SAR Image Change Detection Based On Spatial Threshold Segementation

Change detection of remote sensing images is based on identificatoin and analysis of scene changes by utilizing multi-temporal images of same and other auxiliary data.Synthetic aperture radar(SAR)imaging system works independently of the weather and sunlight conditions and has incomparable advantages over other optical remote sensing.Therefore,SAR image is an important data source for change detection.Currently,SAR image change detection plays an important role in many fields,such as environmental monitoring,agricultural survey,urban research,disaster monitoring,military investigation,etc.Especially when natural disaster occurs,an efficient change detection technique can avoid or reduce the injuries and death of humans as well as the loss of belongings.Change detection methods have the following steps: image preprocessing,difference image generation and difference image analysis.And the analysis of the difference image is crucial for change detection.Threshold methods are commonly used in the analysis of difference image due to their simplicity and high efficiency,making change detection faster and more efficient.This thesis contributes to the body of knowledge mainly in the difference image analysis with the modification of threshold-based technique,which can be summarized as follows:1.Threshold-based methods are usually adopted to produce a high false alarm rate when change areas occupy a low proportion in the entire imaging scene.A novel change detection method for SAR images is proposed based on an iterative Otsu method.Firstly,neighborhood-based ratio operator is improved.By utilizing modified neighborhood-based ratio operator and local information,the difference image is generated with a good performance on the suppression of speckle and preservation of detailed information in SAR images.Then,the difference image is segmented via the iterative Otsu method.With multiple iterations,a target region is extracted where the distribution of changed pixels and unchanged pixels are relatively balanced.With the help of the Otsu method,an optimal threshold of the region is calculated and subsequently used to binarize the difference image by comparing the grayscale of pixels with the optimal threshold.Finally,the result of change detection is obtained.The experimental results on real SAR images with small-area changes demonstrate that the proposed algorithm has an outstanding performance in detection accuracy.In order to verify the universality of the algorithm,the corresponding experiments were carried out on the real SAR images data sets under the condition of images with nonsmall area change and images with different noise levels,our method also outperforms other threshold-based methods in both detection accuracy and robustness to noise.2.For the relatively poor ability of existing conditional distribution models in KI threshold method to fit the difference image,a SAR image change detection algorithm based on KI threshold segmentation method based on K distribution is proposed.The proposed algorithm utilizes K distribution to fit the histograms of changed pixels and unchanged pixels.K distribution model is widely used because it has the multiplicative fading statistical characteristics and exhibits a good ability to fit heterogeneous areas of SAR images.However,it is complicated to estimate the parameters of K distribution.From the perspective of the algorithm efficiency,a fast method using the moments of second and fourth order is utilized to estimate the parameters of K distribution.The experimental results show that the proposed algorithm has an evident improvement on detection performance in comparison of existed KI threshold method.