无人机RGB影像在高寒草地狼毒入侵监测及盖度估算中的应用  

Application of UAV RGB Image in Monitoring and Coverage Estimation of Stellera chamaejasme Invasion in Alpine Grasslands,Qinghai-Tibet Plateau

作　　者：刘咏梅[1,2] 胡念钊 龙永清 王雷[1,2] 盖星华 董幸枝[1] LIU Yongmei;HU Nianzhao;LONG Yongqing;WANG Lei;GE Xinghua;DONG Xingzhi(College of Urban and Environmental Sciences,Northwest University,Xi’an 710127,China;Shaanxi Key Laboratory of Earth Surface System and Environmental Carrying Capacity,Xi’an 710127,China)

机构地区：[1]西北大学城市与环境学院,陕西西安710127 [2]陕西省地表系统与环境承载力重点实验室,陕西西安710127

出　　处：《中国草地学报》2023年第2期1-12,共12页Chinese Journal of Grassland

基　　金：国家自然科学基金项目(41871335);科技援青合作专项(2020-QY-210)。

摘　　要：有毒植物入侵对草地生态系统和生物多样性的影响日益严重,无人机遥感为毒草防治提供了快速高效的监测方法。以青藏高原危害最严重的有毒植物狼毒为研究对象,探讨基于无人机影像的高精度狼毒遥感识别及盖度估算方法。在盛花期获取典型狼毒入侵草甸的无人机RGB正射影像,结合植被指数、色彩变换和纹理滤波方法提取狼毒识别特征,通过ReliefF-VIF/Pearson二次降维筛选出6项最优特征,基于RF、SVM和ANN三种机器学习算法构建狼毒识别模型。结果表明,与原始RGB波段相比,优选特征使狼毒识别精度有效提高4%~7%。三种分类方法的分类总精度和狼毒分类精度均大于81%,基于优选特征的RF和SVM模型的分类总精度和狼毒分类精度达到91%以上,狼毒识别效果最佳。随着统计单元的增大,利用无人机RGB影像分类结果估算狼毒盖度的精度明显下降但稳定性逐渐增加,斑块尺度50~60 cm是狼毒盖度估算的最优尺度。Toxic plant invasion has an increasingly serious impact on grassland ecosystem and biodiversity.UAV imaging provides rapid and efficient monitoring for the control of poisonous weeds.We take Stellera chamaejasme,the most serious toxic invasive weeds on Qinghai-Tibetan Plateau,as the research object,and discuss the high-precision remote sensing identification and coverage estimation method of S.chamaejasme based on UAV images.The RGB orthophotos images of typical S.chamaejasme invading meadows were acquired in full bloom,and the features of S.chamaejasme were extracted by combining vegetation index,color space transformation and spatial texture filtering,six optimal features were selected by Relief F-VIF/Pearson hierarchical dimensionality reduction,and S.chamaejasme recognition models was constructed based on Artificial Neural Network(ANN),Support Vector Machine(SVM),and Random Forest(RF).The result showed that compared the original RGB band,the optimized features can effectively improve the classification accuracy of S.chamaejasme by 4%-7%.The total classification accuracy and S.chamaejasme recognition accuracy under the three classification methods were greater than 81%,and the total classification accuracy and S.chamaejasme recognition accuracy based on the RF and SVM models using the optimized features were more that 91%,and the recognition effect of S.chamaejasme was the best.With the increase of the statistical unit,the accuracy of estimating the coverage of S.chamaejasme using the classification results based on UAV images decreased significantly,but the stability gradually increased.The patch size of 50-60 cm was the best scale for the estimation of the coverage of S.chamaejasme.

关 键 词：植被指数 色彩变换 纹理滤波 机器学习 狼毒 盖度 

分 类 号：S812.6[农业科学—草业科学]