融合无人机多光谱与机器学习的洮河中游云杉人工林林窗识别

doi:10.17521/cjpe.2025.0230

• •

融合无人机多光谱与机器学习的洮河中游云杉人工林林窗识别

万姣姣, 陈国鹏, 尚嗣梁, 鲜骏仁, 杨永红, 王飞

甘肃农业大学林学院, 甘肃 730070 中国
四川农业大学环境学院, 四川 611130 中国
甘肃省白龙江林业科学研究所, 甘肃 733000 中国

收稿日期:2025-06-18 修回日期:2026-05-14

UAV Multispectral and Machine Learning-Based Forest Gap Detection in Mid-Reach Tao River Spruce Plantations

WAN Jiao-Jiao, CHEN Guo-Peng, SHANG Si-Liang, XIAN Jun-Ren, YANG Yong-Hong, WANG Fei

College of Forestry, Gansu Agricultural University 730070, China
College of Environment, Sichuan Agricultural University 611130, China
, Gansu Bailongjiang Forestry Research Institute 733000, China

Received:2025-06-18 Revised:2026-05-14
Supported by:
Supported by the National Natural Science Foundation of China(32260281); the Science and Technology Innovation Fund of Gansu Agricultural University(GAU-QDFC-2023-08); and the College Students' Innovation and Entrepreneurship Training Program of Gansu Agricultural University(202308017)

摘要/Abstract

摘要： 【目的】林窗是驱动森林更新的关键因子，其精确识别对森林生态系统结构分析与可持续管理具有重要意义。传统方法难以高效获取林窗的精细空间信息。【方法】本研究利用无人机多光谱影像和冠层高度模型(CHM)，结合面向对象影像分析(OBIA)与机器学习算法，构建了一套高精度的林窗信息提取方法，旨在揭示洮河中游云杉(Picea asperata)人工林林窗的分布特征。【主要结果】结果表明：(1) 最优影像分割参数组合为形状因子0.3、紧致度0.5、尺度26，能够有效平衡过分割与欠分割问题；(2) 在对比的四种机器学习算法中，贝叶斯分类器表现出最优的分类性能与稳定性，在测试样本上的总体精度达0.963、Kappa系数为0.913；(3) 从林窗组成结构看来，面积大于200 m2的林窗占总面积比例超过35%，表明该区域林窗形成可能主要受风倒、病虫害等干扰事件主导。【结论】研究证实了融合无人机多源数据与贝叶斯算法的OBIA方法可有效实现云杉人工林林窗的精准识别，所揭示的林窗分布格局为理解该区域人工林的干扰历史与更新动态提供了重要依据，研究成果可为退化人工林的结构优化与近自然经营提供技术参考。

关键词: 林窗, 无人机, 面向对象, 多光谱, 分割尺度, 机器学习

Abstract: Aims Forest gaps serve as a key driver of forest regeneration, and their accurate identification is of great significance for assessing forest ecosystem functions and supporting sustainable management. Traditional methods are often insufficient for efficiently acquiring detailed spatial information on forest gaps. Methods In this study, using UAV multispectral imagery and a canopy height model (CHM), combined with object-based image analysis (OBIA) and machine learning algorithms, a high-precision method for forest gap information extraction was developed. The aim was to reveal the distribution characteristics of forest gaps in a Picea asperata plantation in the middle reaches of the Taohe River. Important fingings The results showed that: (1) The optimal image segmentation parameter combination was a shape factor of 0.3, compactness of 0.5, and scale of 26, which effectively balanced over-segmentation and under-segmentation; (2) Among the four compared machine learning algorithms, the Bayesian classifier exhibited the best classification performance and stability, achieving an overall accuracy of 0.963 and a Kappa coefficient of 0.913 on the test samples; (3) In terms of gap composition, gaps larger than 200 m² accounted for more than 35% of the total gap area, suggesting that the formation of forest gaps in this region may be primarily driven by disturbance events such as windthrow, pests, and diseases.This study confirms that the OBIA approach integrating UAV multi-source data and the Bayesian algorithm can effectively achieve accurate identification of forest gaps in spruce plantations. The revealed gap distribution pattern provides an important basis for understanding the disturbance history and regeneration dynamics of plantations in this region. The findings can serve as a technical reference for structural optimization and close-to-nature management of degraded plantations.

Key words: Forest Gap, drone, Object-Oriented, Multispectral, segmentation scale, Machine learning

万姣姣, 陈国鹏, 尚嗣梁, 鲜骏仁, 杨永红, 王飞. 融合无人机多光谱与机器学习的洮河中游云杉人工林林窗识别. . DOI: 10.17521/cjpe.2025.0230

WAN Jiao-Jiao, CHEN Guo-Peng, SHANG Si-Liang, XIAN Jun-Ren, YANG Yong-Hong, WANG Fei. UAV Multispectral and Machine Learning-Based Forest Gap Detection in Mid-Reach Tao River Spruce Plantations. Chinese Journal of Plant Ecology. DOI: 10.17521/cjpe.2025.0230

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.plant-ecology.com/CN/10.17521/cjpe.2025.0230

[1]	付照琦, 胡旭, 田沁瑞, 葛艳灵, 周红娟, 吴小云, 陈立欣. 晋西黄土区2种典型森林树种夜间液流特征及对环境因子的响应[J]. 植物生态学报, 2024, 48(9): 1128-1142.
[2]	王秀英, 陈奇, 杜华礼, 张睿, 马红璐. 基于机器学习的青藏高原高寒沼泽湿地蒸散发插补研究[J]. 植物生态学报, 2023, 47(7): 912-921.
[3]	杜婷, 陈玉莲, 毕境徽, 杨玉婷, 张丽, 游成铭, 谭波, 徐振锋, 王丽霞, 刘思凝, 李晗. 林窗对川西亚高山凋落叶总酚和缩合单宁损失动态的影响[J]. 植物生态学报, 2023, 47(5): 660-671.
[4]	马和平, 王瑞红, 屈兴乐, 袁敏, 慕金勇, 李金航. 不同生境对藏东南地面生苔藓多样性和生物量的影响[J]. 植物生态学报, 2022, 46(5): 552-560.
[5]	江蓝, 魏晨思, 何中声, 朱静, 邢聪, 王雪琳, 刘金福, 沈彩霞, 施友文. 格氏栲天然林林窗植物群落功能性状的变异[J]. 植物生态学报, 2022, 46(3): 267-279.
[6]	田佳玉, 王彬, 张志明, 林露湘. 光谱多样性在植物多样性监测与评估中的应用[J]. 植物生态学报, 2022, 46(10): 1129-1150.
[7]	姜玉峰, 李晶, 信瑞瑞, 李艺. 沿海养殖池塘对红树林生态系统影响的时空变化监测与分析[J]. 植物生态学报, 2022, 46(10): 1268-1279.
[8]	刘兵兵, 魏建新, 胡天宇, 杨秋丽, 刘小强, 吴发云, 苏艳军, 郭庆华. 卫星遥感监测产品在中国森林生态系统的验证和不确定性分析——基于海量无人机激光雷达数据[J]. 植物生态学报, 2022, 46(10): 1305-1316.
[9]	郭庆华, 胡天宇, 马勤, 徐可心, 杨秋丽, 孙千惠, 李玉美, 苏艳军. 新一代遥感技术助力生态系统生态学研究[J]. 植物生态学报, 2020, 44(4): 418-435.
[10]	郭彩虹, 杨万勤, 吴福忠, 徐振锋, 岳楷, 倪祥银, 袁吉, 杨帆, 谭波. 川西亚高山森林林窗对凋落枝早期分解的影响[J]. 植物生态学报, 2018, 42(1): 28-37.
[11]	宋思梦, 张丹桔, 张健, 杨万勤, 张艳, 周扬, 李勋. 马尾松人工林林窗边缘效应对油樟化学计量特征的影响[J]. 植物生态学报, 2017, 41(10): 1081-1090.
[12]	张艳, 张丹桔, 张健, 杨万勤, 邓长春, 李建平, 李勋, 唐仕姗, 张明锦. 马尾松人工林林窗大小对两种凋落叶难降解物质含量的影响[J]. 植物生态学报, 2015, 39(8): 785-796.
[13]	许子乾, 曹林, 阮宏华, 李卫正, 蒋圣. 集成高分辨率UAV影像与激光雷达点云的亚热带森林林分特征反演[J]. 植物生态学报, 2015, 39(7): 694-703.
[14]	徐婷, 曹林, 申鑫, 佘光辉. 基于机载激光雷达与Landsat 8 OLI数据的亚热带森林生物量估算[J]. 植物生态学报, 2015, 39(4): 309-321.
[15]	李晗, 吴福忠, 杨万勤, 徐李亚, 倪祥银, 何洁, 胡义. 亚高山森林林窗对凋落物分解过程中半纤维素动态的影响[J]. 植物生态学报, 2015, 39(3): 229-238.