HI please I have classified my satellite images, done change detection and produced my Land Surface Temperature maps, but I have no access the landuse data and the national topomaps of the study area.
How do I carry out my accuracy assessment for the classification and change detection?
How can I validate the LST?
Thanks
http://www.utsa.edu/lrsg/teaching/ees5083/l9-classaccuracy.pdf
http://www.ethanpublishing.com/uploadfile/2015/1231/20151231033410270.pdf
use enhanced FCC FOR validation ( accuracy calculation)
calculate kappa and overall accuracy.
http://waset.org/publications/7672/satellite-data-classification-accuracy-assessment-based-from-reference-dataset
In satellite images classification and change detection, one will need reference points. One set of reference points is used to train satellite image classification algorithm and other one to calculate accuracy of classified image. Two common methods to get such reference points are:
1) Using already developed landuse/landcover map of study area as a reference (this method is adopted when study area is inaccessible or there is a time limitation)
2) If study area is accessible then best way is to undertake GPS survey of area to collect reference points. This is the best method which a remote analyst considers to get best results. The process is also know as ground referencing or ground truthing.
Once you have reference points then you can develop error matrix and calculate kappa coefficient. Moreover, accuracy of change detection (post classification comparison method) will depend on accuracy of classified images along with accuracy of images registration given that both images are taken from same sensor with same specs.
for validation of any change detection's analysis there is need for accuracy assessment to justified the changes discover of both land use or land cover, the analysis TOOLS is there in ERDAS IMAGINE remote sensing software. to have correct value of KAPPA MATRIX.
Thank You very much all for the help.
Firoz Ahmad sir, I used the reference data according to Mohd Hasmadi Ismail in the following attachment, my overall accuracy is 99% and the other is 99+ % Please is it normal to have high accuracies as high as this?
http://waset.org/publications/7672/satellite-data-classification-accuracy-assessment-based-from-reference-dataset
Asid Ur Rehman Sir, In my case I neither have the developed LULC data nor is the area accessible. so based on the suggested paper by Firoz Ahmad and the attached link, I collected the reference points from the satellite image but i got very high accuracies. Could GPS coordinates (x,y) be helpful?
I would also like to be clear about the accuracy of the classified image and the change detection , are they the same? I mean if my overall accuracy for classification is satisfactory does it mean my change detection result is valid?
https://www.youtube.com/watch?v=FaZGAUS_Nlo
Tanimu Isah Sir please the analysis tool you mentioned, is it for calculating change accuracy? Also is it also available in other software like Envi, ArcGIS or Ilwis,? I dont have Erdas
Hafsat,
Image classification and accuracy assessment is a broader topic. Value of accuracy assessed is really based on number of classes we extract from a satellite image. If one goes for detailed classification like extracting 8-10 classes from LandSat image then accuracy of 70-79 percent is considered good while value range 80-85 percent is considered very high and practically it is very difficult to achieve an accuracy of 90% or greater for such detailed classification of moderate resolution satellite images. Yes, GPS observations of physical objects are considered accurate for good classification but in your case it is not possible as area is inaccessible.
As far as accuracy of change detection is concerned, I did never take literature review on it. In my opinion if you are performing post classification comparison method for change detection then its accuracy should be equal the minimum accuracy value among accuracies of classified images used in change detection.
Asid Ur Rehman Sir, thank you very much. I got GPS coordinates of the area, but they are points where oil spills occurred (longitude and latitude).
Perhaps the high accuracy has to do with the number of my classes. I have only four but I will check that.
Thanks
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing