Research On Underwater Geomorphic Image Denoising Technology In Morphological Wavelet Domain Combined With Spectral Clustering Theory

With the development of Digital Earth and ocean power,the concept of digital ocean came into being.Digital Ocean is to establish a three-dimensional,network-based,continuous and comprehensive observing ocean system,and to obtain marine geological,biological,physical and other massive data.Underwater geomorphology exploration is one of the urgently needed technologies to support digital and scientific ocean construction,and imaging sonar,as the most effective means to directly analyze underwater geomorphology,has been widely concerned.Because of the complexity of underwater environment and the influence of man-made interference,sonar image is usually characterized by low resolution and low contrast.In order to acquire the underwater geomorphologic information accurately,the first step is to denoise the sonar image.Based on the theory of wavelet transform and spectral clustering,combining the characteristics of sonar image,this paper studies the image denoising technology,and discusses the denoising algorithm suitable for sonar image.Firstly,the basic structure of side-scan sonar system,the imaging principle of sonar image and the influence of noise in imaging process are analyzed.In this paper,the model of ocean environment noise which has the greatest influence on the quality of sonar image under the deep sea condition is analyzed.According to the Gaussian property of the ocean environmental noise,a wavelet threshold de-noising algorithm is used to simulate the noisy sonar image by adding Additive white Gaussian noise to the sonar image,the morphological median wavelet transform is constructed,and the improved morphological median wavelet de-noising method is used to simulate the noisy sonar image,the highfrequency coefficients(including noise and image details)after morphological median wavelet decomposition are analyzed.In order to improve the denoising performance,spectral clustering is used in the clustering method,in this paper,spectral clustering is used to classify the high-frequency coefficients,and a denoising method combining spectral clustering with morphological midpoint wavelet is proposed.The experimental results show that the morphological midpoint wavelet algorithm combined with spectral clustering theory has more advantages in edge-preserving denoising because of the nonlinear characteristic of morphological wavelet.With the addition of spectral clustering,the de-noising performance of sonar image under low SNR is further improved.The feasibility and validity of morphological midpoint wavelet algorithm combined with spectral clustering theory are proved.