Hello. I have fine, medium and coarse resolution satellite images with me. I wish to opt for change detection studies, classification methods plus some spectral index ratio methods. I have enough material to generate the outputs but not a single ground truth/field observation to validate those results. I wish to calculate area covered by a particular feature of interest and delineate it from others and also measure its temporal variation. So how do I go about it? How can I validate my digital image processing results with the actual ground values? Kindly help me out. Thank you all. Cheers!!
Why do not you split your training data into 2 parts: one for classification and other for validation?
Attached examples of change detection by cover classification methods for your consideration. Classification methods are superior to spectral index methods in my experience.
Article What Drives Conversion of Tropical Forest in Carrasco Provin...
Article From monospecific to mixed forest after fire?: An early fore...
Article Forecasting the pattern and pace of Fagus forest expansion i...
Dear Hein Sir, thanks a lot for your suggestions. The research work you have carried out is interesting. I will certainly try to follow the track. However, here I am more concerned about the ground observations/field data. I have ample amount of satellite data plus I know the methods to apply to exploit that data to its capacity. But I do not have any field data to validate those results. Here, I would like to know:
1. From where can I get free ground data? Is there any library system / organization which provides ground data (global coverage) free of cost?
2. Can I validate digital image processing results based on another processed satellite imagery? If yes, what is the procedure and accuracy of the same?
3. Any other method?
Also kindly enlighten me about the uncertainty involved in it.
Thank you once again sir!
Thank you for your time and consideration.
Cheers!!
@Parag, why not try Google Earth Pro? It is free for use & I can extract some ground truth from it. There is also a historical view so one can go back in time.
Enjoy
You can use Virtual verification method...which recently used by molan and chandan..even im having same problem..so i thought of adopting this method..
Dear Parag, what area are the images of? In many locations we find organizations/institutions with complimentary if not similar research. A collaboration or data sharing would work in your case. Since you have very high resolution, GE may not add much value.
A proper validation for change detection (or any other type for that matter) should of course count with proper ground truth.
However, in case you do not have it the option of using other satellite data (as has been suggested) is fair, as long as it is a trusted source, but then your best performance will be limited to that sensor's performance, and it might result in hindering your algorithm's results.
Another option, in your case, is to try and locate some reference surfaces of known time bahaviour, e.g. stable in time or of known seasonality. For this, the use of Google Earth, or any other high spatial resolution map would be fine. You could also go to Corine Land Use Map.
If any stable surface reports variance in your data, any variable surface would suffer AT LEAST so much, so you can give an estimate of confidence (minimum error commited, real error will be higher but impossibe to determine without ground truth or additional knowledge on the surface behaviour). In any case make sure that your reference surface has a similar 3D structure than your target ones in order to minimize differences due to changes in illumination. Of course, I assume you have taken into account topography in your processing for any proper change detection.
Hope this helps
You can use some statistical methods (at least two) which have already been implemented along with ground truth data. So in case you don't have ground truth data, you can still use those statistical method by refering that particular paper.
Thank you Mr. Hristo, Ms. Kavita, Mr. Zhang and Mr. Kenneth for your suggestions. I will try to adopt your valuable suggestions according to my needs. Cheers!!
Mr. Ramesh, I will certainly go through the paper. It sounds like an interesting and reliable option. Thank you!
I manually generated ground truth in the following publication. It was proper for validation.
Article Automatic building extraction in dense urban areas through G...
take control point from your area as much as possible and cheek whether it correct or not with your control point data.
- Badges
- Science topic
- Similar topics
- Clinical Pharmacology
- Dosing