Aiming at the problem of monitoring and identification of dike piping phenomenon, a method for dike piping identification based on the YOLO model was proposed. A Piping YOLO model based on region of interest (ROI) extraction was proposed to improve the performance of YOLO v3 network by introducing improved residual block and activation function of replacement model. After the ROI was extracted, the two-dimensional principal component analysis method was used to extract the characteristics of piping phenomenon, which was used as the input of multi-weight neural network to realize the classification and recognition of piping state through training. The experimental platform of dike piping was built, and the data set was established to verify the effectiveness of the proposed method. The results show that the proposed method can effectively identify the phenomenon of dike piping, and has a certain application prospect in the field of unmanned inspection of dike piping.