Automatic Road Extraction From Aerial Photographs/Satellite Imagery Using ArcGIS Pro Part 1
HTML-код
- Опубликовано: 1 окт 2024
- Automatic Road Extraction From Aerial Photographs/Satellite Imagery Using ArcGIS Pro Part 1. Apply the finished model to extract the road network in raster format.
This video is the first part of the few videos about how to extract road network from satellite imagery / aerial photographs in ArcGIS Pro.
Finished DLPK model can be downloaded from : www.arcgis.com...
ArcGIS Pro Deep Learning Libraries can be downloaded from : github.com/Esr...
Second Video already uploaded here: • Automatic Road Extract...
Music by : www.bensound.com
#arcgis #arcgispro #deeplearning #AI #machinelearning #geospatial #geoai #roadmapping #roadextraction #roadnetworkmapping #gis #geographicinformationsystem
How to build DLPK Deep Learning Model using your own data : ruclips.net/video/o1nXoR5ZnHI/видео.html
can the deep learning libraries from the link be used in Arcgis pro 3.1.5?
Esri always publish deep learning libraries according to the respective arcgis version, on later version (3.2 onwards) the deep learning libraries are included in the main installer.
Check them on esri github page
Sir, in case of the high quality data that you are using for road extraction, how can I get the Dolpic data that you are using?
Wah, part 1 udah keren ini. Siap menunggu sekuelnya. :)
Sir, in case of the high quality data that you are using for road extraction, how can I get the DLPK data that you are using?
Check the video description. I put the link there
@@GEO2004 Sir, I understand. If I work with any other data, I will use any speculative data for that dlpk.
Yes, for the next video I will demonstrate how to make the dlpk with our own data
@@GEO2004 Yes sir it would be great to understand how to create dlpk data and how to use it for every high imegery data.
@@GEO2004 Yes sir I am very excited
I'm glad I found this video. I will give it a try and see if I can make it... starting right from 0... and try to download the frameworks and DLPK... thank you for the video.
You can do it! good luck mate!
Sir,please provide the link of the website from which you are downloading aerial photographs.
here is how ruclips.net/video/qtNM3gev-nY/видео.html
Appreciate this video, since I've never heard of deep learning models before. I am wanting to extract road polygons from my local orthoimagery source. I'm using PRO 2.9 and have installed deep learning libraries from the link provided. I run the model using your input parameters and it appears to run successfully. However the output is another 3-banded raster (aerial image) and not polygon layer. Any thoughts?
hmmm thats strange, one thing I should mention is, in ArcGIS Pro 2.9 you don't have to install deep learning libraries because it is already come pre-installed along with arcgis installation, can you confirm that maybe there is some python library conflict? may be trying to install ArcGIS pro 2.9 from the scratch could help
It is really appreciable effort you have put to showcase your understanding/talent on how to automate capturing the road network using ArcGIS Pro Deep Learning tool. It works well with our RGB scanned coloured maps. But I have one problem: I am getting the result of the tif map with more than width of the polygon input features. So, when features are closely located, i am getting overlapping with each others features. Here my question is why the output road width in tif is more than the input polygon feature in shape file. What could be cause of this. Kindly request you to help in this regard. Thanks
There is a better way to handle that, if I am not mistaken, in latest arcgis pro version there is a feature to generate only the centerline of the road, I never tried it by myself
This is not directly answer your question, but I hope it helps
@GEO 2004 Sorry for the late response. Yes, it is not a direct answer. However, you can answer my question. As you mentioned in the video, the inputs are satellite imagery and road layer with polygon feature. This road layer has a width of about 30m, for example. The steps include 1. Exporting, 2. Training and 3. Classify. After classifying the pixesl, we get the output as imagery which can be used for automatic digitization using arcscan. But, width of the road pixels in the output imagery is nearly 120m. Why it is coming almost 4 times when compared with road layer (input) with buffer width of 35m. If the output pixel width of the road layer is 35m, then my problem would be resolved. Thanks.
@@balugdb you are using it for the scanned maps right? in my example I am using satellite imagery, that is why I can't answer your question, I personally never tried this algorithm for scanned maps, so I don't really know the model behavior, thank you
Hello, Thank you for this video. I am looking for a very very simplified version of this tool. Something that would take a scaned topographical map with and convert it into a shapefile. I found this video ruclips.net/video/0bqK0XPs_HA/видео.html but it is for ARcmap and has no audio. I'm wondering if something is already built out for this.
Your mentioned video is just a simple raster to vector conversion. It looks works superb in that video because the raster data is only contain single information. That would be different if the data is raster topographical map. Not only the result will be messed up, you also needs to clean those result with a lot of editing works.
My video is about machine/deep learning which is very different approach compared to raster to vector. Although the method is complex, it could handle almost all kind of raster data includes topographical map. Trust me mate, deep learning implementation in arcgis pro actually is already simple because esri successfully made a GUI to facilitate it, and you don't have to write codes in python or R in order to use these techniques
@@GEO2004 Thank you very much for your response. I should try this deep learning method on one of the scanned maps. Thanks again.
Yes man, you will be surprised with the result. Goodluck in your learning, cheers.
Hi Michael, did you ever figure this out? We're doing something similar, but with geologic contacts in old B&W scanned maps. Our maps also have topographical + road info, so they're too complex for raster to vector, like GEO 2004 says. I figured I could just fine-tune the road extraction model, since a road is conceptually pretty similar to a geologic contact, but I've been running into a lot of technical difficulties.
There are few new deep learning algorithms introducted in latest version of arcgis pro, maybe you can try your luck with those new algorithm
If I don't have access to high resolution sattelite imagery? Can I use screenshot images from Google maps or Esri, georeferencing and mosaic them together into one huge image?
Sure, however it would be impractical and tedious, the model will likely wont perform well because mosaicked screenshot images tend to have varying "true" resolution (not image/raster resolution).
@@GEO2004 thanks for the reply. I see, so what would be a good alternative? My AOI is Ho Chi Minh City in Vietnam but I don't have access to sattelite imagery here :/
No alternative I guess, either you can test the method you describe or get a true satellite imagery
@@GEO2004 I'll try for a small area and report back
Nice