SBIR Phase I: High-Resolution Image Segmentation for Natural Resource Management

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project is to produce currently unavailable high-resolution vegetation maps and analyses that enable stakeholders (i.e., government agencies, academic researchers, land managers, non-governmental organizations, and private companies) to rapidly assess the health of ecosystems that are threatened by human development and environmental change. Producing this information will result in better management of natural lands and their associated services and goods, globally valued at $125 trillion, such as buffering current and future infrastructure from natural disasters (i.e., managing wetlands that dampen storm surge) and improving human health and well being outcomes (i.e., disease prevention and livelihood security, respectively). Compared to traditional ground surveying methods, this project will revolutionize the ecosystem health evaluation and management process by reducing work hours by approximately 50-90% and project costs by approximately 40%-70%. This SBIR Phase I project will demonstrate the feasibility to expand the accessibility and scalability of machine learning image segmentation to the fields of natural resource management, environmental conservation, and ecological research. The technical innovation of this project is a replicable machine learning model for vegetation analysis and ecosystem assessment that will expedite the ability to process and classify aerial imagery into individual species layers that can be used to assess vegetation composition and dynamic changes in species populations most influenced by human activity and climate change on a global scale. While there are examples of employing machine learning image segmentation in these fields, they are specific to regions or species and are incapable of scaling across diverse ecosystems and image resolution levels. The goal of the project is to create a machine learning model that can quickly and accurately delineate vegetation types from aerial imagery across diverse sets of data. This goal will be achieved following a development strategy of model exploration, data collection/annotation, model refinement, testing and evaluation, and model deployment. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.