A real-time and adaptive segmentation algorithm based on sonar raw echo data is proposed for small underwater targets acquired by side-scan imaging sonar. The algorithm uses the statistical characteristics of underwater echo distribution and adopts an adaptive threshold determination method based on Weibull distribution model to achieve the initial segmentation of targets; On this basis, a region growing method based on the characteristics of bright region echo is proposed to obtain accurate segmentation results; After that, Hough detection method is used to merge the segmented region, and small target physical size information is used for size detection to eliminate false targets with large size such as ridges and rocks. The experimental results show that the speed of the detection method using only target physical size can reach 3 s/frame, the target recall rate can reach 100%, and the false alarm rate is only 34.62%