The invention belongs to the technical field of hydraulic engineering, and discloses a hydraulic-engineering crack extraction method and system and an information processing terminal. Original hydraulic-engineering pictures are collected, preprocessing of artificial denoising, image enhancement and the like is carried out on the pictures, and the pictures are divided into two sets of a picture setwith cracks and a normal picture set; a crack marking method based on correlation analysis is designed to accurately mark the cracks of the picture set with the cracks; a multi-layer convolutional neural network is constructed, the picture set with the cracks and the normal picture set are respectively learned, convolution kernels of high discrimination degrees are selected, and different weightsare set for convolution kernels according to discrimination ability to form a crack identification model; and the trained crack identification model is utilized for crack extraction on an obtained picture, and specific parameters of cracks are generated. According to the method, a method of setting the different weights for the convolution kernels is employed, functions of the convolution kernelsof the high discrimination degrees can be effectively improved, and thus accuracy of crack identification is improved.