Accurate Vegetation Subtraction Tools for Disparate Data Products
- Applied Research LLC
Conventional approaches to digital terrain model (DTM) extraction use LiDAR and radar, which are expensive. Moreover, LiDAR point clouds may not be dense enough. In this research, our key objective is to use color (RGB) and near infrared (NIR) images for accurate DTM extraction. One key advantage is that it is low cost to using RGB and NIR bands. However, using RGB and NIR bands cannot differentiate different vegetation types: grass, shrub, and trees. In Phase 1, we conducted four extensive investigations using two actual datasets in the public domain. Two experiments are on vegetation and land cover classification using advanced deep learning and machine learning algorithms. Another two are on accurate DTM extraction by removing the vegetation and possibly manmade structures from the digital surface model (DSM). Our results clearly showed that the vegetation classification using only color and near infrared images is very accurate. Moreover, the DTM maps generated by using the inpainting results are also accurate and much better than the conventional bicubic interpolation algorithm. Therefore, we have successfully demonstrated the feasibility of the proposed approach. Two conference papers and one book chapter have been published. Two additional journal papers are in preparation. We also filed two invention disclosures, which we will file for patents in Phase 2.
- Research Organization:
- Applied Research LLC, Rockville, MD (United States)
- Sponsoring Organization:
- USDOE Office of Science (SC)
- DOE Contract Number:
- SC0019936
- OSTI ID:
- 1706728
- Type / Phase:
- SBIR (Phase I)
- Report Number(s):
- Final
- Country of Publication:
- United States
- Language:
- English
Similar Records
A COMPRESSED SENSING METHOD WITH ANALYTICAL RESULTS FOR LIDAR FEATURE CLASSIFICATION
Improving Land Cover Classification Using Extended Multi-Attribute Profiles (EMAP) Enhanced Color, Near Infrared, and LiDAR Data