/rs-img-classification

Semantic segmentation of remote sensing images.

Primary LanguagePython

rs-img-classification

Semantic segmentation of remote sensing images.

Development enviroment

  1. Operation system: Ubuntu 16.04.3 LTS
  2. CUDA version: 8.0
  3. Programming language: Python 2.7
    Modules:
    tensorflow-gpu==1.4.1, scikit-image==0.13.0, tifffile==0.13.5,
    pandas==0.20.3, keras==1.2.2, h5py==2.7.0, tqdm== 4.19.4

Model

  • Reference: kaggle-dstl 3rd blog
  • Framework:
    Model structure
  • Input and output: 1024*1024*3 (in), 1000*1000*1 (out), row*column*channel

Result

left: image, middle: ground truth(label), right: prediction

  1. Airport airport238
    airport238
  2. Baresoil baresoil542
    baresoil7181
  3. Building building1675
    building6634
  4. Farmland farmland2833
    farmland4662
  5. Road road1003
    road4298
  6. Vegetation vegetation3882
    vegetation4799
  7. Water water4627
    water5375