Change Detection in SAR Images Using a Modified Genetic Algorithm

This letter introduces an unsupervised method forchange detection of SAR images using a modified GeneticAlgorithm. This method is based on searching the optimummask in the solution space using evolutionary concept. Thealgorithm is implemented on the difference image computed bydifferencing of corresponding pixels of two multi-temporalsatellite images. To manage the time of implementation, weused a small size moving window to select sub-images ofdifference image. Each sub-image is partitioned into twodistinct parts, named changed and unchanged, using the binarychange detection mask realization from the Genetic Algorithm.The weighted sum of Mean Square Errors (MSE) for thechanged and unchanged areas is used as a cost value of thecorresponding binary mask of selected sub-image. The finalchange image is derived by merging these small size changemasks. To achieve the efficiency of the proposed method weused two sets of test images. The simulated images and the realones acquired from ASAR sensor of Envisat satellite. The realdata shows the coast of Miyagi prefecture, where a hugetsunami swept away houses and thousands of people on March2011. The qualitative and quantitative results show saving andmanaging of time besides preserving proper accuracy.

کلیدواژه ها:

Change Detection ، Genetic Algorithm ، MeanSquare Error ، sliding window ، Synthetic Aperture Radar