Soft supervised classification: An improved method for coral reef classification using medium resolution satellite images

This paper evaluates two soft/fuzzy supervised classification algorithms for coral reef mapping using Landsat-8 satellite images. The work is unique in its nature, since it introduces for the first time supervised soft pixel classifiers for coral reef mapping. A comparison was made between the fuzzy maximum likelihood (FML) and a supervised version of the fuzzy c-means (FCM) with traditional hard maximum likelihood classifier (MLC) and support vector machine (SVM). The comparison was based on benthic substrate data collected in a study conducted on reef site in Australia. Subpixel confusion uncertainty matrix (SCM) and Root Mean Square Error (RMSE) are used for accuracy assessment of soft classification. Experimental results demonstrate the accuracy and advantages of the proposed soft supervised classification algorithms in comparison to traditional classifiers for coral reef mapping.