Due to the complexity of the underwater environment, the unpredictability of submerged objects and the limitation of the detected image quality, the underwater implementation of the simultaneous localization and mapping (SLAM) algorithm based on image information-assisted positioning and navigation has always been a challenging task. This paper proposes a SLAM algorithm based on image sonar salient object detection to optimize the cumulative error of the autonomous underwater vehicle (AUV) inertial navigation system. By adopting the saliency object detection method of image sonar data to enhance the perception of the external environment, it constructs a complex and changeable underwater SLAM system, improving error-free autonomous vehicle navigation. This experiment shows that the extraction of landmark features based on the salient object detection of image sonar can accurately construct an underwater SLAM system and reduce the navigation error of an AUV. It is helpful for long-range AUV detection in marine archaeology, rescue search and hydrogeology.