Automatic Detection of Individual Trees from VHR Satellite Images Using Scale-Space Methods
Milad Mahour1,2, Valentyn Tolpekin1,3, Alfred Stein1
1Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, 7514 AE Enschede, The Netherlands.
Sensors (Basel, Switzerland)
|December 18, 2020
Summary
This study introduces a novel two-step method using Gaussian scale-space theory to accurately detect individual trees from very-high resolution satellite images. The refined model improves detection rates and tree size accuracy in diverse orchard settings.
Area of Science:
- Remote Sensing
- Computer Vision
- Forestry
Background:
- Accurate individual tree detection is crucial for forest management and agricultural monitoring.
- Very-high resolution (VHR) satellite imagery offers potential for detailed tree-level analysis.
- Traditional methods struggle with precise tree delineation and size estimation.
Purpose of the Study:
- To develop and validate a robust method for detecting individual trees in VHR satellite images.
- To improve the accuracy of tree crown identification, size estimation, and positional accuracy.
- To assess the method's effectiveness across different tree species and geographical locations.
Main Methods:
- Application of scale-space theory, specifically Gaussian scale space, for feature detection.
- A two-step procedure involving a Gaussian blob model and an improved tree crown model.
- Testing on VHR satellite data from vitellaria, walnut, and oil palm tree sites in Mali, Iran, and Indonesia.
Main Results:
- The refined Gaussian blob model significantly improved discrimination between true and false tree detections.
- Accurate identification of tree size and position was achieved.
- A 10-20% increase in true positive detections compared to existing methods was observed.
- Successful application demonstrated across diverse tree types and environments.
Conclusions:
- The presented two-step modeling procedure using Gaussian scale space is effective for automatic individual tree detection in VHR satellite imagery.
- This approach offers a significant improvement over traditional methods for tree mapping and inventory.
- The method shows broad applicability for various tree species and regions.
