+
Executive Summary
Aeroxis Enterprises partnered with the National Oceanic and Atmospheric Administration (NOAA) on the "Digital Twin for Earth Observations (EO-DT) Using Artificial Intelligence" project. The project's objective was to enhance the accuracy and reliability of satellite images from the GOES and VIIRS satellites by detecting and correcting errors.
Our role in this project was to develop an AI-driven binary error detection system. This system was designed to automatically identify and report errors in satellite images, ensuring the data's integrity. The AI component we created achieved a remarkable accuracy rate of 99.2%, significantly improving NOAA's capability to trust and utilize their satellite data for earth observation and analysis.
Examples of Anomalies
GOES-16 - CH08
GOES-17 - CH11
GOES-17 - CH03
GOES-18 - CH13
In the News
Insights and Research
Subscribe for expert analysis, industry trends, and actionable strategies. Stay informed and inspired.
Technologies
Project Highlights
Client:
National Oceanic and Atmospheric Administration (NOAA)
Project:
Digital Twin for Earth Observations (EO-DT) Using Artificial Intelligence
Objective:
Detect and correct errors in satellite images from GOES and VIIRS satellites
Solution:
AI-driven binary error detection system
Outcome:
Achieved 99.2% accuracy in error detection.
Detection under 1 second.
Challenge
NOAA needed a system to handle and ensure the precision of satellite pictures, from the GOES and VIIRS satellites. The main issue was the errors in these images making it difficult to rely on the data, for studying Earth and its analysis. The previous way of detection looked at pixel intensity, and calculating the percentage of pixels that went over a certain threshold, which wasn’t the most reliable.
Solution
Our team developed an AI-driven binary error detection system designed to automatically identify and report errors in satellite images received by the digital twin. The focus was on ensuring high accuracy and reliability in error detection to improve the overall quality and usability of the satellite data.
Implementation
During the project, we employed advanced artificial intelligence techniques to build a binary error detector. This detector was integrated into the digital twin system to continuously monitor incoming satellite images from the GOES and VIIRS satellites. The detector's primary function was to identify and flag erroneous images automatically, ensuring that only accurate data was used for subsequent analysis and processing.
The detection system was built using PyTorch. PyTorch was used instead of Tensorflow because the library is more “pythonic” as it is newer, and had better integration with the Python interpreter for debugging purposes. Over the course of the project, we built multiple tools to acquire data, setup sample dataset, and train on the datasets.
Multiple models were created, and evaluated against real data to determine the quality of the model. Lessons learned from each trial were used to refine the training data, and retrain the model.
ResNet-18 Used to Classify GOES Images
Results
The AI error detection component I developed achieved an impressive accuracy rate of 99.2%. This high level of precision significantly enhanced NOAA's ability to maintain the integrity of their satellite data, leading to more reliable earth observation insights and decision-making.
Testing if image is “invalid” (positive state)
Accuracy: 99.2%
(04/01/23 - 04/05/23)
Lessons
Learned
One key insight from this project was the critical importance of precision in AI-driven error detection systems, especially for applications involving large volumes of high-stakes data. Another lesson was the value of collaboration between AI specialists and domain experts in developing tailored solutions that address specific operational challenges.
How can we help?
If you're facing similar challenges with data accuracy and need a cutting-edge solution, contact us to discuss how our expertise in AI and digital twin technology can help you achieve your goals.