Recipient Organization
STATE UNIV OF NEW YORK
(N/A)
SYRACUSE,NY 13210
Performing Department
Environmental Resources Engineering
Non Technical Summary
The use of Unmanned Aerial Systems (UAS) opens a new era for remote sensing and forest management, which requires accurate and regular quantification of resources. The goal of the project is to develop novel geospatial and remote sensing image processing methods using UASs imagery for application in detailed forest monitoring. Specifically, the objective of the project is twofold: forest height estimation and forest type classification using UAS technologies and advanced image processing and machine learning methods. The outcome will be a software toolbox for automatic classification of forest in 3D using UAS data. The outcome will have applications in areas such as crop classification and monitoring, forest resources management, reforestation, timber harvesting, and urban forest. This project will provide a foundation for a larger future research project focused on regional scale forest monitoring using multi-source earth observation data from air-borne and space-borne RADAR and optical sensors, UAS and LiDAR technologies.
Animal Health Component
30%
Research Effort Categories
Basic
10%
Applied
30%
Developmental
60%
Goals / Objectives
The objective of this proposed research is twofold: To develop image processing methods for (a) forest height estimation and (b) forest type classification down to stand and single tree level using data acquired by UAS (unmanned aerial susyems). The first component of the project will enhance the existing tree height estimation methods by improving a) the DSM (digital surface model) of forest areas generated by stereo/multi-view UAS imagery and b) the reconstruction of DTM (digital terrain model) of the bare ground leading to accurate forest height estimation. The second part of this research is to develop an automatic framework for classification of forest type using an object-based random forest classifier. The outcome of this project will be an automatic image processing framework of UAS data that can be used for very detailed and timely mapping and monitoring of forests in 3D.
Project Methods
This section describes the procedures and methods of the project. The proposed methodology addresses the two project parts: tree height estimation and tree type classification using stereo or multi-view multi-spectral UAS imagery.UAS Data Plan: SUNY ESF currently has an inventory of both quadcopter (3DR Solo and DFI Phantom 3 StandardTM) and fixed-wing (senseFly eBeeTM) UASs in various research labs. The Co-PIs have purchased and own UAS with permission to fly. Dr Bevilacquaa and his team have gained necessary training and certification to deploy and operate UAS over forest areas. A combination of field-based data and UAS imagery from a variety of forest types - i.e., monoculture conifer plantations of various ages, and mixed species northern hardwood stands - located at Heiberg Forest, Tully, NY and a variety of willow biomass plantations throughout central New York is available. Furthermore new image acquisition over specific site has been planned for summer 2019 and 2020. This will provide all the data required for the proposed project and allows us to focus on the objectives of the project which is to develop processing methods for such data.Tree Height Estimation: UAS can be programmed to acquire overlapping (end and side) images of target areas (e.g., forests) in multiple flight lines. This stereo data can be processed to extract a DSM of forest areas. DSM represents the highest surface recorded in the imagery, which is the top of the tree canopy in forest areas. However, to extract and map tree heights, the bare ground elevation (DTMs) are also required. Accordingly, by subtracting DTM from the DSM in the forest areas, the canopy height model (CHP) or normalized DSM (nDSM) can be extracted (Salehi et al., 2014).The image processing framework developed for this component will, first, generate DSM from stereo UAS imagery and, second, generate DTM from the DSM. Finally, the DTM is subtracted from the DSM yielding CHP (or nDSM).DSM generation using stereo UAS imageryPoint Cloud generation: Tie point selection using matching algorithms. Our improved SLIM matching algorithm (Yavari et al 2018) will be adopted and improved.DSM generation using co-linearity and co-planarity methods combined with UAS GPS/ IMU info and Ground Control PointsMosaic DSM from all stereo pairsDTM reconstruction from DSMAutomatic selection of ground point candidates using a local minimum finder algorithm.Interpolation to find DTM of non-tree areas.Reconstruct DTM for tree and non-tree areas.Tree height estimation (CHP or nDSM) Tree Type Classification: The flowchart of the proposed object-based classification approach is illustrated in Figure 3. A summary of steps involved follows:UAS data pre-processing: Ortho-rectify and mosaic UAS images using DSMImage segmentation: Multi-resolution segmentation of UAS and DSM data with trees as targets.Feature extraction: Automatic extraction of optimum features for tree type classification.Random Forest classification (RFC): Conduct RFC to classify dataAccuracy assessment: Using sample test data confusion matrix will be derived and different accuracy assessment metrics will be calculated.Classification map: shape files and tiffs formats are produced with attributes representing tree types.