An autonomous underwater robot path planning algorithm for underwater dam surface apparent crack detection was designed based on bio-inspired neural network in order to acquire continuous local images of long cracks on underwater dam surface. The algorithm identifies the crack orientation and calculates the raster it points to, and then applies the pointing result to the path decision by means of an activity gain. It is confirmed by simulation tests that the algorithm can continuously acquire local images of long cracks with a greater degree of continuity than the paths planned by the cow plowing method, while ensuring full coverage of the detection object.