Development of a Test Method for Evaluating Staining on Structures as an Indicator of Bat Presence

About the report:

To protect threatened and endangered bat species in Virginia, VDOT conducts bat inventory surveys on bridges, structures, and dwellings to identify whether bats are using the infrastructure. Observing guano droppings and staining is a common indicator of bat presence, but it can be difficult to verify whether certain stains originated from bats or other sources such as water seeps, asphalt leaching, or other structural deterioration mechanisms. While bat indicators are hard to distinguish by humans without training, from a computer vision perspective they show different features and patterns that, when coupled with expert opinion, can be used for automated detection of bat presence. To facilitate the detection of bat presence in structures maintained by VDOT and streamline bat surveys, this project leverages recent advances in visual recognition using deep learning to develop an image classification system that Identifies bat indicators. To overcome the shortage of data needed to train a deep learning model, the bat identification task used the parameters previously trained on large-scale image data sets to transfer the learned feature representation. Using a pool of data collected through VDOT, a visual recognition model was developed and achieved 92.0% accuracy during testing. To facilitate the application of the developed model, a prototype web application was created to allow users to upload images and receive classification results from the developed model.  The study recommends that VTRC staff: (1) work with the VDOT Information Technology Division (ITD) to host the image classification web app and make it accessible for use by VDOT bridge inspectors and environmental staff, and (2) conduct a pilot evaluation of the web app in several VDOT Districts before widespread deployment of the web app. 

Authors