/Satellite-unet-Segmentation-master

Primary LanguagePython

unet-Satellite-Segmentation

This is a satellite remote sensing segmentation project wirtten by Keras based on U-Net.

main ideas

  1. segmented by U-Net
  2. loss-balanced_cross_entropy loss
  3. image overlap ensemble

training original data

link：https://pan.baidu.com/s/1lQZDfriZpNL0yqWY_pvfKg 提取码：ubcs

testing data

link：https://pan.baidu.com/s/1hMC3eQmJTnRotgnF_mTBwA 提取码：10me

--Satellite-unet-Segmentation-master

--data

     :training original data

--test

     :testing data

--unet

     :gen_dataset.py
     
     :unet_train.py
     
     :unet_predict.py
     
     :combind.py
    
     --lib
     
          :define_loss.py
          
     --predict
     
          :single class predict result
          
     --final_result
     
          :total class combind predict result

--unet_train

     :training data

1step: generater training data

python gen_dataset.py

2step: unet training

python unet_train.py

3step: single class predict (plant,building,water,road)

python unet_predict.py

4step: total class combind predict result

python combind.py