基于改进YOLOv11的树种识别检测算法

Tree Species Identification and Detection Algorithm Based on an Improved YOLOv11

摘要: 树种的精准识别是森林资源调查、动态监测、科学管理及生态保护的核心环节。传统乔木林调查依赖实地勘测，成本高且受环境限制，难以满足现代林业信息化对精确、高效数据的需求，针对这个问题，提出一种改进YOLOv11的无人机影像树种识别算法，引入权重机制，通过对树种类别不均衡进行适应性校正，提升了稀有树种识别精度，并增强了模型在复杂林业遥感影像下的分类准确性与鲁棒性。本研究以东北地区常见的4种树种——兴安落叶松(Larix gmelinii)、胡桃楸(Juglans mandshurica)、榆树(Ulmus pumila)和白桦(Betula platyphylla)为对象构建影像数据集，并基于YOLOv11算法建立了树种实例分割模型，用于树冠分割与树种识别；在此过程中引入类别权重机制并优化权重配置，最终确定最优比例；经过220轮训练，模型逐渐收敛，各项性能指标稳定在较高水平，其中mAP@50为92%，Precision为91%，Recall为87%，F1 Score为88%。结果表明，引入权重机制的YOLOv11改进算法能够有效缓解类别不均衡带来的负面影响，在复杂林区场景下显著提升了对不同树种的识别精度与稳定性。

Abstract: Accurate tree species identification is a core component of forest resource surveys, dynamic monitoring, scientific management, and ecological conservation. Traditional tree surveys rely on field surveys, which are costly and constrained by environmental factors, making it difficult to meet the demands for precise and efficient data in modern forestry informatization. To address this issue, this study proposes an improved UAV image tree species identification algorithm based on YOLOv11. By introducing a weighting mechanism and performing adaptive correction for imbalanced tree species categories, the algorithm enhances the identification accuracy of rare tree species and improves the classification accuracy and robustness of the model under complex forestry remote sensing images. This study constructs an image dataset using four common tree species in Northeast China—Larix gmelinii, Juglans mandshurica, Ulmus pumila, and Betula platyphylla—and establishes a tree species instance segmentation model based on the YOLOv11 algorithm for canopy segmentation and species identification. A category weighting mechanism was introduced and optimized to determine the optimal weight distribution. After 220 training iterations, the model converged with stable performance metrics at high levels: mAP@50 of 92%, Precision of 91%, Recall of 87%, and F1 Score of 88%. Results demonstrate that the weighted YOLOv11 algorithm effectively mitigates the negative impact of class imbalance, significantly enhancing recognition accuracy and stability for diverse tree species in complex forest environments.