Remote Sensing Technology and Application, Volume. 39, Issue 5, 1115(2024)

A Multi-Class Object-Level Change Detection Method for Identifying Human Disturbance in Ecological Red Line Areas

Xiaokun GUAN, Xinsheng ZHANG, Luyang ZAN, Pan CHEN, Zhaoming WU, Yunfan XIANG, and Mingyong CAI
Author Affiliations
  • Airborne Remote Sensing Center, Aerospace Information Research Institute, Chinese Academy of Sciences,Beijing100094,China
  • show less
    Figures & Tables(13)
    Technical flowchart
    The architecture of object-level change detection based on YOLO v5
    Prediction flowchart
    Overview of the samples in Lijiang
    Comparison of change detection inference results
    Results of MobileNet v2 scene classification model
    Results of multi-class object-level change detecttion
    • Table 1. Scene classification dataset taxonomy

      View table
      View in Article

      Table 1. Scene classification dataset taxonomy

      一级人类活动二级人类活动数据数量/张
      编码类型类型
      01矿产资源开发
      矿石开采9 715
      采砂场111
      尾矿堆放地2 385
      02工业开发
      工厂10 130
      工业园871
      03能源开发
      水利水电设施1 376
      风电设施74 051
      光伏设施6 555
      火电设施3 738
      04旅游开发
      旅游用地/度假村560
      高尔夫球场696
      寺庙919
      05交通开发
      机场1 241
      港口/码头910
      铁路560
      硬化道路600
      其他道路226
      06坑塘开挖
      开挖沟渠124
      围填水域610
      07农田6 412
      08林地7 363
      09草地6 017
      10水域湿地6 835
      11荒漠175
      12居民点1 804
      13其他开发活动
      大棚9 104
      施工工地7 121
    • Table 2. Comparison of classification modelse on ImageNet-1k

      View table
      View in Article

      Table 2. Comparison of classification modelse on ImageNet-1k

      模型

      参数量

      /M

      浮点运算数

      /G

      top-1精度

      /%

      top-5精度

      /%

      ResNet1811.691.8269.989.43
      EfficientNet-b05.290.4276.7493.17
      MobileNet v23.50.3271.8690.42
    • Table 3. The architecture of MobileNet v2

      View table
      View in Article

      Table 3. The architecture of MobileNet v2

      阶段操作分辨率层数步长
      1Conv2d224×224×312
      2BottleNeck112×112×3211
      3BottleNeck112×112×1622
      4BottleNeck56×56×2432
      5BottleNeck28×28×3242
      6BottleNeck14×14×6431
      7BottleNeck14×14×9632
      8BottleNeck7×7×16011
      9Conv2d 1×17×7×32011
      10AvgPool 7×77×7×1 2801-
      11Conv2d 1×11×1×1 280-
    • Table 4. Experimental environment configuration

      View table
      View in Article

      Table 4. Experimental environment configuration

      硬件环境软件环境
      中央处理器:Intel Xeon E5-2620操作系统:

      Linux

      Ubuntu16.04

      图形处理器:NVIDIA TITAN Xp编程语言:Python3.8
      内存:128 GB深度学习框架:Pytorch1.8
    • Table 5. Comparison of change detection models

      View table
      View in Article

      Table 5. Comparison of change detection models

      模型精度/%召回率/%APIoU=.50/%APIoU=.50:.05:.95/%
      FC-Siam-conc40.6459.5020.609.70
      FC-Siam-diff49.2349.7325.9013.90
      CDNet44.7457.2629.5015.90
      Ours81.9063.6068.8057.20
    • Table 6. Comparation of scene classification models

      View table
      View in Article

      Table 6. Comparation of scene classification models

      模型

      召回率

      /%

      F1分数

      /%

      top-1精度

      /%

      top-5精度

      /%

      占用显存

      /M

      ResNet1888.6989.4290.8399.821 580
      EfficientNet-b090.7490.8991.7199.797 208
      MobileNet v290.5390.8491.8199.833 369
    Tools

    Get Citation

    Copy Citation Text

    Xiaokun GUAN, Xinsheng ZHANG, Luyang ZAN, Pan CHEN, Zhaoming WU, Yunfan XIANG, Mingyong CAI. A Multi-Class Object-Level Change Detection Method for Identifying Human Disturbance in Ecological Red Line Areas[J]. Remote Sensing Technology and Application, 2024, 39(5): 1115

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category:

    Received: Jul. 15, 2023

    Accepted: --

    Published Online: Jan. 7, 2025

    The Author Email:

    DOI:10.11873/j.issn.1004-0323.2024.5.1115

    Topics