Urban Mapping

Microwave Toolbox
#1

Hello Everyone , I’m using Sentinel-1 data for mapping urban areas please suggest best way to for accuracy assessment. I’m confused how i will explain my final result

How do I perform Accuracy Assesment for a Radar processed and classified image in SNAP?
Change detection
Urban Accuracy assessment
Classification problem (nullpointerexception)
#2

Did you use GRD? Or Interferogram coherence?

In case of GRD, What are the steps processes did you follow?

I think it is possible to apply SVM classifier and getting the accuracy of the classification.

1 Like
#3

to have an objective accuracy assessment you need to have a reference dataset for comparison.

You can use the global urban footprint, for example: https://www.dlr.de/eoc/en/desktopdefault.aspx/tabid-9628/16557_read-40454/

Alternatively, you digitize points inside and outside the urban area and give them a binary attribute (1: urban or 0: non-urban). Then you intersect these points with your result, e.g. with the pin manager, and calculate the sensitivity of your result (that means how much true positives you have: https://en.wikipedia.org/wiki/Sensitivity_and_specificity)

2 Likes
#4

Thanks for your time
I’m Using Coherence and amplitude of SLC. Now suggest me best way

#5

Thanks for your time

if i want to comprasion with in one dataset not with other dataset. I think i should go for your second suggestion classifying image but question is here how can i validate these results of classification?

#6

I think for coherence you could apply RF-classifier, in case you have multi raster bands, also the suggestions of @ABraun both could be followed, my personal opinion this one is more realistic in this case,

Here you could find an example,

URBAN CLASSIFICATION WITH SENTINEL-1 Case Study: Germany, 2018

1 Like
#7

you cannot perform an accuracy assessment without external information.

You have created a result. The only way to find out how good it is, is to compare it with an independent reference. There is no information about accuracy within the result itself.

Collect points in Google Earth, for example, label them as urban or non-urban. Then import them into SNAP and check how many of the urban points are classified as urban in your result. This in relation to all urban points is the true positive value. The points which are labeled as non-urban and are non urban in your result is the true-negative value. And so on…

4 Likes
Forest species classification
#8

But the scale in this case is different, more detailed scale collected urban points from google earth could be classified as non-urban points,

#9

I agree that the scale is different, but as long as the reference is more detailled than the product, I see no problem. It is at least an honest measure because it tells if the SAR product is representing the real conditiions and therefore if it’s useful. At least many studies argue like that.

#10

I have gone this training already
it is very helpful but i just want add more comprehensive results after making footprint mapping of urban areas .
Is it good to compare SAR result with Optical data or go for data fusion?

#11

I got your point so if i want to make a accuracy assessment for my ROI then i have to compare with other dataset like comprasion of results of SAR with Optical?

#12

Yes, of course it is good approach for comparing both, data fusion could give you also a comparison, but this will bring us back to this post of our colleague @ABraun source of the post

1 Like
#13

Okay I will keep in mind. I think i should go for comprasion after doing classification on both comparative datatset. is it a good approach?