/Semantic-Segmentation-using-U-Net

Implementing Semantic Segmentation on Satellite images using U-Net architecture

Primary LanguagePython

This project illustrates how to implement sematic segmentation with the help of U-Net architecture

1. Basic U-Net Architecture

u-net-architecture

2. Kaggle Dataset which are goingo to train our model on

Satellite images of Dubai, the UAE segmented into 6 classes The dataset consists of aerial imagery of Dubai obtained by MBRSC satellites and annotated with pixel-wise semantic segmentation in 6 classes. The total volume of the dataset is 72 images grouped into 6 larger tiles. The classes are:

  1. Building: #3C1098

  2. Land (unpaved area): #8429F6

  3. Road: #6EC1E4

  4. Vegetation: #FEDD3A

  5. Water: #E2A929

  6. Unlabeled: #9B9B9B

Sematic Seg Kaggle

4. After pathifying and One-hot encoding :

After one-hot encoding

5. Model summary which we are going to implement using TensorFlow.

Model_summary

6. Training for 30 epochs and bathx_size of 16:

Training images

7. Training and Validation loss:

Training and val loss

8. Training and validation IoU:

Training and val IoU

9. MODEL PREDICTION AFTER TRAINING USING U-Net Architecture

Model Predcition

THANK YOU FOR YOU TIME !