测绘通报 ›› 2022, Vol. 0 ›› Issue (4): 16-19,43.doi: 10.13474/j.cnki.11-2246.2022.0103

• 工程测量分会年会(2021年)优选论文 • 上一篇    下一篇

基于DeeplabV3+的建筑垃圾堆放点识别

刘小玉1, 刘扬1,2, 杜明义1,2, 张敏1, 贾竞珏1, 杨恒1   

  1. 1. 北京建筑大学测绘与城市空间信息学院, 北京 102616;
    2. 北京建筑大学北京未来城市设计高精尖创新中心, 北京 100044
  • 收稿日期:2021-07-14 修回日期:2022-02-10 出版日期:2022-04-25 发布日期:2022-04-26
  • 作者简介:刘小玉(1996-),女,硕士生,主要研究方向为遥感影像图像识别。E-mail:2108570020072@stu.bucea.edu.cn
  • 基金资助:
    国家重点研发计划(2108YFC0706003);北京建筑大学市属高校基本科研业务费专项资金(X20070);促进高校内涵发展定额项目(31081021004)

Research on construction and demolition waste stacking point identification based on DeeplabV3+

LIU Xiaoyu1, LIU Yang1,2, DU Mingyi1,2, ZHANG Min1, JIA Jingjue1, YANG Heng1   

  1. 1. School of Geomatics and Urban Spatial Informatics, Beijing University of Civil Engineering and Architecture, Beijing 102616, China;
    2. Beijing Advanced Innovation Center for Future Urban Design, Beijing University of Civil Engineering and Architecture, Beijing 100044, China
  • Received:2021-07-14 Revised:2022-02-10 Online:2022-04-25 Published:2022-04-26

摘要: 针对遥感影像中建筑垃圾非法堆放,难以快速、准确、有效地识别建筑垃圾堆放的位置、类型、面积和体量等动态信息的问题。本文基于卷积神经网络模型,对多光谱遥感影像和全色遥感影像进行NNDiffuse算法融合处理,以提高图像分辨精度,并深入分析遥感影像中建筑垃圾堆放点的特征信息,采用DeeplabV3+网络模型,采用编码器将目标所需的浅层特征和高层语义特征相结合,从图像样本数据平衡的角度,调整样本权重系数,进一步提高识别精度。试验结果表明,使用DeeplabV3+网络对建筑垃圾堆放点的识别精度达82%,有利于实现建筑垃圾动态监测与管理。

关键词: 建筑垃圾, 遥感影像, 语义分割, DeeplabV3+, 图像融合

Abstract: It is difficult to identify the location, type, area and volume of construction waste piled up illegally in remote sensing images quickly, accurately and effectively. In this paper, based on convolution model, the multi-spectral remote sensing image and panchromatic remote sensing image on its NNDiffuse pan sharpening algorithm fusion processing, it improves the precision of image resolution, in-depth analysis the characteristics of the construction waste pile up some information in remote sensing image. Use DeeplabV3+ network model and the encoder to target the shallow features and high-level semantic feature. From the perspective of image sample data balance adjust the sample weight coefficient to further improve the identification accuracy. Experimental results show that the identification accuracy of construction waste dumps using DeeplabV3+ network reach 82%, which is beneficial to realize the dynamic monitoring and management of construction waste.

Key words: construction waste, remote sensing image, semantic segmentation, DeeplabV3+, image fusion

中图分类号: