On Nature Inspired Computation Based SAR Image Segmentation Techniques

Due to the character of high resolution and the fact that it can be used to observe in all weather and all time, Synthetic Aperture Radar (SAR) has been widely used in military areas, resource observation, environment detection, archeological areas, etc. Therefore, the analysis and processing of SAR images have been important research topics. SAR image segmentation is an important step in SAR image analysis and processing and has attracted much attention. This thesis addresses the key issues in SAR image segmentation and makes research and proposes algorithms and technologies for SAR image segmentation. The thesis includes the following research works and new results:1. The problem of SAR image segmentation is converted into the problem of optimization of spatial matrix by combining spatial characters and gray information of SAR image. The segmentation threshold can be automatic obtained by using intelligent searching algorithm. The simulation results show that the algorithm based on spatial matrix can have good performance for artificial objects of images.2. SAR images often have abundant texture information. According to the texture characters derived from the computation of Gray Level Co-occurrence Matrix (GLCM), an unsupervised immune clustering algorithm is proposed based on the immune algorithm. The simulation results indicate that high segmentation resolution can be achieved for images with certain textures.3. This thesis theoretically proves that wavelet transform can reduce the effect of coherent noise and improve the precision of texture analysis and then proposes a segmentation algorithm based on wavelet transform. The simulation results indicate that the wavelet transform based algorithm has high segmentation resolution.4. Hidden Markov Model (HMM) is used to model SAR image. The parameters of the statistical model can be obtained by learning. Distance between images can be computed by using the obtained parameters. A segmentation algorithm is proposed based on the image distance. The simulation results show that the algorithm is robust to texture scale and details and not sensitive to the coherent noise hence and has low complexity and better segmentation.5. A data cube model for SAR image is constructed to describe the characters of SAR image. The multi-dimensional description of the SAR image database and image data mining techniques can help to achieve fast segmentation and object recognition of SAR image.