Partial Differential Equation Method For Sonar Image Segmentation

Today's development in the marine sector is increasingly important to a country and the world,and sonar technology is the main technology for its development.For the obtained sonar image,it is generally required to perform sonar image filtering processing,sonar image segmentation processing,and final sonar image feature extraction and target recognition processing in sequence.The results of the sonar image segmentation process in these steps are critical to the final result of sonar image recognition.Only by filtering the sonar image can the sound image segmentation be better.As for the application of partial differential equations(PDE equations)in this paper,it can make complex and variable image processing problems into more intuitive mathematical problems in the field of image processing.It is suitable for complex and variable sonar image processing.The anisotropic diffusion model(proposed by Perona and Malik,abbreviated as PM model)has advantages in both noise removal and edge detail preservation,which makes PM model filtering a research hotspot in the field of image filtering.However,PM model filtering still needs to be improved in terms of noise removal and edge preservation.Combining the filtering performance of the diffusion function with the local information parameters of the image,this paper constructs a method of adaptively selecting the diffusion function according to the local information parameters of the image,and proposes an improved anisotropic diffusion model for sonar image filtering.The model not only can solve the phenomenon that there are a large number of isolated noise points in the classical PM model filtering,but also can balance the edge preservation of the sonar image and the noise removal of the sonar image.Image segmentation is often an area in which images are represented by different gray levels that do not overlap each other.The local binary fitting model(LBF model)can rely on the local details of the image to achieve accurate segmentation of grayscale non-uniform images.However,the local characteristics of the LBF model make it more sensitive to the position of the initial zero level set curve.Aiming at this defect in the LBF model,this paper proposes a global and local superposition of the fitted energy active contour model.It consists of a locally fitted energy term,a global fitting energy term,and a variance term for the fitted and mean values for each region.Therefore,the model is not only insensitive to the position of the initial zero-level set curve,but also can segment the boundaries with similar gray levels.