多波束声呐图像是进行海底底质分类与目标识别的主要数据源之一。原始反向散射强度经过改正处理后仍存在残余误差,需要进行图像镶嵌以削弱残差的影响,构建大区域声呐图像。本研究针对多波束声呐图像镶嵌中的拼接线消除与信息综合问题,提出一种在NSST域内结合PCNN与混合加权的图像镶嵌方法。首先,对单条带声呐图像进行NSST分解,再对重叠区低频项按照混合加权进行镶嵌,高频项按照PCNN进行镶嵌。实验结果表明,在拼接线消除方面,镶嵌后条带拼接线两侧声强均值差由-6.18降为-1.29dB;在信息综合方面,镶嵌结果的信息熵相对于混合加权法提升2.5%,空间频率提升31.6%。本研究方法能够在削弱拼接线的同时,有效综合相邻条带纹理信息,实现多波束声呐图像的镶嵌。

Multibeam sonar images are one of the main data sources for seafloor sediment classification and target detection. However, there are still residual errors in backscatter intensity after correcting, and image mosaics are needed to reduce the impact of residual errors to build a large-area sonar image. This paper proposed an image mosaic method combining pulse coupled neural network(PCNN) and blending weighted method in the non-subsampled shearlet transform(NSST) domain to eliminate seam lines and fuse information. Firstly, the single sonar image was decomposed by NSST and the low-frequency terms were fused according to the blending weighted method while the high-frequency terms were fused according to PCNN. The results show that in terms of seam line elimination, the mean difference of backscatter on both sides of the seam lines after the mosaic reduced from -6.18 to -1.29 dB. In terms of information fusion, the information entropy in the mosaic result was increased by 2.5% compared with the blending weighted method, and the spatial frequency was increased by 31.6%. The proposed method can effectively fuse the texture information of adjacent strips while weakening the seam lines, thus realizing the mosaic of multibeam sonar images.