POLARIMETRIC SAR (SYNTHETIC APERTURE RADAR) FOR LAND CHANGE DETECTION

Abstract

Land cover change detection, carried out by using remote sensed images, is a challenging problem having a key function in many practical application areas such as deforestation assessment, damage assessment, disaster monitoring and urban expansion. Growing importance of change detection for environmental assessment is major impetus to development of remote sensing based change detection. In this work both unsupervised and supervised change detection technique is used for change detection. In unsupervised change detection we do not have any prior need of information about ground terrain feature. In this work algebraic based unsupervised technique were used. Image differencing and image ratioing are two algebraic techniques used in dissertation. As unsupervised change detection is easy in implementation but this method has lack of information because we can't access information about terrain features. These methods only inform about change and no change of a particular pixel. Speckle noise in both supervised and unsupervised change detection is cause of degradation of SAR image so speckle noise needs to be done away with. Speckle noise can't be reduced to zero level but can be reduced significantly by Multllooking and spatial filtering. For spatial filter, Lee filter is used. Another major confrontational task is to decide thresholding. In this work histogram based method is used for thresholding that is standard deviation. In supervised changed detection technique post classification change detection is used. For that first pair of polarization and classification technique for maximum classification accuracy is achieved. For that purpose four polarization states were classified by two unsupervised ISODATA and K-means and two supervised Maximum likelihood and Mahalanobis techniques. For computing change detection statics HH_I-IV polarization of Maximum likelihood is selected because it gives maximum accuracy of classification. Most advantage of post classification change detection is having access about classes which are terrain feature. One thing one has to take into consideration is that image must be properly geo-referenced and co-registered otherwise pixel mismatch would cause inaccuracy in change detection