fidelity of the original image. It is the most accurate atmospheric compensation technology in the industry, the science and benefits of which are explained in this academic paper authored by DigitalGlobe’s Fabio Pacifici and Nathan Longbotham. For further analysis of AComp’s effectiveness, read the Master’s thesis by DigitalGlobe’s Michael J. Smith.

Watch what happens when AComp is applied to a DigitalGlobe image taken under very challenging conditions in notoriously hazy Beijing:

And it works not only with all DigitalGlobe imagery, but on other imagery sources as well, like this large-area, low-resolution scene captured by Landsat 8:

In addition to making imagery more visually appealing, AComp can also drastically improve the accuracy of the image mining and deep learning algorithms that are used in our Geospatial Big Data platform (GBDX). While an imagery analyst could study these two images for hours to pinpoint all of the instances of change, a well-trained algorithm can extract this information almost instantly.

AComp can also drastically improve the accuracy of the image mining and deep learning algorithms that are used in our Geospatial Big Data platform (GBDX). While an imagery analyst could study these two images for hours to pinpoint all of the instances of change, a well-trained algorithm can extract this information almost instantly.

When a change detection algorithm is run against the original, uncorrected images, it produces the output shown below on the left. This result incorrectly implies that there have been many changes, as indicated by the heat map. After AComp is applied, however, the result on the right provides a much more accurate assessment of where structures have been built up or torn down — dramatically reducing the amount of manual interpretation required to produce a decision-ready result.

When a change detection algorithm is run against the original, uncorrected images, it produces the output shown below on the left. This result incorrectly implies that there have been many changes, as indicated by the heat map. After AComp is applied, however, the result on the right provides a much more accurate assessment of where structures have been built up or torn down — dramatically reducing the amount of manual interpretation required to produce a decision-ready result.

AComp is just one of many technologies DigitalGlobe has developed to optimize our imagery for different customer needs. For example, our Location-Based Services customers like Mapbox require a basemap that’s optimized for global consistency. With our Basemap +Vivid and +Metro products, our sophisticated algorithms identify the best images from our massive imagery library. Then, for country-scale products, these algorithms adjust for aspects like contrast and tone to transform a patchwork quilt of image strips taken by multiple sensors during different seasons and times of day into a mosaic product with unrivaled consistency, completeness, and visual appeal.

These are just a few of the innovations that DigitalGlobe has pioneered over the last two decades that have allowed us to become a trusted partner of many of the world’s largest and most sophisticated users of satellite imagery and geospatial intelligence. With a 15-year time-lapse image library, a global ground infrastructure, and powerful machine learning and cloud computing technologies, DigitalGlobe is extending its industry leadership position and delivering even more powerful and insightful solutions for our customers.