Development of a web-based blind test to score and rank hyperspectral classification algorithms

Remotely sensed hyperspectral imagery plays an important role in land cover classification by supplying the user with additional spectral data as compared to high-resolution color imagery. The web application described in this paper enables users to test their classification algorithms without the risk of bias by withholding the majority of the true classification data and only providing a small section of the truth data to be used for training user algorithms. After downloading the dataset, users run their classification algorithms and upload their results back to the web application. The blind test site automatically scores and ranks the uploaded result. The Classification Blind Test web application can be found at: http://dirsapps.cis.rit.edu/classtest/.