Land Cover Classification using a Simple Deep Learning Model (TensorFlow) with Earth Engine Data

Hi Geospatial Enthusiast!
Would you like to classify land cover classification using a deep learning model such as Convolution Neural Network? In this video, I will show you how to use Earth Engine imagery and samples to build a land cover classification model in TensorFlow (Keras).
GEE script: code.earthengine.google.com/4...
Colab script: colab.research.google.com/dri...
The image used is the 2023 Landsat 8 and 9 composite of Jambi City, Indonesia. The land cover classes I use are built-up, bare land, water, wetland, herbaceous, dry shrub, wet shrub, palm oil, and plantation forest. Other than the spectral band from Landsat, the features or variables I use are the spectral indices and elevation from SRTM. The images/variables and extracted samples will then be exported to Google Drive where they will be mounted and used in Colab.
For the model, I am using a combination of Dense and Convolutional 1-dimensional layers. My model is far from perfect, you could improve it more. Also, this is 1 dimensional (tabular), you could use more dimensions using spatial data instead of extracted tabular data.
I hope this video helps with your research or project.
Email: [e-mail]
Linkedin: / ramiqcom
GitHub: github.com/ramiqcom

[@GiancarloAlciaturiDiaz]

Looks so good! I am going to try it.

[@user-xk7hi8ur9z]

thank you very much my dear realy you give me a big solution for extracting stack data and load them in colab

[@user-xk7hi8ur9z]

thank you a lot for script also

[@ahmedameen4587]

Good work

[@mtkavak]

Thanks, ❤❤

[@_zerofox]

🎯 Key Takeaways for quick navigation: 00:00 🌱 *Land Cover Data Preparation* - Preparing data from Earth Engine for land cover classification, - Masking, adding variables like elevation, and sampling land cover types. 01:10 📦 *Setup and Data Export* - Setting up Google Colab for training, - Exporting data for TensorFlow land classification. 02:01 📊 *Data Visualization and Exploration* - Visualizing data with packages like Matplotlib, - Exploring image and sample data before model training. 03:39 🖼️ *Image and Sample Processing* - Reading and shuffling image and sample data, - Splitting data into training and testing sets. 05:01 🧠 *Model Architecture* - Constructing a simple deep learning model using TensorFlow, - Configuring layers and input shape. 06:33 ⌛ *Model Training* - Training the model with Adam optimizer and early stopping, - Adjusting batch size for efficient training. 08:31 📉 *Model Evaluation and Optimization* - Evaluating model performance and predicting land cover classes, - Considering model complexity and potential improvements. 10:09 📊 *Comparison with Alternative Models* - Comparing model performance with a random forest model, - Analyzing accuracy and potential model generalization.

[@ramiqcom]

Thanks

[@dwiprabowo4511]

Hi Ramadhan, great tutorial ;) i am just missing about how to create 'sample lc', is there any another video for this? Huge thanks

[@ramiqcom]

it is in the script too, I make my own sample in GEE

[@user-is3se7fw1g]

Thank you sir, please explain for us drought assessment using Google earth engine

[@ramiqcom]

idk how tho

[@user-tm8ck3qx2b]

Thank for your code and demonstration. But the thing is why did you mention you have used Tensor Flow as a deep learning method? In your code you have used the ML Random Forest classification. In which section you use the TF method??

[@ramiqcom]

Please watch the video and read the description

[@nakulchoudhari6711]

Hi Ramadhan, I tried to use your code for my ROI and tried to take samples in 2 ways, first using polygons like in your code, and using the marker. But in both the techniques, the train size and test size are 0. Please help me out.

[@user-xk7hi8ur9z]

can you please do regression analysis with deep learning after

[@ramiqcom]

sure

[@zj1281]

Hi Ramadhan, nice video. I tried running your code on google earth engine but it didn't show up like how it did in your demo video. Could I also check where did you get your data from for the respective FeatureCollection(s)? Thank you.

[@ramiqcom]

The feature collection? I just made it there in the script

[@kavyajeetbora2585]

Hi Ramadan, is there any standard library for state of art AI models that can be directly used for classifying the landuse data based on given image collection ?

[@ramiqcom]

I dont know

[@shivkarnik3253]

Hey, can we use similar method to classify sentinel 1 sar data?

[@ramiqcom]

Maybe, I have not try it

[@user-rq7lm8ec1f]

Dear Ramadhan thanks for your informative lesson. I need your excel file to arrange my file according to that how can I get your csv file ?

[@ramiqcom]

I think I export it from the GEE

[@iqbalhabibiehabibie5689]

Hi Ramadhan. It is a very interactive video. I have some problems with my QGIS 3.28.11 Firenze it automatically unexpectedly close. Could you send me the link for the right tools? or any idea how to solve it. Also, Could you attach the RGB color to the description? and what the color represents? Thanks.

[@ramiqcom]

use the LTR QGIS. Check my script to see what does each color represent.

[@iqbalhabibiehabibie5689]

@@ramiqcom I could open the file raster file is okay on my QGIS but when I try to change the color palette close itself.

[@iqbalhabibiehabibie5689]

@@ramiqcom could you give me the link for LTR QGIS? Thanks

[@ramiqcom]

@@iqbalhabibiehabibie5689 www.norbit.de/~jef/QGIS-OSGeo4W-3.28.11-1.msi?US

[@shrabanikar8646]

Sir, can you help me to do CNN-2dconvolution-based LandUse LandCover Classification using vector sample points?

[@ramiqcom]

You need to convert vector to raster first

[@shrabanikar8646]

@@ramiqcom can you gave me any tutorial for this work.

[@nakulchoudhari6711]

I'm trying to add these tif file as an asset in google earth engine but everytime it is giving error code 13. Please help me out with it.

[@ramiqcom]

What is the message of code 13?

[@livewiregirl]

What deep learning method used?

[@ramiqcom]

just dense layer

[@saraswatisahoo2816]

why did you merge L8 and L9 data?

[@ramiqcom]

Using merge method

[@saraswatisahoo2816]

@@ramiqcom no, I'm asking that why did you merge L8 and L9. you could have used only L8 data. Is it necessary to merge L8 and L9 ?

[@ramiqcom]

@@saraswatisahoo2816 It give more clear data

[@user-qk7xq6xb4v]

hi can i get your drive i can t find it in my drive

[@ramiqcom]

Have you export it?

[@sitiruswandyah640]

halo Pak., bisa kah saya dapatkn script code nya. terima kasih.

[@ramiqcom]

Di description video

[@shirasenderling4272]

🌺 "Promo sm"

[@Health_Geography]

Thank you. But I'm not interested. Because, we use codes to solve our complex issues. You are over complicating a simple topic and it is confusing. A code should easily allow its specialized users to extract data. Not to confuse them. Thank you.

[@ramiqcom]

Maybe skill issue

[@9858695189]

Hi ramdan, this is very helpful in my research . can you please share your jupyter notebook file.