Change Detection Of Harbor Ship Target Based On Structrue Information

Change detection technique in remote sensing images has been widely used in many fields such as earth resource and environment monitoring, battle situation analysis, strike effect evaluating, etc. Harbor ship target change detection can be used to learn the ship change in time, which is very important to monitor ships in seaport, apperceive battle trend and analyse the enemy strategic attempts. So far, quite a lot of change detection methods have been designed. According to the level of the information being used, these algorithms can be divided into pix-based change detection and feature-based change detection. Image structure information is one of image features, and it can not be easily affected by the change of lumination and other noises. Meanwhile, it's also the main clue of the man-made target change detection in high-resolution remote sensing images. Harbor ship target has apparent structure feature and abundant structure information available. Our investigations are mainly carried out around the harbor ship target change detection based on structure information.In this thesis, image structure information is deeply studied, and its definition is given. Through practice experience and theoretical analysis, we divide image structure information into two kinds. One is the structure information based on local neighborhood which is extracted from finite-scale neighborhood and reflect the image content of this local area, such as ET vector, differential invariants, etc. The other is global structure information, which is used to reflect the feature's geometric structure relationship globally, such as target contour, framework, etc.We propose a harbor ship target change detection method based on local neighborhood structure information, and adopte a new conception named"rough segmentation, refined detection". Firstly, remote sensing images are segmented by"Canny"edge detector in order to reject the non-structural change regions. In this step, no much attention is paid to the accuracy. Then, multi-resolution and multi-orientation wide-band Log-Gabor filters are employed to extract local neighborhood structure information to realize change detection.Based on the second structure information, we propose a harbor ship target change detection method which aimed at high-resolution remote sensing images. Firstly, we adopte a new sea-land segmentation approach to obtain the global structure information, and then XOR operation and mathematical morphology processing are used to realize change detection.Experiment results of simulated data and field images show that the methods proposed in this thesis are robust to noise and also more efficient than other approaches.