摘要
以国产GF-2为数据源,使用自制影像标签,对长春市宽城区部分区域进行土地利用分类,提出将传统的Unet卷积神经网络模型与全连接条件随机场(CRF)结合的方法,并与Segnet网络模型进行对比分析。结果表明:Unet+CRF分类方法的Kappa系数达到0.711,F1-score达到0.795,具有较好的精度,与真实地物具有高度一致性。
In the research,the landuse in part of Kuancheng District,Changchun City is classified by using GF-2 image as data source and self-made image labels.A method of combining the traditional Unet convolutional neural network model with fully connected conditional random fields(CRF)is proposed.Compared with the Segnet model,the results show that the Kappa coefficient of the Unet+CRF classification method is 0.711,and the F1-score is 0.795,indicating good accuracy and high consistency with the real objects.
作者
师超
姜琦刚
段富治
史鹏飞
SHI Chao;JIANG Qi-gang;DUAN Fu-zhi;SHI Peng-fei(College of Geo-exploration Science and Technology,Jilin University,Changchun 130026;School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079)
出处
《世界地质》
CAS
2021年第1期146-153,共8页
World Geology
基金
中国地质调查局项目(202012000000180606)。