AI-Enabled Avian-Solar Interaction Monitoring
Photovoltaic (PV) solar energy facilities have both positive and negative impacts on birds, but field surveys do not provide sufficient data to accurately understand their nature and magnitude. Continuous observations of bird behaviors or motions around solar panels fill critical data gap and help understand causes of mortality, sources of attraction, and benefits to birds.
Argonne’s EVS and Strategic Security Sciences divisions have developed a technology to monitor avian interactions with PV solar facilities (e.g., perching, fly-through, and collisions) in a project funded by the U.S. Department of Energy Critical Minerals and Energy Innovation Office. The technology incorporates a computer vision approach supported by machine learning (ML) models. The avian monitoring technology allows continuous, cost-effective, simultaneous data collection of avian-solar interactions at multiple locations across regions to better understand potential avian impacts associated with solar energy facilities.
How Does the Technology Work?
Our AI computer vision approach uses a 4-step approach to generate avian-solar interactions data from videos.
- Detect moving objects: a background subtraction algorithm (i.e., Gaussian Mixture Model) detects moving objects and creates a sequence of image subsets known as tracks.
- Differentiate birds from other objects: an image-by-image classification model or a multiple instance learning model with the entire sequence of image subsets classifies moving objects into birds and non-bird objects.
- Detect collisions: a hybrid ML model determines if a bird collided with facility infrastructure.
- Classify non-collision activities: a Fusion bidirectional Long Short-Team Memory model classifies bird activities in 5 categories—fly over above, fly through, perch on panel, land on ground, and perch in background.
Types of Bird Activities
Our current technology extracts five bird activities: (1) perch on panel, (2) fly through between panels, (3) fly over panels, (4) land on ground, and (5) collide with panel. Extracted video clips contain more detailed behaviors than the five activities, providing solar facility managers for information about how bird may be using their facilities.