MULTITEMPORAL UNSUPERIVISED KERNEL-BASED MULTIPLE CHANGE DETECTION

The change detection (CD) is a process utilized the differences procedure between the two different phenomena at the time of measurement. Nowadays, with the advancement of remote sensing technology, it is possible to obtain spectral band in satellite image format in hundreds of wavelengths. These images are known as hyperspectral images, which can be identified with a very high accuracy changes. Many past years, caused to improve temporal resolution and quality of hyperspectral imagery, there is a huge interest in extraction change information using of multitemporal images. This paper presents a hybrid semi-supervised CD method for land use monitoring by utilizing multi-temporal hyperspectral images. By incorporating unsupervised clustering method, kernel image differencing (KID), this method can detect any changes. The proposed method is implemented in three main phases: (1) optimization kernel parameters using of training data, (2) the corrected data by using of KID converted data to similarity space, (3) the third phase is to make a decision about the nature of pixels and extracted ‘multiple-changes’ by unsupervised clustering. The main advantage of the proposed method is being semi-supervised with simple usage, low computing burden, and high accuracy. The efficiency of the presented method has been evaluated by using Hyperion multi-temporal hyperspectral imagery. The used dataset related to agriculture fields located in southern Iran, Khuzestan. The results of two real data sets show high efficiency and accuracy with low false alarm rate by using the proposed method compared to common CD methods with overall accuracy more than 85%, kappa coefficient of 0.75.