Pulse Coupled Neural Network Methold For Sonar Image Mosaic

With the rapid growth of global population and the increasing shortage of land resources,countries around the world are paying more and more attention to the development and utilization of marine resources.Sonar image is a significant tool for us to explore the ocean.The single sonar image has a small image size.When the target is large,the sonar system cannot capture a complete target at one time.We need to use image mosaic technology to stitch the sonar images with overlapping parts collected multiple times into a clear and complete image.Sonar image mosaic mainly adopted feature-based methods and corner is an important tool for image feature expression.Pulse coupled neural network has achieved good results in many fields of image processing.This thesis attempts to use pulse coupled neural network to extract the corners in the sonar image,so as to solve the problem of sonar image mosaic.The main contents of this thesis are as follows:(1)This thesis analyzes in detail the electronic circuit model of a single neuron in the standard PCNN model and its corresponding mathematical expressions in analog form and discrete expressions.In view of the shortcomings of the standard PCNN model,which has complex structure and many parameters,this thesis improves the standard PCNN model while preserving its synchronization pulsedistribution characteristics.The improved PCNN model has simple structure,few parameters and it is easy to set parameters.(2)This thesis introduces the principles of several classic corner detection algorithms in detail.Then based on synchronous pulse distribution characteristics of PCNN,corner detection algorithm based on standard PCNN,namely Algorithm 1 in this thesis,is proposed.At the same time,corner detection algorithm based on improved PCNN,namely Algorithm 2 in this thesis,is also proposed.(3)This thesis reasonably sets up image mosaic experiments to test the performance of different algorithms.In corner detection stage,Harris algorithm,SUSAN algorithm,FAST algorithm,SIFT algorithm,Algorithm 1 in this thesis and Algorithm 2 in this thesis are used respectively;In corner description stage,the six algorithms uniformly use 128-dimensional corner descriptors;In image registration stage,firstly,the six algorithms use the nearest neighbor sub-nearest neighbor algorithm to calculate the corner pairs participating in the matching between the images to be stitched,and then use RANSAC algorithm to calculate the geometric coordinate transformation model between the images to be stitched,so as to complete the image registration.Finally,the six algorithms uniformly adopt the image fusion algorithm of gradual in and gradual out weighted average to improve the quality of sonar image mosaic.The experimental results show that: this thesis provides two good methods to solve the problem of sonar image mosaic.The corners detected by Algorithm 1 in this thesis have high accuracy and strong adaptability.Algorithm 2 in this thesis adopts improved PCNN model with fewer parameters and less difficulty in parameter setting.This algorithm has strong adaptability and always keeps high matching rate in the experiment of image mosaic.