Refinement curve length of shrimp is an important factor in its weight prediction. Morphological filter was required to reduce branches of the refinement curve in order to get this value. The existence of branches was correctly judged by 3×3 neighborhood analysis of every pixel on the refinement curve. Then the length of refinement curves which contain branches could be calculated by erasing the endpoints to remove branches and average compensation method. The conclusion showed that the correlation was obvious, with the correlation coefficient of 0.894 between its refinement curve length and weight, 0.939 between the size and weight, and prediction correlation coefficient of 0.954 between comprehensive details both of the length and size of shrimp and the weight after modeling. The results indicate that it's promising to increase the predicting accuracy with refinement curve length.