基于深度学习的建筑物识别及占用耕地建房自动化监测--以湖南省长沙市X村为例

Building recognition and automated monitoring of occupying farmland for construction based on deep learning:Taking X Village in Changsha City,Hunan Province as an example

针对农村普遍存在的占用耕地建房现象,基于深度学习和图像分析提出了一种自动化监测方法,通过对高分辨率遥感影像数据的预处理,构建基于卷积神经网络的自动化监测模型,有效判定目标影像中的每个像元格是否占用耕地建房。以湖南省长沙市X村为例,横向比较U-Net、SegNet、DeepLabV3p模型的识别能力。结果表明,当学习率为0.01、批大小为2、迭代次数为100次时,U-Net模型对建筑物的识别结果最佳;该模型共发现66宗潜在占用耕地建房案例,识别结果准确率高且耗时短;该模型充分运用了现代信息技术及方法,可在一定程度提高土地执法监察的工作效率、节省工作时间及资源。

In response to the common phenomenon of occupying farmland for building houses in rural areas,an automated monitor-ing method based on deep learning and image analysis was proposed.By preprocessing high-resolution remote sensing image data,an automated monitoring model based on convolutional neural networks was constructed to effectively determine whether each pixel cell in the target image occupied farmland for building houses.Taking X Village in Changsha City,Hunan Province as an example,horizontally compared the recognition capabilities of U-Net,SegNet,and DeepLabV3p models were.The results showed that when the learning rate was 0.01,the batch size was 2,and the number of iterations was 100,the U-Net model had the best recognition results for buildings;the model found a total of 66 cases of potential occupation of farmland for building houses,with high recognition accura-cy and less time consumption;this model fully utilized modern information technology and methods,which could improve the efficien-cy of land law enforcement and supervision to a certain extent and save work time and resources.