Image classification method based on lightweight residual network

Abstract

The invention discloses an image classification method based on a lightweight residual network, and mainly solves the problems of excessive network model parameters and insufficient utilization of image feature information of the existing image classification method. The implementation scheme is as follows: obtaining a training sample set and a test sample set; changing a traditional residual unit, establishing five different lightweight unit blocks, and sequentially cascading the five different lightweight unit blocks with a full connection layer and a classifier to form a lightweight residual network image classification model; training the constructed image classification model by using the training sample set and adopting a back propagation algorithm; and inputting the test sample set into the trained lightweight residual network image classification model to obtain a classification result. According to the method, high image classification accuracy can be obtained in an image classification task, the network model parameter quantity is reduced, the network operation speed is improved, and the method can be used for recognition of human faces, traffic scenes and medical images, image retrieval and automatic classification of photo albums.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to an image classification method, which can be used for recognition of human faces, traffic scenes, and medical images, image retrieval and automatic classification of photo albums. Background technique [0002] Image classification is a technology that obtains features that can represent the content of the image from an image, and then uses a computer to analyze the features to obtain the category to which the image belongs. In the actual image classification task, several image category information are often given in advance, and the computer will give the probability value of the image belonging to each category, and finally determine the category with the highest probability value as the category to which the image belongs. Therefore, in the image classification task, the features learned and extracted from the image are the key factors to determine the image cat...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem of too many parameters in the network model of the above method, the researchers proposed a lightweight network that reduces the computational complexity by changing the convolution method, so that the effect of fewer parameters and faster speed can be achieved under the premise of ensuring the accuracy of image classification. Most of the lightweight networks rely on a single convolution method, and there are still problems with too many parameters, and there is no obvious improvement in the ability to express network features.

[0006] In view of the lack of information exchange between different groups of channels in the above method of group convolution, the researchers use the channel shuffling operation to strengthen the interaction of feature information. This operation can promote the fusion of information between channels, change the channel dimension, and enhance the expression ability of network features. , however, the simple channel shuffling operation does not reduce the number of parameters in the network, and the speed of the network has not been improved

Images