Applying concept of (Res-net) + (U-net++) for cloud detection
#Outline of project
- reading images names from csv file
- Reading images and then mask of corresponding images
- Image augmentation & processing
- Batch generation for training, validation & testing
- Neural Network building
- Selecting optimizers and losses
- Training & validation of neural network
- Testing on test image dataset
#Dataset description
- Using dataset of LANDSAT-8 satellite known as 38-cloud dataset.
- This dataset consist of 18 images with 8400 patches for training.
- 20 images with 9201 patches are used for testing.
- Each patch of image has 4 channels – red, green, blue & nearly infrared.
#Model Features
- 36M parameters
- Model image is available as model img file
- furthur expanded image of model is also available as model img expanded file
- keras generated graph is also available as model file