Assessing the accuracy of image classification algorithms using during-flood terrasar-x imagery

Extracting and determining floods affected zones is one of the most crucial stages in floods hazard management to reduce damages caused by floods. Assessing and determining the accuracy of image classification algorithms is essential in producing accurate flood hazard maps. This study put forward the application of Remote Sensing and GIS computer programs to carry out a comparative analysis of three image classification algorithms: neural network, parallel-pipe and minimum distance to test which technique best classifies the 2010 during-flood TerraSAR-X image of Perlis, Malaysia. Confusion matrix was calculated to assess the accuracy of each algorithm. The best result of the flood extent extraction model is from the network algorithm classification with an overall accuracy of 99.7661% and a kappa coefficient of 0.9862. These findings could be used to assist the Government to design appropriate measures to safeguard the lives and properties of the residents of Perlis.