结合面向对象卷积神经网络和随机森林的马尾松识别  

Pinus Massoniana Identification Based on Object-oriented Convolutional Neural Network and Random Forest

作　　者：吴瑞姣 WU Ruijiao(Fujian Geologic Surveying and Mapping Institute,Fuzhou 350011,China)

机构地区：[1]福建省地质测绘院,福建福州350011

出　　处：《测绘与空间地理信息》2024年第10期50-53,58,共5页Geomatics & Spatial Information Technology

基　　金：福建省科技厅创新资金项目(2022C0024);福建省地矿局生态地质项目(20210306)资助。

摘　　要：本研究旨在解决南方丘陵地区树种组成复杂、空间尺度较大的马尾松识别问题。采用面向对象和深度学习方法,构建了一种结合卷积神经网络和随机森林的马尾松识别模型,并在福建省建瓯市高分二号卫星影像上进行了应用。实验表明:结合面向对象卷积神经网络和随机森林的模型识别结果优于仅面向对象卷积神经网络和仅用随机森林分类算法,分类结果总体精度较好;在三调林地数据基础下对马尾松林的提取也取得了良好的效果。在此基础上,对建瓯市马尾松的空间分布进行了分析,能较好地预测马尾松林的空间分布,具有一定的实用价值。This study aims to address the challenge of identifying Masson pine in the complex tree species composition and large spatial scale of southern hilly areas.An identification model combining convolutional neural networks(CNN)and random forest(RF)was developed using object-oriented and deep learning methods and applied to high-resolution imagery from the Gaofen-2 satellite in Jianou,Fujian province.The experiment showed that the model combining CNN and RF outperformed both the model using only CNN and the model using only RF classification algorithm,with overall good classification accuracy.The extraction of Masson pine forest under the guidance of the forest inventory of the third national land survey also achieved good results.Based on this,the spatial distribution of Masson pine in Jianou city was analyzed,which can effectively predict the spatial distribution of Masson pine forests and has practical value.

关 键 词：卷积神经网络 随机森林 多尺度分割 深度学习 树种识别 

分 类 号：P237[天文地球—摄影测量与遥感]