Research On Sonar Image Mosaicing Method Based On GA-BP Neural Network

As the largest unexplored coverage area in the world,the ocean has received great attention from all over the world due to its huge biomass and mineral resources.Due to the physical properties of the ocean,optical imaging instruments cannot provide effective technical support for human exploration of the ocean.Therefore,the acoustic detection device,the sonar system,has become the most powerful tool for acquiring information on seabed targets.It is widely used in modern military and civil activities,and its unique accuracy and stability have become the key equipment for detecting information such as seabed resources and targets.However,due to the imaging angle and distance of the sonar device,the interest target cannot be completely presented in the same sonar image,which is not conducive to manual interpretation,recognition and tracking of interest targets.Therefore,it is of far-reaching significance to splicing the target sonar images collected by the device by matching.In this paper,the sonar image of actual acquisition and computer simulation is taken as the research object.The noise pollution in the sonar image and the matching performance of the sonar image feature are not good.The principle is analyzed and the solution is given to solve the core problem of sonar image mosaic.The work of this paper mainly includes:(1)The basic structure,working principle of BP neural network and the basic methods of BP neural network design are studied.The principle and basic steps and characteristics of standard genetic algorithm are analyzed,which provides theoretical basis for the feature matching work in the following text.The sonar image stitching of the-BP neural network lays the theoretical foundation.(2)For the different sources of sonar image noise,the noise is statistically modeled,and the guided filtering is improved.An adaptive weighted filtering method is proposed.Compared with several commonly used classical image filtering methods,the same group of sonar images are filtered.The comparison and analysis of the processing results verify the effectiveness and accuracy of the proposed adaptive weighted filtering method,and the denoising effect is better.Well,it can effectively remove the noise of the sonar image speckle while retaining the image features,providing effective preprocessing for subsequent stitching work.(3)For the requirements of sonar image splicing related technology in the fields of seabed mapping,seabed search,seabed sediment identification,etc.,starting from the characteristics of sonar image,a feature matching algorithm based on GA-BP neural network is proposed to reduce image features.The wrong match of the points gives the registration result for the sonar image.In terms of registration accuracy,the proposed GA-BP neural network feature matching algorithm is compared with several commonly used registration operators.The results show that the proposed method not only has high stitching precision,but also multi-angle affine transform sonar image.The mosaicing effect is good.