The invention relates to the technical field of image processing and in particular to a method for identifying and classifying high-similarity images. The method comprises the steps of firstly, extracting a main body object in an image through a partitioning algorithm to realize appropriate separation of an image theme and a background so as to prevent the image background from interfering in classification of the main body object and improve the performance of accurate classification of the image; decomposing the image into different frequency domain sub-graphs by use of wavelet transform ina data preprocessing stage, and extracting characteristics of each frequency domain sub-graph; then taking output of wavelet transform as input of a convolutional neural network, performing update optimization on a network weight parameter through multiple trainings, and acquiring a neural network parameter with a highest classification accuracy rate to obtain an optimal classification model. According to the method for identifying and classifying the high-similarity images, through combination of the wavelet transform and the convolutional neural network, the precise identification and classification for the images can be realized, and the method can be used for image processing.