Sir i dont know how to express my happiness for this wonderful content, This is really very helpful for my academic project. Thank u thank u sooooooo much. We need more content like this...........
From this tiles, it is not possible because there are the tiles from random location. But for such work, I recently wrote the library named as "geotile" which will help you to create tiles as well as merge tiles. Library github: github.com/iamtekson/geotile
This video is very informative. Such types of tutorial are very rare. I am also working on landslide it will help me for my research. My best wishes for you and i am waiting for such videos on landslide inventory and prediction
Great video. Kindly ask that for the Sentinel 2 images which platform was used to download it, maybe different imges download platform has diffenent outcomes. Thanks!
I am also exactly not sure, which platform was used to download the original imagery. Please have a look to the published paper or landslide4sense website.
the dataset provided, is it from a specific study area, im very new to ml and related topics but find it interesting. i wanna try this with a dataset used in a research paper for a specific geographic area, it state that it used Digital terrain model along with Enhanced Natural Terrain Landslide Inventory (ENTLI). it would also be beneficial for me to create something for a specific area. Also some papers mentioned other factors like rainfall and such, can this project use factors like that as a parameter. This topic was more complex than i anticipated so im asking alot of questions
@@geodev okay, I understand that random forest method needs for dependent data such as yes/no or landslide/non landslide. But you run the model only with landslide data, is that right?
@@OmenJap Not really. In segmentation tasks you have masks/patches of the corresponding satellite image which acts as a label (for model training) and these masks/patches includes pixels from both the landslide and non-landslide class. So, such models can identify both landslides and non-landslides pixles and segment only the class of interest.
model = unet_model(128, 128, 6) #model.summary() checkpointer = tf.keras.callbacks.ModelCheckpoint("best_model.h5", monitor="val_f1_m", verbose=1, save_best_only=True, mode="max") #earlyStopping = tf.keras.callbacks.EarlyStopping(monitor='val_f1_m', patience=10, verbose=1, mode='max') callbacks = [ #earlyStopping, checkpointer ] history = model.fit(x_train, y_train, batch_size=16, epochs=100, verbose = 2, validation_data=(x_valid, y_valid), callbacks=callbacks) model.save("model_save.h5") I am getting an error like this. What should I do to execute this code.
ValueError: The filepath provided must end in `.keras` (Keras model format). Received: filepath=best_model.h5
I really like your videos. I want to know the dataset you have take above in landslide detection it's corrupted file. I tried to download train data and it shows file is corrupted. What should i do?
![Machine Learning Predicts Floods and Landslides [2024] | AI Project](/img/n.gif)
I really appreciate it, so far the best video about the ML application in natural hazards.
More contents related to ML/DL are coming on satellite imagery. Stay tuned!
Sir i dont know how to express my happiness for this wonderful content, This is really very helpful for my academic project. Thank u thank u sooooooo much. We need more content like this...........
It is my pleasure! All the best and stay tuned for future similar tutorials
Great Tutorial!!! Congratulations, and of caourse, thanks for not being jealous with your knowledge and thanks for sharing everything you did.
My pleasure! Glad you liked it!
Thank you so much...❤️ expecting more deep learning tutorial for geosciences application which is very rare in UA-cam. No body teaches that
Can you tell me how can we merge this patches to create a single output if we have a geotiff image and how can we convert h5 into geotiff
More tutorials on the way. Stay tuned!
From this tiles, it is not possible because there are the tiles from random location. But for such work, I recently wrote the library named as "geotile" which will help you to create tiles as well as merge tiles. Library github: github.com/iamtekson/geotile
@@geodev Thank you so much. I would like to request one video on how to create our own dataset(patches and masks) from satellite data.
@@rahulds001 definitely, i will creat. Stay tuned😃
This video is very informative. Such types of tutorial are very rare. I am also working on landslide it will help me for my research. My best wishes for you and i am waiting for such videos on landslide inventory and prediction
Glad it was helpful! More videos are on way, stay tuned!
@@geodev sir please share with me your contact email id and ph. no. i have more work related to landslide, we can do with collaboration.
Great video 👍🏻 hope you make more such tutorials
Thank you, Sure I will create more tutorials on deep learning. Stay tuned!
Great video and important topic Tek!
Glad you think so-:) Thank you Mikey!
Great video
Glad you enjoyed it
Thank you so much professor.
You are very welcome
Great video. Kindly ask that for the Sentinel 2 images which platform was used to download it, maybe different imges download platform has diffenent outcomes. Thanks!
I am also exactly not sure, which platform was used to download the original imagery. Please have a look to the published paper or landslide4sense website.
How to get mask data for validation dataset... its not provided by land4sense too!!
Can u help with doing that
For the validation set, you can test and generate the result. Sorry I haven't tested the model for validation set.
Thank you Sir much awaited topic
Always welcome
the dataset provided, is it from a specific study area, im very new to ml and related topics but find it interesting. i wanna try this with a dataset used in a research paper for a specific geographic area, it state that it used Digital terrain model along with Enhanced Natural Terrain Landslide Inventory (ENTLI). it would also be beneficial for me to create something for a specific area. Also some papers mentioned other factors like rainfall and such, can this project use factors like that as a parameter. This topic was more complex than i anticipated so im asking alot of questions
Very knowledgeable video
I want to know
What deep learning model is it ???
Is it CNN??
Yes it is CNN. To be more precise, It is Unet model
Why would you set NoData to 0.000001 instead of 0 or 256?
Do you have any references or literature to understand the theory of this method?
Can I do similar analysis with LiDAR data? If it is possible please do a video please
I think it will be possible with RGB imagery along with LiDAR point cloud. At the end, we need to create the DEM/DSM for landslide detection.
Thank you for the video sir 🙏
Could you please help to know about How to import utils?
Im getting error in importing
Hi, you need to write utils.py file as well. Please check the github repo and download the full code.
sir, i need full dataset link for project, thanks for the info
Your lecture is very useful. Please let me know where I can download your dataset. I can't download it at this moment.
How can i use arcgis for data collection
You can manually digitize the labels and produce the image tiles using "Export Training Dataset Using Deep Learning" tool.
How can I convert from TIFF to H5 or H5 to tiff, or any gis software can do it ?
H5 format doesn't come with coordinate system. But anyway if you want to export as a image, write it using rasterio or gdal.
Did you only use landslide location to run the model? So, We don't need to include non-landslide points in the inventory data.
Sorry, I used the data from landslide4sense challenge, which is not geolocational data.
@@geodev okay, I understand that random forest method needs for dependent data such as yes/no or landslide/non landslide. But you run the model only with landslide data, is that right?
@@OmenJap Not really. In segmentation tasks you have masks/patches of the corresponding satellite image which acts as a label (for model training) and these masks/patches includes pixels from both the landslide and non-landslide class. So, such models can identify both landslides and non-landslides pixles and segment only the class of interest.
bought your course on udemy
Great! Thanks for the purchase.
can you share your drive link where you have stored your project because i am not able to downlaod from site
please
Hi, I have removed the data from my drive but you can get the same dataset here: www.kaggle.com/datasets/tekbahadurkshetri/landslide4sense
Can you please make such video for air pollution prediction😊
model = unet_model(128, 128, 6)
#model.summary()
checkpointer = tf.keras.callbacks.ModelCheckpoint("best_model.h5", monitor="val_f1_m", verbose=1, save_best_only=True, mode="max")
#earlyStopping = tf.keras.callbacks.EarlyStopping(monitor='val_f1_m', patience=10, verbose=1, mode='max')
callbacks = [
#earlyStopping,
checkpointer
]
history = model.fit(x_train, y_train, batch_size=16,
epochs=100,
verbose = 2,
validation_data=(x_valid, y_valid),
callbacks=callbacks)
model.save("model_save.h5")
I am getting an error like this. What should I do to execute this code.
ValueError: The filepath provided must end in `.keras` (Keras model format). Received: filepath=best_model.h5
I really like your videos. I want to know the dataset you have take above in landslide detection it's corrupted file. I tried to download train data and it shows file is corrupted. What should i do?
Did you download the data from here: www.iarai.ac.at/landslide4sense/challenge/? I have worked on this data and it is not corrupted.
sir, Is this code is possible for real time images
Yes, If you have an real time images, it will definitely works.
arey hindi bol na
I found your project, it's very well done, I have to contact with you for some problems plz help me
I really appreciate you, would you please share your email