Image classification network compression method based on convolution kernel shape automatic learning

Abstract

The invention relates to an image classification network compression method based on convolution kernel shape automatic learning, and belongs to the technical field of image processing and recognition. According to the invention, multiple sparse regular constraints are applied to parameters of all positions in a conventional convolution kernel, internal parameters of the convolution kernel are sparse in the network training process, a clipping threshold value is set according to a compression ratio, and then an automatically learned convolution kernel shape can be obtained, so that internal redundant parameters of the convolution kernel can be effectively eliminated. When the invention is applied to an image classification task, the classification accuracy can be ensured, the compression ratio of a network model can be further improved, the parameter quantity and the calculation quantity of the model are reduced, and deployment and application in mobile equipment with limited resources are facilitated.

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 automatic learning of convolution kernel shape. 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 ...

AI Technical Summary

Problems solved by technology

This method belongs to the structured pruning method, and the convolution channel is used as the smallest unit for pruning, but the redundant parameters inside the convolution kernel cannot be removed.

Images