Abstract:

Aiming at the problem of low registration caused by image rotation or background transformation in large-scale image classification processing, a convolutional neural network image classification method based on edge enhancement was proposed.This method extracted image features through the VGG19 network model, and used cosine similarity for image classification judgment, used edge enhancement to highlight the edge features of image subject, and reduced the impact of image rotation or background transformation on the classification performance of the VGG19 network.Experiments prove that this method can effectively improve the similarity between the same subject rotated image and background transformed image and the original image, and is suitable for the classification of various images.

Key words: convolutional neural network, VGG19, cosine similarity, edge enhancement