Abstract: Objectives: Side scan sonar is one of the main means to acquire the submarine geomorphology image. Sea bottom line is the most prominent feature of the side scan sonar waterfall image. Accurate detection and tracking of the bottom line is the basis of the fine processing of the side scan sonar data. Methods: The traditional threshold methods and related image feature detection algorithm are difficult to achieve automatic, accurate, and efficient extraction of the sea bottom line due to the interference of water environment noise and scattering sound from the hull, water surface, and other suspended objects in the water. This article fully considers the image edge feature of the sea bottom line and its spatial characteristics that distribute along the track line, thus forming a kind of sea bottom line tracking method with the combining of adaptive edge direction density-based spatial clustering and clustering chain screening. The main work includes three parts: firstly, calculating image edge gradient and direction with a convolution template defined by the first-order derivative of Gaussian function, and non-maximum suppression also used to thin the edge character; secondly, improving the density-based spatial clustering algorithm to direction adaptively by setting the narrow strip search neighborhood and adjusting the long axis of the search neighborhood in real-time according to the direction of the edge gradient; thirdly, constructing the sea bottom line tracking strategy based on the clustered edge chains set, those strategies includes setting experience range, constructing the clustering seed set by threshold method, the long chain principle, the exclusion principle, the symmetry principle, the trend extension principle, the repair principle, etc. Results: Through experimental verification and comparative analysis, the results show that the accuracy and stability of the proposed method are superior to the traditional threshold method, even in the case of common water echo interference such as continuous noise and complex suspended objects, the average single ping detection time is only 0.661ms. Conclusions: The sea-bottom line detection method of side-scan sonar image described in this paper has good stability and anti-interference performance, and has great potential to be a generalized method for onboard data acquisition and post-processing of side-scan sonar.