Hi, the point file you use at 8:07. What kind of file is this? What kind of information does it have? Is it just points with the right class of that place?
I need to delineate shoreline boundary . 1. Can I able to do for the coast boundary. 2. Is there any method for classifying the panchromatic satellite image along the coast
Actually It depends what you want to achieve. For crop classification you need to have a time series And several layers. If you want to classify for example 2 classes like water and land one layer should be enough for this. Best regards.
Yes of course. At train vector classifer stage you have choose random forest instead of ANN. And of course subscribe my chanell if you like it. Beat regards.
Sorry to bombard you with questions! In 'Train Vector Classifier' for 'Field names for training features', why do you select only mean values and not var values?
Sir, I want to work on Sentinel-2 data in QGIS using OTB but after downloading the Dataset from ESA Copernicus open access hub, I can't display or add the dataset in my project so I can't start analysis using OTB. Please help me or suggest any tutorials if you have any.
I've worked through this now and I seem to be missing some of the finer scale features. Is there a way to reduce the size of the segments to take finer scale features into account? Also is there anywhere that describes the variables that you are changing? Is it a case of adjusting the variables manually until you get a good result?
Best way is to test parameters on small raster. The most important parameter is range radius (smaller values produce smaller and more homogenous segments). You can also change minimum segment size.
Centroids were made from learn and control samples. Learn and control samples were obtain from cadastral parcels from LPIS system and farmers declarations about crops.
You don't need to downgrade. Version 7 is mostly the same as 6. The main difference is that they added zonal statistics for raster. But it you need older version you can download at CNES archive repository. Just Google it.
Thank you very much sir, it is a very useful video. I am interested in agricultural field segmentation; Should I have a large number of images (large time series) to get a precise segmentation ??
excuse me qir, but in the new otb versions, we should write all the names of the field when we want to Tainvectorclassifier; I did that but no results.
非常棒的视频，最近正在从事相关工作。非常幸运，及时看到了这个视频。
Great video, working on it recently. I was lucky enough to see this video in time.
Best regards. Thanks for subscribing.
Thank you for that video! Great idea to use points for an "in-situ" sampling and then later add the labels to the polygons.
You welcome. Best regards.
Hi, the point file you use at 8:07. What kind of file is this? What kind of information does it have? Is it just points with the right class of that place?
It’s a point layer with learn samples mostly crops. It is used for selecting segments for machine learning.
@@MrGIS thanks!
Hello, can you upload the exercise data? Thanks!
Hi! Thancks for the video. Where can i get the data?
very good video, but I want to ask at minute 23.08, where did you get the classification_result file from? thank you very much for the tutorial
I did the process before the tutorial so the file was already there and I overwrite it. At this step You nedd to type file name. Best regards.
I need to delineate shoreline boundary . 1. Can I able to do for the coast boundary. 2. Is there any method for classifying the panchromatic satellite image along the coast
Method is flexible but you need to test parameters for segmentation and classification based on your source datasets.
Wanted to know the file you've used for extracting the polygon features...is it a simple segmented file or did you run zonal statistics on the file?
I did segmentation and the result is segments shapefile with mean value of each band of an raster stack.
@@MrGIS so if i want to run this on a single image then the desired result will not be received right?
Actually It depends what you want to achieve. For crop classification you need to have a time series And several layers. If you want to classify for example 2 classes like water and land one layer should be enough for this. Best regards.
@@MrGIS so i'm looking for building extraction portion and was wondering if this tool will work best for me....
awsome tutorial mate, i wanted to ask is there any way to apply random forest classifier in this process ?
Yes of course. At train vector classifer stage you have choose random forest instead of ANN. And of course subscribe my chanell if you like it. Beat regards.
Sorry to bombard you with questions! In 'Train Vector Classifier' for 'Field names for training features', why do you select only mean values and not var values?
Variance do not improve classification. It is helpful for excluding outliers in the learning samples.
@@MrGIS thanks!
Hi, thanks for the video. How many dates and which S2 bands did you use in the raster stack?
About 20 dates for S1 and about 4 dates for S2
@@MrGIS Thanks for the reply, which S2 bands did you use or was it a vegetation index?
@@MrGIS Sir, can you explain to us what you ve used also sentinel 1 ??
Why I wanted to say ??
@@cars_worldcw488 For dense time series. In my country there is a lack of clear sky for S2 so the only alternative are SAR images.
Sir, I want to work on Sentinel-2 data in QGIS using OTB but after downloading the Dataset from ESA Copernicus open access hub, I can't display or add the dataset in my project so I can't start analysis using OTB. Please help me or suggest any tutorials if you have any.
Check out Semi Automatic Classification Plugin for QGIS for downloading and preprocessing Sentinel 2 datasets.
@@MrGIS in 4:30 you have added an input file in "mapla". So from where you have downloaded and processed that?
I've worked through this now and I seem to be missing some of the finer scale features. Is there a way to reduce the size of the segments to take finer scale features into account? Also is there anywhere that describes the variables that you are changing? Is it a case of adjusting the variables manually until you get a good result?
Best way is to test parameters on small raster. The most important parameter is range radius (smaller values produce smaller and more homogenous segments). You can also change minimum segment size.
@@MrGIS will do thanks! Can I add other datasets as well such as a texture layer?
Yes you can add but remember that range radius refers to the raster values.
Thank you for the video. The OTB was recently updated to 7.0. "User definied input centroids" has been added. Can you explain how to process the data?
Centroids were made from learn and control samples. Learn and control samples were obtain from cadastral parcels from LPIS system and farmers declarations about crops.
@@MrGIS Thank you very much, I am using OTB 7.0. Do you know how to downgrade to OTB6.6?
You don't need to downgrade. Version 7 is mostly the same as 6. The main difference is that they added zonal statistics for raster. But it you need older version you can download at CNES archive repository. Just Google it.
@@MrGIS Thank you for your kind reply.
Great video! Im new with geoprocessing. How can I create an agriculture mask vector as in the video? Thanks a lot!
It's from administrative LPIS database. Check out databases available in your country.
What did you click to open this? 1:19
You mean layers names? They were opened in notepad from envi hdr raster format.
Thank you for videolesson. Could you tell me what file format of input image did you use (5:55)?
ENVI hdr
@@MrGISsir can you tell me if Can I use a merged .tif satellite image as an input ??
@@cars_worldcw488 yes you can
Can you inform me about your points please? Are they your "training samples" and you are just using points instead of polygons?Thanks!
Points are insitu samples and im using them for selecting segments for training subset.
Okay, thank you! Great video by the way!
You're welcome. Best regards.
Thank you very much sir, it is a very useful video.
I am interested in agricultural field segmentation; Should I have a large number of images (large time series) to get a precise segmentation ??
You need to calculate statistics for time series. Also it improves segmentation accuracy.
@@MrGIS Can you explain to me more? Which statistics? for which object: pixel, images or what sir?
For every segment. Based on this statistics machine learning and final classification will obtain.
@@MrGIS yes sir. But I mean the qualite of the large_scale_segmentation
I have a problem sir: for new OTB versions, how can we write the list of field names for training features
As an integer number.
excuse me qir, but in the new otb versions, we should write all the names of the field when we want to Tainvectorclassifier; I did that but no results.
so how we can write them and the same problem with the output file extension
In attribute table should be column describing training feature as an integer number.
@@MrGIS thank you sir. I am talking about the meanB columns
where did you make the training samples?
In situ
@@MrGIS what qgis version you used for this tutorial? im trying to do this tutorial in qgis 3.10.5 and it failed
@@MrGIS also, did you use SCP plugin for processing the raster data?
As I remember it was 3.12
No. Everything OTB
How do you get the Sentinel 1 and Sentinel 2 data to match up?
Best way is to do coregistering and stacking in snap or ENVI.
@@MrGIS could you do a tutorial on that? :)
Here is a tutorial how to do it in QGIS ua-cam.com/video/91a41GTCmZ0/v-deo.html
@@MrGIS Awesome thanks! I'm just working through the tutorial with some Sentinel 2 data. Really appreciate your tutorial!