Young White Pine Detection Using UAV Imagery and Deep Learning Object Detection Models.

Area of Science:

• Forestry
• Remote Sensing
• Artificial Intelligence

Background:

• Forest regeneration monitoring is crucial for sustainable forest management.
• Traditional fieldwork for assessing young tree stands is labor-intensive and time-consuming.
• Integrating advanced technologies can improve the efficiency and accuracy of forest monitoring.

Purpose of the Study:

• To evaluate the effectiveness of combining unmanned aerial vehicle (UAV) imagery and deep learning (DL) for monitoring white pine (Pinus strobus) regeneration.
• To compare the performance of various DL object-detection models for identifying and assessing white pine seedlings.
• To determine the optimal UAV-DL system configuration for operational forest regeneration assessment.

Main Methods:

• Acquisition of high-resolution RGB and multispectral orthomosaics using UAV flights.
• Evaluation of 20 deep learning object-detection models within ArcGIS Pro.
• Testing model performance across different white pine size classes and densities in St. Lawrence County, NY.
• Utilizing confusion matrix analysis to determine overall accuracy.

Main Results:

• The Faster R-CNN (F-RCNN) model demonstrated superior performance, achieving an average precision of 0.88.
• The F-RCNN model achieved 91% and 90% overall accuracy in medium and high-density white pine stands, respectively.
• Model performance was consistent across both RGB and multispectral imagery types.

Conclusions:

• UAV-based deep learning systems provide an accurate and efficient method for monitoring white pine regeneration.
• This technology significantly reduces the need for extensive manual fieldwork.
• The findings support the operational implementation of UAV-DL for forest regeneration assessments.