Data Warehouse Design for Multiple Source Forest Inventory Management and Image Processing
This research developed a prototype data warehouse to integrate multi-source forestry data for long-term monitoring, management, and sustainability. The data warehouse is intended to accommodate all types of imagery from various platforms, LiDAR point clouds, survey records, and paper documents, with the capability to transform these datasets into machine learning (ML) and deep learning classification and segmentation models. In this study, we pioneered the integration of unmanned aerial vehicle (UAV) imagery and paper records, testing the merged data on the YOLOv11 model. Paper records improved ground truth, and preliminary results demonstrated notable performance improvements. This research aims to implement a data warehouse (DW) to manage data for a YOLO (You Only Look Once) model, which identifies objects in images. It does this by integrating advanced data processing pipelines. Data are also stored and easily accessible for future use, including comparing current and historical data to understand growth or declining patterns. In addition, the design is used to optimize resource usage. It also scales easily, not affecting other parts of the data warehouse when adding dimension tables or other fields to the fact table. DW performance and estimations for growing workloads are also explored in this paper.