Evaluation of Simulation Framework for Detecting the Quality of Forest Tree Stems

The advancement of harvester technology increasingly relies on automated forest analysis within machine operational ranges. However, real-world testing remains costly and time-consuming. To address this, we introduced the Tree Classification Framework (TCF), a simulation platform for the cost-effective testing of harvester technologies. TCF accelerates technology development by simulating forest environments and machine operations, leveraging machine-learning and computer vision models. TCF has four components: Synthetic Forest Creation, which generates diverse virtual forests; Point Cloud Generation, which simulates LiDAR scanning; Stem Identification and Classification, which detects and characterises tree stems; and Experimental Evaluation, which assesses algorithm performance under varying conditions. We tested TCF across ten forest scenarios with different tree densities and morphologies, using two-point cloud generation methods: fixed points per stem and LiDAR scanning at three resolutions. Performance was evaluated against ground-truth data using quantitative metrics and heatmaps. TCF bridges the gap between simulation and real-world forestry, enhancing the harvester technology by improving efficiency, accuracy, and sustainability in automated tree assessment. This paper presents a framework built from affordable, standard components for stem identification and classification. TCF enables the systematic testing of classification algorithms against known ground truth under controlled, repeatable conditions. Through diverse evaluations, the framework demonstrates its utility by providing the necessary components, representations, and procedures for reliable stem classification.