Replication for: ForestSAM: A Novel Integration of DeepForest and SAM2 for Oil Palm Crown Segmentation in Aerial Imagery.

The oil palm industry in Colombia faces significant challenges in large-scale plantation management, particularly in the early detection of diseases that cause morphological changes in the crown of palm trees, severely impacting crop health and productivity. Early identification of these symptoms is typically performed manually through structural examination of the tree crowns, a process that is time-consuming, labor-intensive, and prone to human error. Although recent studies have made progress using deep learning techniques for tree counting, detection, and crown classification, the application of these methods for detailed crown segmentation remains largely unexplored. In this study, we propose ForestSAM, the first integration of DeepForest for oil palm tree detection with Segment Anything Model 2 (SAM2) for precise crown segmentation using drone and aerial imagery. The proposed method achieves a Dice coefficient (DSC) of 0.826 and a Jaccard index (IoU) of 0.712, showing its effectiveness in accurately segmenting oil palm crowns and laying the basis for timely disease detection, carbon capture metrics, and optimized resource allocation.哥伦比亚油棕产业在大规模种植园管理领域面临诸多严峻挑战，其中尤以精准识别引发棕榈树冠部形态变化的病害为甚，此类病害会严重损害作物健康与生产效能。传统上，这类症状的早期识别需通过人工对树冠开展结构检查，该过程耗时费力、劳动强度大且极易出现人为误差。尽管近期已有研究借助深度学习技术在树木计数、目标检测及树冠分类领域取得进展，但针对树冠精细化分割的相关应用仍鲜有探索。本研究提出ForestSAM方法，首次将用于油棕树检测的深度森林 (DeepForest) 与用于精准树冠分割的分割一切模型2 (Segment Anything Model 2, SAM2) 相结合，所用数据源自无人机与航空影像。该方法的戴斯系数 (Dice coefficient, DSC) 达0.826，雅卡尔指数 (Jaccard index, IoU) 达0.712，证明其可有效实现油棕树冠的精准分割，为及时开展病害检测、碳捕获量测算及资源优化配置奠定了坚实基础。