***Coastline Detection through Image Segmentation using UNet***
Authors: Donovan Dahlin, Kristine Veneles
Mentors: Dr. Dulal Kar, Jacob Hopkins
***Research Goal***
Create a Convolutional Neural Network that can detect water at shoreline in order to track coastline.
Architecture used is UNet proposed by Olaf Ronneberg
VGG16 is used as the pre-trained model for transfer learning to train model faster, provided by Keras.
***Dependencies***
tensorflow 2.1.0
keras 2.3.1
segmentation-models 1.0.1
numpy 1.20.2
matplotlib 3.3.4
python 3.7.10
opencv 3.4.2
***Environment Set-up***
Jupyter Lab via Anaconda
GPU: NVIDIA GeForce RTX 2070 SUPER
***Data Set-up***
Folder name 'data' has 6 folders:
image names corresponds with masks
- train_img
- train_mask
- test_img
- test_mask
- valid_img
- valid_mask
---- Masks were created using LabelMe ----
- LabelMe from Github creates annotated masks
source: https://github.com/wkentaro/labelme
- Annotated Masks were transformed to PNG file via GitHub source
source: https://github.com/pei223/labelme_json_to_png
***Video Files Used***
cape_cod_edit.mp4
bob_hall_pier.mp4
---- Frames were retrieved and used as the images for train, test, valid ----
-- Other images of shorelines collected from pixabay
***ShorelineDetection Folder within KristineVeneles***
ShorelineDetection Folder consists of the jupyter notebook,
data and the folder where the model is saved.
***ALERT***
This is an unfinished work. It needs a bit of work and features to be added in
order to fulfill its ultimate goal of detecting coastlines. In summary, the goal
is to use image semantic segementation to find coastline.
***ALERT***