Sonar Image Matching Optimization Using Convolution Approach Based on Clustering Strategy

Sonar image matching is the fundamental for large-scale seafloor image acquirement and multisource data fusion. However, initial matching results include numerous errors, and current approaches to eliminate mismatches have some drawbacks. Segmenting image for matching does not consider the distribution of feature points, and the existing mismatching elimination algorithms need a predefined coordinate transformation model. To solve the two problems, this letter proposed a method to eliminate mismatches by combining the clustering operation of feature points and convolution approaches. In the experiment, we used side scan sonar (SSS) and multibeam echo sounder (MBES) images to conduct the matching optimization. Experimental results indicate that the correct ration of image matches reaches 64% with clustering strategy, which proved that the proposed method is valid and can be transferable to the matching of other kinds of sonar images.