1/3. Data Processing for the Satellite Imagery (th-cam.com/video/3Xn21RT-y7Y/w-d-xo.html) - Data Preparation &Satellite Data Processing 2/3. Deep Learning with Satellite Image Data (th-cam.com/video/UBzMgr6yfpw/w-d-xo.html) - Deep Learning Model Training, Prediction with model performance - Local Debugging 3/3. Advance Deep Learning with Satellite Image Data (th-cam.com/video/i40ulpcacFM/w-d-xo.html) - Remote debugging with Weights & Biases - Model Saving and Reloading with custom metrics and loss function - Activation/Gradients outputs with heatmap, - Model Deployment application and Model Serving app on Hugging Face
Thank you so much for your wonderful comment, appreciate it sincerely and such comments are the motivation behind keep creating such content.. Thank you one more time!!
Hello sir, I need to integrate satellite data, Geospatial data like flood point and non flooded point As well ndvi nwdi etc in ML to trained my model Could you please help me out with this
Thanks for your comment. The initial data may have been multispectral however when I was processing in my tutorial it was all RGB. I haven't performed any multispectral analysis in my workshop.
@@650AILab Can you please tell me how initial data was multispectral? It was 3 channels right from the beginning,right? And also how would one go for multispectral analysis? Thanks! Your tutorials help me a lot
My training result shows a linier increasing of validation loss on epoch 50 to 100, but the training loss is decreasing through the training epoch, is it a sign of overfitting? and how do I overcome it? Note that I only used 1 dataset from Kagge, I haven't tried using 38GB satellite data (due to lack of storage). Thank you for the tutorial, waiting for the next part!!.
Thanks for the comment... Yes, if your validation loss is not going down subsequently in the later part of training epochs, ether it is not generalizing or just overfitting.. You can play with model params as well as dropout rate (start with .1) to to make sure u can generalize your model well.. Appreciate your comment and part 2/3 is out, please enjoy.
I never comment normally on youtube videos, but the way you teach and walk through the code is awesome. Normally what people do is just write the final code and explain what is happening, but you code it as if someone would normally do it for the first time, printing to see the results, checking our logic on a single line and then making a function out of it. You are a great teacher!
Appreciate your comment sincerely. I do because I enjoy it most sharing with folks like you. I have done the coding for living, built great products in past used by large global enterprises. I don't do it for clicks or advertisement revenue that's why If I do it, I do it with heart and enjoy doing it. Thanks.
I am starting my thesis by creating a model that can predict LULC maps from a particular area from time series of the exact location in the past that can be used to train deep learning models that generate future developments in advance. I hope this Workshop will be of great help..thanks in advance
Sounds great! I am glad that you find it useful and if you need more help just send me an email and I will be happy to assist you, no problem at all. All the very best for you and appreciate your comment.
hello, your video is very useful. I am still a newbie with neural networks but I learnt that for training the model, we need a great number of labelled datasets which I am currently missing. I do have access to Lidar and Hyperspectral imagery covering a specific area and was wondering if its a good idea to use CORINE land cover classes as labels and then make prediction of LCC using Lidar and Hyperspectral and maybe SAR? or the quality of training data (labels) could be too bad? Corine LC2018 is made from Sentinel and Landsat8. Kind regards
Thanks for the comment and its a great use case. I am currently busy with some other engagements however if you have a simple prototype (even if it is not working, but at least it has data collection part working) I can extend it to make it work as you suggested. Please reach out to me at my channel email. Thanks again.
Thanks for the comment, appreciate it sincerely. Not all the images are same size to start with so before all the images are going to processed for the model, patching is needed to make sure all the images are fitting based on source images.
Its very easy. you just need Google Gmail account and then please follow the given TH-cam video to get Colab Free working for you: th-cam.com/video/izlugRckG0Q/w-d-xo.html Good luck. Thanks for reaching out.
Hello sir thank you so much for this video series, youre doing a great service for the society. Is it possible for you to make a video on semantic change detection of satellite images using unet where we implement a dissimilarity decoder ?
Thanks for your comment and and I will be happy to help you. Please reach out the me through my channel email. I will be happy to assist you. Please create a notebook to collect the data first and process it, make it ready for feeding into the network and rest I will take care of it. Thanks.
I am a complete beginner when it comes to all things deep learning. Can you go over some basic steps about how to apply the model to new images? Thanks.
Thanks for clearing my basic on how to process images, mask and associate labels with them. @Prodramp do you also have any tutorial on how to do unsupervised learning on hyperspectral images ? Unsupervised modeling on satellite images will help me greatly because I have some hyperspectral images but they are not labelled.
Hello, I am a student from China, and my English level is not very good. So there is a question when listening to your lecture. Why do you need to convert BGR to RGB for masks images, but not for jpg images? This is translated by translation software. If the words or tone are not very good, I hope you will forgive me. Thank you for your selfless sharing!
thank you for this amazing tutorial but I tried and the color changed in the images and when I apply the BGR2RGB like mask the images now are the same like masks, can anyone conform that ?
Nice. But you spent a lot of time on explaining the file system and how to manipulate that in Python unnecessarily. This confuses and deviates the focus of this lecture.
Thanks for your comment, appreciate it. So if I am understanding it correctly your segmentation images are TIFF, is that right? - For TIFF segmentation image, you can use Python PILLOW library to extract each TIFF layer as PNG and then combine all the layers back to final PNG segmentation image.. If you could please share ~10 images and I can share code with you on how to get it all working with this exercise.
@@650AILab Yes Sir the images are TIFF, I can definitely share the images with you, please let me know which way would you prefer one drive or google drive? and where I should send you the link?
Hello Dear Chauhan, i watch your tutorials about Deep learning but i want to know how to create dataset using Landsat imagery or Sentinel-2. Especially how to create mask and label of image Tile. I create tiles using Geotile. Thanks for your help
Thanks for your comment and sorry for the delayed response as I was traveling. Please send me an email with the details about your Sentinal-2 output data format with a few samples. I will get back to you directly over email and once we have a solution I will create a separate video on this topic.
@@tankformations1048 Hello, I am new to deep learning, and I am currently working with Sentinel-2 data. I was wondering if you could share any experiences or projects related to deep learning and Sentinel-2 data that I can learn from. Thank you!
1/3. Data Processing for the Satellite Imagery (th-cam.com/video/3Xn21RT-y7Y/w-d-xo.html)
- Data Preparation &Satellite Data Processing
2/3. Deep Learning with Satellite Image Data (th-cam.com/video/UBzMgr6yfpw/w-d-xo.html)
- Deep Learning Model Training, Prediction with model performance
- Local Debugging
3/3. Advance Deep Learning with Satellite Image Data (th-cam.com/video/i40ulpcacFM/w-d-xo.html)
- Remote debugging with Weights & Biases
- Model Saving and Reloading with custom metrics and loss function
- Activation/Gradients outputs with heatmap,
- Model Deployment application and Model Serving app on Hugging Face
Instead of using Dubai Dataset can i use Sentinel-2 3 months composit image for classification? can yoou share a video how to do it?
nice video
you are one of the few influencer who are really adding values to the community.
Thank you so much for your wonderful comment, appreciate it sincerely and such comments are the motivation behind keep creating such content.. Thank you one more time!!
At 24:38 it should be "for i in range(1,9):" instead of "for i in range(1,8)" as there are 8 ids because python skips the last digit.
Sir can u suggest which type of hardware is okay for deep learning rtx ddr4 ram 8gb i7 processor is okay ?or i need more adavanced system
Hello sir,
I need to integrate satellite data, Geospatial data like flood point and non flooded point
As well ndvi nwdi etc in ML to trained my model
Could you please help me out with this
is this multispectral analysis ? cuz all i saw is RGB
Thanks for your comment.
The initial data may have been multispectral however when I was processing in my tutorial it was all RGB. I haven't performed any multispectral analysis in my workshop.
@@650AILab Can you please tell me how initial data was multispectral? It was 3 channels right from the beginning,right?
And also how would one go for multispectral analysis?
Thanks! Your tutorials help me a lot
I want to do multifeature and multiscale high spatial resolution image segmentation is it possible by using U-net
Sir I want to detect clouds to satellite images is this tutorial help to achieve this
Why can't I find you till now? You are amazing. Thank you very much.
You are most welcome with this sincere comment, thanks for your feedback, appreciate it sincerely.
Great masterclass! Greeting's from Chile
Thank you so much for your contribution, effort and time. This channel is a valuable treasure on TH-cam.
My training result shows a linier increasing of validation loss on epoch 50 to 100, but the training loss is decreasing through the training epoch, is it a sign of overfitting? and how do I overcome it? Note that I only used 1 dataset from Kagge, I haven't tried using 38GB satellite data (due to lack of storage). Thank you for the tutorial, waiting for the next part!!.
Thanks for the comment...
Yes, if your validation loss is not going down subsequently in the later part of training epochs, ether it is not generalizing or just overfitting.. You can play with model params as well as dropout rate (start with .1) to to make sure u can generalize your model well..
Appreciate your comment and part 2/3 is out, please enjoy.
Thanks for creating such an amazing tutorial, waiting for the next part.
Thanks for your comment, Part 2 to ready and will be live by Wednesday 6:00 AM PST, enjoy.
Hello sir, how to find the height and weight of the road from satellite images
In deep learning?
You are simply legend sir thanks for these lectures sir.
Hello Sir
If you can suggest some tutorial on how to create these multi label mask images for satellite data, that will be very helpful.
I am really 💪😲🤳 excited for this project! ❤
Thank you so much, sincerely appreciate it..
I never comment normally on youtube videos, but the way you teach and walk through the code is awesome. Normally what people do is just write the final code and explain what is happening, but you code it as if someone would normally do it for the first time, printing to see the results, checking our logic on a single line and then making a function out of it.
You are a great teacher!
Appreciate your comment sincerely. I do because I enjoy it most sharing with folks like you.
I have done the coding for living, built great products in past used by large global enterprises.
I don't do it for clicks or advertisement revenue that's why If I do it, I do it with heart and enjoy doing it.
Thanks.
Instead of using Dubai Dataset can i use Sentinel-2 3 months composit image for classification? can yoou share a video how to do it?
where can we find those data sets sir
I am starting my thesis by creating a model that can predict LULC maps from a particular area from time series of the exact location in the past that can be used to train deep learning models that generate future developments in advance. I hope this Workshop will be of great help..thanks in advance
Sounds great! I am glad that you find it useful and if you need more help just send me an email and I will be happy to assist you, no problem at all. All the very best for you and appreciate your comment.
May I know about this topic? Can you give me your email? I have some questions.
Helo brother I want to talk with you about my fyp project plz reply me
hello, your video is very useful. I am still a newbie with neural networks but I learnt that for training the model, we need a great number of labelled datasets which I am currently missing. I do have access to Lidar and Hyperspectral imagery covering a specific area and was wondering if its a good idea to use CORINE land cover classes as labels and then make prediction of LCC using Lidar and Hyperspectral and maybe SAR? or the quality of training data (labels) could be too bad? Corine LC2018 is made from Sentinel and Landsat8. Kind regards
Thanks for the comment and its a great use case. I am currently busy with some other engagements however if you have a simple prototype (even if it is not working, but at least it has data collection part working) I can extend it to make it work as you suggested. Please reach out to me at my channel email.
Thanks again.
Hey Avkash, Can you please explain why you are patching each image?
Thanks for the comment, appreciate it sincerely.
Not all the images are same size to start with so before all the images are going to processed for the model, patching is needed to make sure all the images are fitting based on source images.
when use patchify function to execute subset, why you doesn't consider overlap?
why did after u did patched_image.shape,it showed (2,1,1,256,256,3),i understood the latter but i didn get the 2,1,1 ,can anyone please help?
Asalamalikum, sir I have a query regarding the Deep image output , when there were coordinates on earth how we present those values with map
Great job! Thank you so much for this very helpful content.
Thanks for the comment, appreciate it.
Sir how to get into the google colab please inform me
Its very easy. you just need Google Gmail account and then please follow the given TH-cam video to get Colab Free working for you:
th-cam.com/video/izlugRckG0Q/w-d-xo.html
Good luck. Thanks for reaching out.
I have a question why we convert to RGB in mask images only when we use CV2.imread ??
Would you help me downloading the images ? It seems like there is no available link in the website provided
hey, can you suggest a way to find satellite images for training a deep learning model(wildfire detection)?
The full hands-on coding workshop is exactly what I needed to get started. Thanks for sharing this valuable content!
46:00
44:00
34:00
15:00
BROOO thankyou so much, this really helped and the tutorial was really easy to use as well :)
Thank you so much, I am glad you find it helpful, appreciate it.
you are a Legend
Thank you so much, appreciate your comment. Will try my best.
Very nice
At 33:42, there is inconsistency in the image size written in the code. Also, there is typo in the for loop. You are not taking the 8th tile.
How can I create a dataset with any satellite image like this dubai dataset?
Hello sir thank you so much for this video series, youre doing a great service for the society.
Is it possible for you to make a video on semantic change detection of satellite images using unet where we implement a dissimilarity decoder ?
Thanks for your comment and and I will be happy to help you. Please reach out the me through my channel email. I will be happy to assist you.
Please create a notebook to collect the data first and process it, make it ready for feeding into the network and rest I will take care of it. Thanks.
sir what is you channel mail id?
I am a complete beginner when it comes to all things deep learning. Can you go over some basic steps about how to apply the model to new images? Thanks.
can i do the same with drone arial imagery ?
Thanks for clearing my basic on how to process images, mask and associate labels with them. @Prodramp do you also have any tutorial on how to do unsupervised learning on hyperspectral images ? Unsupervised modeling on satellite images will help me greatly because I have some hyperspectral images but they are not labelled.
Hello, I am a student from China, and my English level is not very good. So there is a question when listening to your lecture. Why do you need to convert BGR to RGB for masks images, but not for jpg images? This is translated by translation software. If the words or tone are not very good, I hope you will forgive me. Thank you for your selfless sharing!
The problem is about 45 minutes in the video. The code is this sentence
if image_type == 'masks':
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
thank you for this amazing tutorial but I tried and the color changed in the images and when I apply the BGR2RGB like mask the images now are the same like masks, can anyone conform that ?
I am having the same problem too!
clear guide worked well with for sharing.
Thanks for the comment, appreciate it sincerely.
can i get the shape files so that i can access it in QGIS .
are you using colab pro?
god bless u xdd
Thank you so much, appreciate your comment.
Nice. But you spent a lot of time on explaining the file system and how to manipulate that in Python unnecessarily. This confuses and deviates the focus of this lecture.
You are so welcome! Appreciate your comment.
I have to work on Tiff file formats, which is very confusing can you please tell me how I can use tiff format for all this workshop?
Thanks for your comment, appreciate it.
So if I am understanding it correctly your segmentation images are TIFF, is that right?
- For TIFF segmentation image, you can use Python PILLOW library to extract each TIFF layer as PNG and then combine all the layers back to final PNG segmentation image.. If you could please share ~10 images and I can share code with you on how to get it all working with this exercise.
@@650AILab Yes Sir the images are TIFF, I can definitely share the images with you, please let me know which way would you prefer one drive or google drive? and where I should send you the link?
Perfect sir
thank you are an inspiration
Hello Dear Chauhan, i watch your tutorials about Deep learning but i want to know how to create dataset using Landsat imagery or Sentinel-2. Especially how to create mask and label of image Tile. I create tiles using Geotile. Thanks for your help
Thanks for your comment and sorry for the delayed response as I was traveling. Please send me an email with the details about your Sentinal-2 output data format with a few samples. I will get back to you directly over email and once we have a solution I will create a separate video on this topic.
@@650AILab thanks for your reply. I send you an email with data. I am very happy to hear you. See you soon.
@@tankformations1048 Hello, I am new to deep learning, and I am currently working with Sentinel-2 data. I was wondering if you could share any experiences or projects related to deep learning and Sentinel-2 data that I can learn from. Thank you!
did you get a solution for this@@tankformations1048
👌🏻 great
Thanks for the comment, appreciate it.
thank you so much for sharing this method with !
Thanks for the comment, appreciate it sincerely.
Genuis
Thank you for your amazing lessons. I learned a lot!