Thak you so much for your tutorial,I think this type of tutorial is very much helpful for any user if you show the data preparation process using raw data and classify this according to any deep learning process. Thank you again.
Thank you very much for your tutorial, it will really help all users who wish to use Deep Learning technology to solve their respective problem. However, if you can also show the production process of this model, it will be even more interesting. Because many developers stop after saving and analyzing training results. But do not really affect the deployment of these models under a platform or locally. In short, if you can also show how to classify land cover features with this model as you often do on Earth Engine with other classification algorithms. Thank you very much.
@@geodev Thank you very much for considering my suggestion. I will be the first person to like and share the Deep Learning model release tutorial you are going to develop. Because it hardly exists on RUclips. Thanks very much!
I concur, i urgently need a video that will show how to infer this model for feature classification. On a different note, i am trying to replicate this model for my research work but on (step 6: Model training) i am getting a NameError: 'convulutional_block' is not defined. How is this possible, i installed and imported all the packages as listed on this video. Do you have a suggestion on how i can resolve this?
Hlo. I m having error while loading the model after epochs. It says no training configuration found in the save file, so the model was not compiled . Compile it manually.. what do i do
Hey. Thank you for the informative video. I have a simple doubt. I am working for crop classification for SAR imagery using CNN and my ground truth dataset is in the shp file format. I want to prepare training and testing files but I am confused for how to separate the SAR imagery based on my ground truth for the training and test files. Can someone help me solve this? Thanks
Sorry I forgot to explain this step. But I write the code for it in the confusion matrix section. The model is written for the batch of the images. So to predict the class, you need to pass the batch of images (i.e. 32 images) and predict the label as below, preds = model.predict(image_batch) y_pred = np.argmax(preds, axis = 1) The y_pred will be list of 32 labels. If the value of y_pred is 0, that means, it is the first class image ("annualCrop"), if the value is 1, that means, it is second class ("Forest") and so one. I hope it will help you to understand how to predict the images. Sorry for the late response.
I have carefully seen your class, thank you very much. I have a question, I understood that you did the classification only for some unknown area based on the Eurosat dataset. That is, training classes and test classes were formed from the Eurosat dataset. Please tell me how can I make a classification for a selected area (say for my city) based on Sentinel 2 using training classes and test classes that I will generate from the Eurosat dataset?
I don't think you can classify the images from sentinal2. If you want to create the test images, you need to make sure that the properties (such as x and y dimensions, data types, bands) of your test image matches with training image.
Thank you for the video, It is really helpful. I need some clarification regarding the dataset used. In the data description it is said that "A novel dataset based on sentinel-2 satellite images covering 13 spectral bands". However, when I saw the dataset each image has only one layer in .jpg format. All the training and testing were done on these single layered satellite images. My questions are what is the purpose of all the remaining bands that were used in the dataset generation.? Does the deep learning models(RESNET50) consider spectral values of the sentinel bands for classification or will it use the color and feature patters from the image. Thank you in advance.
Hi Geospatial lab, the official eurosat dataset contain all the multi-spectral bands. For make the tutorial simpler, I have used RGB as the training dataset. One things is sure that, we can't use all the bands (it is computationally too costly). But you can select few other bands (such as NIR, SWIR) in the training dataset. If you want to know, how to add those bands on training dataset, please refer to my another tutorial here: ruclips.net/video/1kl5RNPkz-g/видео.html Talking to your second question about RESNET50 model, it is just a model. You can use other models too (such as Unet). We have to see, which model works best for LULC classification.
You can use model.predict() function for predicting the result. Here is the complete example of image classification problem: keras.io/examples/vision/image_classification_from_scratch/
recently I wrote the python library called "geotile" which helps to create the image chips as your need. Please have a look to this library here: github.com/iamtekson/geotile
Thak you so much for your tutorial,I think this type of tutorial is very much helpful for any user if you show the data preparation process using raw data and classify this according to any deep learning process.
Thank you again.
For the data preparation, just need to create the tiles using satellite imagery. After that, the data should be keep in the corresponding folders.
Thank for the tutorial , it is very informative and the quality is outstanding. As @Fils Magao mentionned the deployment part will help me alot
Glad it was helpful! Definitely, I am going to publish such tutorial in near future. It is already on my list. Stay tuned!
Thank you very much for your tutorial, it will really help all users who wish to use Deep Learning technology to solve their respective problem. However, if you can also show the production process of this model, it will be even more interesting. Because many developers stop after saving and analyzing training results. But do not really affect the deployment of these models under a platform or locally.
In short, if you can also show how to classify land cover features with this model as you often do on Earth Engine with other classification algorithms.
Thank you very much.
That is a great suggestion. Definitely in the near future, I will come up with the production process for model. Thanks for the suggestion.
@@geodev Thank you very much for considering my suggestion. I will be the first person to like and share the Deep Learning model release tutorial you are going to develop. Because it hardly exists on RUclips.
Thanks very much!
@@fils2magao130 I am really happy to hear that. Stay tuned, Thanks.
I concur, i urgently need a video that will show how to infer this model for feature classification. On a different note, i am trying to replicate this model for my research work but on (step 6: Model training) i am getting a NameError: 'convulutional_block' is not defined. How is this possible, i installed and imported all the packages as listed on this video. Do you have a suggestion on how i can resolve this?
@@yingisanichabalala105 I think you forgot to run the one cell for convolution_block function.
Hlo. I m having error while loading the model after epochs. It says no training configuration found in the save file, so the model was not compiled . Compile it manually.. what do i do
Thanks mhan really great work ❤️
My pleasure 😊
Hi. It was very helpful for me. Thank you so much.
Glad it was helpful!
Great ❤
Hii sir i need some guidance regarding generating synthetic images with triplegan on Eurosat, can you please help me.... Thanks in advance sir
Hey. Thank you for the informative video. I have a simple doubt. I am working for crop classification for SAR imagery using CNN and my ground truth dataset is in the shp file format. I want to prepare training and testing files but I am confused for how to separate the SAR imagery based on my ground truth for the training and test files. Can someone help me solve this?
Thanks
could you assist me how to create custom dataset like that 10 classes? It will help me a lot.
Without running epoch we can't get the predicted output
excellent explanation.
please can you provide code to print a batch of predicted images, or maybe a custom image
Please
Thank you
Sorry I forgot to explain this step. But I write the code for it in the confusion matrix section. The model is written for the batch of the images. So to predict the class, you need to pass the batch of images (i.e. 32 images) and predict the label as below,
preds = model.predict(image_batch)
y_pred = np.argmax(preds, axis = 1)
The y_pred will be list of 32 labels. If the value of y_pred is 0, that means, it is the first class image ("annualCrop"), if the value is 1, that means, it is second class ("Forest") and so one. I hope it will help you to understand how to predict the images.
Sorry for the late response.
I have carefully seen your class, thank you very much. I have a question, I understood that you did the classification only for some unknown area based on the Eurosat dataset. That is, training classes and test classes were formed from the Eurosat dataset. Please tell me how can I make a classification for a selected area (say for my city) based on Sentinel 2 using training classes and test classes that I will generate from the Eurosat dataset?
I don't think you can classify the images from sentinal2. If you want to create the test images, you need to make sure that the properties (such as x and y dimensions, data types, bands) of your test image matches with training image.
Thank you for the video, It is really helpful. I need some clarification regarding the dataset used.
In the data description it is said that "A novel dataset based on sentinel-2 satellite images covering 13 spectral bands". However, when I saw the dataset each image has only one layer in .jpg format. All the training and testing were done on these single layered satellite images.
My questions are what is the purpose of all the remaining bands that were used in the dataset generation.?
Does the deep learning models(RESNET50) consider spectral values of the sentinel bands for classification or will it use the color and feature patters from the image.
Thank you in advance.
Hi Geospatial lab, the official eurosat dataset contain all the multi-spectral bands. For make the tutorial simpler, I have used RGB as the training dataset. One things is sure that, we can't use all the bands (it is computationally too costly). But you can select few other bands (such as NIR, SWIR) in the training dataset. If you want to know, how to add those bands on training dataset, please refer to my another tutorial here: ruclips.net/video/1kl5RNPkz-g/видео.html
Talking to your second question about RESNET50 model, it is just a model. You can use other models too (such as Unet). We have to see, which model works best for LULC classification.
what about the prediction? how can i do it?
You can use model.predict() function for predicting the result. Here is the complete example of image classification problem: keras.io/examples/vision/image_classification_from_scratch/
hello and thank you for informative tutorial, I need to visualize my predicted model as a classified map please help me to add this line of code it
thankyou for the video.
are you Nepali by any chance?
Yes, I am Nepali Guy. Thanks for visiting!
@@geodev dhanyabad dai lai
kei sodhnu cha bhane hajur lai sodhna milcha hola hai
@@arpankhanal2980 you are welcome vai!
how to create this image chips ?
recently I wrote the python library called "geotile" which helps to create the image chips as your need. Please have a look to this library here: github.com/iamtekson/geotile
@@geodev After installing the pakage inside many environment, I still have error when running "from geotile import GeoTile" . I don't know why
how you prepare the data set ?
please search for end-to-end deep learning with google earth engine (geodev) and you will get the playlist explaining everything!
Sa
-:)