基于改进DeepLabV3+的高分辨率遥感影像建筑物提取  

Building Extraction from High-resolution Remote Sensing Images Based on Improved DeepLabV3+

作　　者：乔志勇[1] QIAO Zhiyong(Xiamen Kingtop Information Technology Co.,Ltd.,Xiamen 361008,China)

机构地区：[1]厦门精图信息技术有限公司,福建厦门361008

出　　处：《地理空间信息》2024年第12期69-73,共5页Geospatial Information

基　　金：厦门市科技计划资助项目(3502Z20231038)。

摘　　要：针对管道传统的高后果区识别方式不但工作效率低,而且很难进行定量判断的问题。提出了一种基于改进Deep⁃LabV3+的遥感影像建筑物提取算法,采用基于全卷积神经网络识别算法,用于高分辨率遥感影像语义分割,以提取出兴趣区域(AOI)的像素点,并采用形态学方法进行后处理,以获得最终的提取结果并应用于高后果区识别。实验结果表明:该文所提算法对高分辨率遥感影像中建筑物识别效果具有较大增强,且该算法也适用于油气管道高后果区的识别工作。此外还可结合其他数据以更精确地实现高后果区的识别。According to the issue of low efficiency and difficulty in quantitative judgment in traditional high consequence area identification methods for pipelines,we proposed a remote sensing image building extraction algorithm based on improved DeepLabV3+.In order to extract pixels of the area of interest(AOI),we used recognition algorithm based on a fully convolutional neural network to conduct semantic segmentation of high-resolution remote sensing images.Then,we employed morphological method to conduct post-processing,and obtained the final extraction results,which were applied to high consequence area identification.Experimental results demonstrate that the algorithm proposed in this paper significantly improves the recognition of buildings in high-resolution remote sensing images.Additionally,this algorithm is suitable for identifying high consequence areas in oil and gas pipelines.Furthermore,it can be combined with other data for a more accurate identification of high consequence areas.

关 键 词：遥感影像识别 全卷积神经网络 建筑物识别 高后果区 

分 类 号：P237[天文地球—摄影测量与遥感]