Vegetation Risk Analysis on Utility infrastructure
In this project, we are trying to determine the most important factors that leads to power outage due to tree failure during extreme weather conditions like hurricane, snowstorms, thunderstorms etc. Variables are categorized into different categories like environmental variable, geographical variables, management related variables etc. Also, accuracy of various machine learning approaches like Random Forest, XGBoost will be compared for classification Device exposure zone (DEZ) based on the risk level to tree failure .
Unhealthy Tree Crown Detection using Deep Learning Approach
Tree failure is the major source of power outage and utility companies spend millions of dollar every year in maintenance. Dead, decaying and defoliated tree are more prone to falling in extreme weather conditions. Hence, there is strong need of identification and management of such trees. We are trying use state-of-art methods to detect unhealthy tree crowns along the roadside that can be possible threat to the utility infrastructure. A Very High Resolution Imagery (VHRI) like NAIP, planetary image products and Lidar derived products will be used in this analysis. U-net and MaskRCNN deep learning algorithms will be used in this process.
Dashcam videos for roadside vegetation analysis
Airborne remote sensing images only provide the aerial view of the object that lacks providing the existing real condition on the ground. The vertical profile of the objects in the earth surface may differ than the image captured from above. Thus, there is need of the ground level dataset that provides the vertical profile of the forest trees so that we can more accurately describe the forest characteristics. Ground-level images/videos are the potential datasets that can fulfill this demand. We want to use Dashcam video data for the roadside vegetation monitoring. These videos have higher spatial resolution than remote sensing data and are provide more information on the entire tree that is comparable to the expert view while assessing the tree manually. Since the data acquisition can be done in a more flexible way and are not expensive, they can be important source of information to open new perspective in the forest characterization and management. Deep learning approach will be used to produce real-time monitoring of roadside vegetation.