An application of image change detection-urbanization

The goal of this paper is to develop an efficient method for finding changes in land areas that undergone changes over a period of time. Land monitoring and observation is such an area that is both important and time consuming task of remote sensing applications. Land changes often occur due to the seasonal changes of land covers, deforestation, natural disasters and many other factors. Existing methodologies do the change detection procedure only between two images taken at same area at two different time instances. A method is proposed for finding changes between images of same set occurred at various time intervals may be between years or various dates. The image is allowed for pre-processing using multiple steps to obtain a clear and an efficient pre-processed image. The spatial domain Image registration methods say transformations are supported in the proposed method. Unsharp Filters are used to remove the noise and to highlight the low intensity regions. Selections of Control Points are included in the Pre-processing step itself to choose a particular region from those input images. The histogram of the input image is used for selection of threshold and the objects in the image are detected. The objects in the detected image are compared with the base image objects to obtain a difference image. Image segmentation is done on edges by Canny edge detector as this is known for its popularity. The selected regions are fed up into neural network tool where Classification training and validation of the data is performed and mean square error is considered as a performance measure. The performance parameters like true positive, false positive, true negative and false negative are obtained by comparing the change detected image with the ground truth image. Graphical representation of training state such as gradient and validation check is also performed. ROC is plotted against false positive and true positive values. The experimental results determine the efficiency of the proposed method by testing it with different data sets.