Hello Everyone,
I really need some expert advice in my project. I have two objects (permanent and intermittent specifically flood water bodies) that have very similar spectral characteristics. For my project, I am using Landsat-5 TM remote sensing images. Could you please suggest me some statistical methods that are able to distinguish or extract some information of two highly correlated objects. Any suggestions would be highly appreciated. I will require some distinct methods to extract information to train my classifier.
If you want a good classifier: Random Forest
Belgiu, M., & Drăguţ, L. (2016). Random forest in remote sensing: A review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 114, 24–31. http://doi.org/10.1016/j.isprsjprs.2016.01.011
Thanks very much Oliver. I am already reading the Random Forest Classifier. I will definitely look at the paper. But I am not sure whether I would have to define some specific parameters as my training parameters that train the classifier? It is hard to distinguish between flooded water or intermittent water from permanent water bodies (river, lakes, reservoirs etc).
Thanks Gahul for your suggestion. But in my case I do not thing ANOVA is a good method. I used this method before. ANOVA can only tells me the variance degree between the variables or in my case two different class types. I need some parameters that can able to collect some good data to differentiate between those two.
Yes, I recomend you train the classifier by using a good sampling. It is a critical point. Furthermore, there are two paths you can take: Pixel-based analysis and Object-Based analysis. OBIA is very useful when your object have a lot of spectral similarities and need other parameters for train the clasiffier (e.g. shape, topology)
You are right Oliver. I have faced a lot of problems as doing Pixel based sampling in my previous experiments. OBIA might be a good choice to my sampling work.I have started reading the Journal paper you have recommended and thanks a lot again, it is really a good reference for my further analysis. I really appreciate your help.
Dear Chandrama,
I do not know what you mean by distinct methods to extract information to train my classifier? which classifier? In addition, since they are water and they have similar characteristics but they are either permanent or intermittent in this case I would use different LANDSAT 5 TM images for different dates to identify permanent form temporary (temporal analysis). In addition, I would use NDWI "Normalized Differences Water index" to identify flooded area.
see the following paper:
Suwarsono, Jalu Tejo Nugroho, and Wiweka, IDENTIFICATION OF INUNDATED AREA USING NORMALIZED DIFFERENCE WATER INDEX (NDWI) ON LOWLAND REGION OF JAVA ISLAND, International Journal of Remote Sensing and Earth Sciences Vol.10 No.2 December 2013:114-121
Thank you for your suggestions. Awad. By stating "distinct method to extract information to train my classifier" I would like to say that if I am going to use a machine learning technique, eg. Random Forest Classifier, I would definitely have good training samples from both of the classes. Now in every classification, selection of training samples is a tough step. Now to select flood and permanent water features I am searching for some statistical method like mean, median or skewness that I can apply to select the training sample. then I can use the classifier by using those training samples. Thanks for the journal paper suggestion, I will go through it definitely.
- Badges
- Science topic