Hello. I'd like to know if there's any way to validate digital image processing results without having to take ground truth data. Example: I have processed Landsat-8 imagery and I want to validate it using another processed satellite imagery. Is that possible? If it is, then what is the best qualifications of the satellite imagery for validation itself?
Thank you very much.
No way
The goal is to accuracy the information with the ground land cover
No you can not. The ground truth is required to set training points for the software to classify the image. Otherwise noise can be there and you can not be able to segregate the pixel of two object of similar features...
No way
The goal is to accuracy the information with the ground land cover
In my opinion, you can not! Real field data is needed to classify other parts and accuracy assessment.
Field data is the best validation option, but there are many instances where authors use higher resolution imagery and manual photointerpretation to validate classification accuracy; this IS an acceptable practice in the field of remote sensing (see PDF for more). Depending on the scope of your study, it may be logistically impossible to collect enough field data to validate a classification workflow. Think about what you are trying to classify, and at what spatial resolution those classes/objects are very obvious to the human eye. If you are classifying Landsat imagery (30m pixel resolution), you would ideally want to use another imagery source with a 10m or less pixel resolution. Keep in mind that generally, high-resolution imagery is relatively new, so if you are classifying Landsat imagery pre 2010, you may have a hard time finding higher resolution imagery for validation purposes. Some possible missions to use include Quickbird, RapidEye, Sentinel.
Hi,
Frankly this question is a non-starter. Anybody trying that is strongly discouraged.
As has also been responded by others before me, ground-truth validation of any image interpretation is UNAVOIDABLE if reliability and reproducibility is to be achieved.
Having said that, the only one 'exceptional' situation that can bypass ground truth (hypothetically) is when there is authentic 'map' data available for comparison. However, this exception comes with a caveat that the map data must be of a scale and resolution comparable to your image data. Unless that can be established, this exception should not be tried.
The other part of this question also is 'how much ground truth data is required?'. A thumb rule for this is that the ground-truth must cover all 'variable' or 'doubtful' classifications in the image interpretation.
All the best.
Okay, If the researcher can not validate digital image processing results without ground truth data, how then can one validate digital image processing results of the year 2017 in the year 2020 ??
It is good to validate results obtained from digital image analyses, however, in case where the study area is not accessible, other higher resolution digital information, as well as virtual interviews or surveys could be an alternative way.
You may be right...but remote sensing technique is used only when the area is not accessible or study of a large area is required to be carried out. On the other hand virtual interviews and surveys are themselves the ground truthing carried out by the person whom one has to interview.
In agriculture, there are certain crops which shows least difference in the signatures such as wheat and barley. So in such cases you have to analysed the data by using ground control points chosen for ground truthing....
Field data and reference map are important or main reference for validtion. Before attempting few spot specific field
validation, some conceptual validation may also be attempted which would give first level confidence on the results.
These validations could be scene driven and no field information is required.
- Badges
- Science topic
- Similar topics
- Geoscience
- Remote Sensing