With　the　advent　of　sonar　technology,　our　understanding　of　the　ocean　has　become　more　comprehensive,　especially　for　deep　sea　biology　and　geology.　However,　the　sonar　image　is　easily　degraded　during　the　underwater　acoustic　channel　acquisition　process,　which　affects　the　later　research　work.　To　this　end,　this　paper　compares　and　analyzes　multiple　saliency　models　and　combines　them　with　PSNR,　SSIM　and　GSIM　to　explore　an　effective　sonar　image　quality　evaluation　method.　Finally,　an　experimental　analysis　on　the　newly　established　sonar　image　quality　database　shows　that　the　difference　of　the　significance　model　in　predicting　human　attention　has　a　performance　gain　effect　on　the　image　quality　evaluation　method　when　fused　with　the　saliency　model.　©　Published　under　licence　by　IOP　Publishing　Ltd.