My Daily Activities

Monday, 10 June

Fistly, i explore some resources and data

Create a first-try with Openstreetmap Coastline in https://edx.tavenel.fr/coastalerosion/index2.html

Tuesday, 11 June

I've done a presentation of my works

Define the topic: Interactive Coastal Erosion Map with AI Predictions

I improve my first-try

Wednesday, 12 June

I've done a first try with Flask project with NASA API based on NASA Earth Imagery API

Began developing the website https://erosion-ai.tavenel.fr/ to showcase my work at the Coastal Engineering & Ocean Climate (CEOC) center. This includes my research on the Interactive Coastal Erosion Map with AI Predictions topic.

Discover a new tool : https://livingatlas.arcgis.com/wayback/

Correct errors on the tool and made a pull request on an opensource project: https://github.com/Esri/wayback

Create a local implementation of the ESRI | World Imagery Wayback

Thursday, 13 June

New prototype idea

The idea is to use World Imagery Wayback to create a custom basemap with ESRI and ArcGIS, and then combine that map with Mapbox.

Using Machine Learning algorithms with historical previews, I could identify the coastline.

With historical coastlines, I could create local animations of coastal erosion.

For example, displaying the coastline in different colors to show the erosion over time, with a 2024 map to illustrate coastal changes.

The data would be stored in JSON format with corresponding years.

But encounter some issues to proceed like that (cannot download pictures)

Find an interesting article https://www.sciencedirect.com/science/article/pii/S0034425722001584

Find a useful database: https://openmldata.ukho.gov.uk/

Friday, 14 June

Fit a model in order to find where the coastline is on a picture

You could check the training and validation metrics in this link

Create the github repository with python code in Repository github coastline-ML

Monday, 17 June

Add the coastline openstreetmap project on erosion-ai [Ireland implemented]

Check the coastline image project on erosion-ai

Add the satellite-imagery interface on gitub

Code project on github

Try to give to coastline-ML the following picture

And i obtain the following result

And me who just want to obtain the coastline automatically from google earth studio , what could i do?

The coastline-ML needs to have a tif image with a label picture where the coastline is identified in order to generate a coastline prediction

So i want to find a way to generate automatically the label picture if i want to use coastline-ML to predict the coastline.

In addition i would like to obtain the coastline but with geoJson data like that

Find interesting resources to explore

Interpreting a Semantic Segmentation Model for Coastline Detection interpreting-coastline-unet SWED-edge-detection

Tuesday, 18 June

Try to implement a Semantic Segmentation Model for Coastline Detection by transform original image to label

Create a new github code project by forking the existing code project https://github.com/conorosully/SWED-edge-detection

Add my work on that project here :click here

Wednesday, 19 June

Improve the powerpoint presentation

Implement France in the openstreetmap coastline interactive map

Thursday, 20 June

Try to train a unet model in order to automatically generate the label image with accuracy apprixmatively 0.46

Check training metrics here!

Create the coastal sentinel repository on github

Repository code project

Improve the mapbox appearance for interactive maps

Friday, 21 June

Create the first prototype : Check the first prototype with Dounam bay

Monday, 24 June

I create a word document to write my work for the last two weeks.

Looking for new labelling methods

Tuesday, 25 June

Implement new models with new labeling methods

Improve the repository CoastalSentinel github project : Check here!

Wednesday, 26 June

I improve the coastal sentinel Jupyter notebook

Prepare the tomorrow presentation

Thursday, 27 June

Held a presentation of my works for the 3 weeks to CEOC group online meeting

The pdf presentation file

Friday, 28 June

Prepare a new dataset for classification in Machine Learning to make the difference between mainland and water

Here is the dataset
Here is the dataset (Updated), still in progress

Next, try to combine the classification dataset with an u-net model to make autolabelling for the dataset and find the coastline automatically.

Monday, 1 July

Improve the new dataset (~1000 images)

Implement my code trials and ideas (still chaotic)

Here is the chaotic dataset (Updated), and code, still in progress

Tuesday, 2 July

Develop the classifcation model algorithm to obtain the difference between mainland and water

Wednesday, 3 July

Try to implement a manual loss function to guide the classification if very wrong

New presentation

Here is the new slides

Thursday, 4 July

Meeting with Hachem Kassem at 11:00

The working in progress presentation

Commit a lot of changes in coastal sentinel

Friday, 5 July

Fix bugs in coastsat software

View the issue written here

I was able to start manipulating the software

Monday, 8 July

I created a pull request on coastsat project to solve the issues i encountered

Coding the new prototype , work in progress ... (Check here)

Improve the new prototype by adding linear trend and cross shore change

Tuesday, 9 July

Begin to implement Openstreetmap coastline

Improve popup style and add new events on click

Download coastsat datas

Wednesday, 10 July

Try to manipulate the coastsat software but encounter many issues and bugs

Thursday, 11 July

Try to manipulate the coastsat software but encounter many issues and bugs

Slides meeting CEOC 11/07/2024

Friday, 12 July

Try to make AI predictions on None values in shoreleins datas but loss on 6*10^22 so still working on progress ...

Prediction model

Monday, 15 July

Prediction model easiest

Add a canvas for all transects in a shoreline popup

Tuesday, 16 July

Make time series forecasting on shore cross datapoints

Combine filled approaches to find values behind None, and apply trend method forecasting to predict the future (up to 2030) .

Wednesday, 17 July

Implement methods to predict the past (1970)