Implementation of Deep Learning to Map Dredge Tailings from Hyperspatial Aerial Imagery

Additional Funding Sources

The project described was supported by the USDA Forest Service - Boise National Forest, as well as an Institutional Development Award (IDeA) from the National Institute of General Medical Sciences of the National Institutes of Health under Grant #P20GM103408.

Presentation Date

7-2019

Abstract

FireMAP’s research team is developing a convolutional neural network to identify archaeological sites including roads, dredge tailings, and hand-stacked tailings in support of a collaborative relationship with the Boise National Forest. A mask regional convolutional neural network (Mask RCNN) has been trained to identify the desired landmarks. This project focuses on using the Mask RCNN and the collection and labeling of hyperspatial, aerial photos of dredge tailings in order to provide a georeferenced shape feature extracted from provided orthomosaic. The Mask RCNN was able to detect numerous dredge tailings from provided testing imagery with high precision after training and implementation. Obtaining additional aerial imagery of dredge tailings would likely improve the Mask RCNN’s performance further, allowing for increased accuracy in detection.

Comments

T102

This document is currently not available here.

Share

COinS
 

Implementation of Deep Learning to Map Dredge Tailings from Hyperspatial Aerial Imagery

FireMAP’s research team is developing a convolutional neural network to identify archaeological sites including roads, dredge tailings, and hand-stacked tailings in support of a collaborative relationship with the Boise National Forest. A mask regional convolutional neural network (Mask RCNN) has been trained to identify the desired landmarks. This project focuses on using the Mask RCNN and the collection and labeling of hyperspatial, aerial photos of dredge tailings in order to provide a georeferenced shape feature extracted from provided orthomosaic. The Mask RCNN was able to detect numerous dredge tailings from provided testing imagery with high precision after training and implementation. Obtaining additional aerial imagery of dredge tailings would likely improve the Mask RCNN’s performance further, allowing for increased accuracy in detection.