Appendix F: GSFC Computational and Information Sciences and Technology Office (CISTO): Environmental Feature Recognition: A Hybrid Machine Learning and Statistical Method for Environmental Feature Recognition

The recent advances in computational technology have enabled the machine learning algorithms practical on a wide range of large and complex problems. One such field is environmental studies in which high-resolution remote sensing images and smart information has contributed to large datasets. Although such data bring a lot of opportunities, it also requires developing advanced computational methods that can perform processing, retrieval, analysis, storage and also extract more meaningful patterns. Machine learning can address some of the above issues by making them automatic and extract patterns that the available methods cannot identify. Such applications, however, requires an appropriate implementation of the machine learning algorithms to produce more predictive method without human’s intervention. One of the pillars of machine learning algorithms is accessing big data and accurate responses. Providing such data for a wide range of variability might not be possible to aid a successful performance. Providing such data for large-scale environmental problems that require days of interpretation is difficult. Furthermore, even the machine learning techniques are more sensitive to the input data as only simple image operators (e.g. rotation, scaling, noise, flipping, and cropping can enrich the dataset. In this research, we aim to diversify the input data by incorporating a stochastic method that can build a more diverse set of data (i.e. input and output) using the already available data. This method aims to increase the accuracy of predictions and feature recognition in environmental data/images. The proposed method can take the manually generated data and produce a much larger dataset that can be used in deep learning algorithms. Our final algorithm will allow NASA to conduct more accurate classification and feature extraction for the satellite images and, thus, reduce the costs and time. Furthermore, one of the preliminary results of this project is creating a very large and comprehensive training data at small resolution (e.g. sub 5m spatial) to identify forest patches using AI. This approach will pave the way for using the state-of-the-art methods developed in machine learning to Earth science and environmental problems in which large datasets with accurate responses are not easily available. The final results can also be used in environmental assessment and monitoring, city planning, and land cover change detection.

This is a mature technology (TRL 7+) — the realistic path in is usually NASA's Technology Transfer Program: licensing an existing NASA patent, or a Space Act Agreement to use NASA facilities/expertise directly. NASA also runs a startup licensing program with no upfront fee for companies formed to commercialize a specific NASA technology.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.