The invention discloses an underwater dam crack detection method based on local and global clustering. The method comprises a step of collecting an underwater dam surface image and transmitting the image to an image database, a step of preprocessing the image, initially equalizing image background illumination and enhancing a target area, a step of processing the image and equalizing the gray level of the image and then realizing image binarization by a binary threshold segmentation method, a step of extracting an image block feature and calculating a two-dimensional feature space through a cluster analysis method to obtain an image block containing a crack, a step of extracting all connected domains, taking each connected domain as a sample, extracting the features of the connected domains to form a three-dimensional feature space, and detecting the crack through the cluster analysis method, and a step of positioning the detected image containing the crack to a dam and thus determining an area where the image containing the crack is located. According to the method, the automatic detection of cracks on the surface of the dam below a water level is achieved, time and labor are saved, the cost is low, the non-destructive detection can be achieved, and the requirements of accuracy and real-time performance are satisfied.