Image classification network compression method based on convolution kernel of any shape

Abstract

The invention relates to an image classification network compression method based on a convolution kernel of any shape, and belongs to the technical field of image processing and recognition. A conventional convolution kernel in an image classification network is replaced by a convolution kernel in any shape, network training is performed on a data set by using a gradient descent algorithm, and a converged network model can be used for image classification; a conventional convolution calculation process is divided into 1 * 1 point convolution, partial operation in the disassembly process is removed to realize a convolution kernel with any shape, so that redundant parameters in the convolution kernel can be effectively eliminated, the convolution kernel is applied to an image classification task, and the image classification efficiency is improved. The compression rate of the network model can be further improved while the classification accuracy is ensured.

Description

technical field [0001] The invention belongs to the technical field of image processing and recognition, in particular to an image classification network compression method based on arbitrary shape convolution kernels. Background technique [0002] Image classification and recognition is an important topic in the field of machine vision. Early image recognition methods mainly relied on manually extracted features, which had low accuracy and limited applicability to different scenarios. With the emergence of deep learning methods, convolutional neural networks have made great achievements in the field of machine vision such as image recognition and target detection. Deep neural networks can effectively extract advanced semantic features in images, and have been able to achieve recognition beyond human ability. [0003] However, while the network performance is improving, the network structure is becoming more and more complex, and the requirements for the storage capacity an...

AI Technical Summary

Problems solved by technology

This hybrid pruning method can obtain a high compression rate while ensuring the accuracy of the model, but the weight pruning is an unstructured pruning, which will cause irregular sparsity in the network structure, and requires a special sparse storage method or a dedicated Computing unit can exert its performance, and it cannot be effectively applied to general-purpose computing equipment

However, if only the convolution kernel is pruned, the redundant parameters in the convolution kernel cannot be removed, and a high compression rate cannot be achieved.

Images