基于深度学习的地表覆盖变化检测方法研究

Research on the Change Detection Method of Land Cover Based on Deep Learning

研究地表覆盖的自动变化检测对于常态化地理国情监测具有重要意义。结合全卷积神经网络(FCN)和孪生(Siam)神经网络,设计了一种全卷积孪生网络模型(FCSCN),并通过构建城市地表覆盖变化样本库、模型训练和测试、精度评价,得到了适用于沈阳城市地表覆盖变化检测的深度学习模型。以2022年度地理国情监测项目局部区域为试点,开展了实践探索,结果发现该方法可以提高作业效率,对于高频次、全覆盖的地理国情监测具有一定的实践参考意义。

The research on automatic change detection of land cover is of great significance for the normalization of geographical national conditions monitoring.Combined with the full convolution neural network(FCN)and the twin(Siam)neural network,a full convolution twin network model(FCSCN)is designed.Through the construction of the sample database of urban land cover change,model training and testing,and precision evaluation,a deep learning model suitable for the detection of urban land cover change in Shenyang is obtained.By constructing a sample database of urban land cover change,then conducting model training and testing,and finally carrying out precision evaluation,a suitable deep learning model for Shenyang urban land cover change detection is obtained.Taking the local area of the 2022 geographical national conditions monitoring project as the pilot,the practical exploration was carried out.The results showed that this method can improve work efficiency,and has certain practical reference for the high-frequency and full-coverage geographical national conditions monitoring.