Target classification method and system based on sparse network

Abstract

The invention relates to the technical field of target classification, and provides a target classification method and system based on a sparse network, and the method comprises the steps: obtaining an input image; inputting the input image into a sparse network to obtain a target classification result; the sparse network refers to a network in which a sparse layer replaces an original layer of the classification network; the classification network is obtained based on sample and label training; the sparsification layer is obtained by sparsification of parameters of the original layer according to the sparsity rate of the original layer; the sparse rate of the original layer is determined based on dichotomy iteration by taking a classification accuracy threshold value of the classification network as a constraint. According to the method, different sparse rate values can be given for different original layers according to the importance degrees of the original layers, and efficient pruning is performed on the network structure on the premise of ensuring the classification accuracy, so that the operation speed of the network is further improved, and the computing resource requirements are reduced.

Description

technical field [0001] The invention relates to the technical field of object classification, in particular to an object classification method and system based on a sparse network. Background technique [0002] The task of target classification is to identify the category of the object in the picture (in some cases, the confidence level corresponding to the category will also be given). The commonly used methods include using the HoG and SIFT features of the image combined with SVM classifier and other algorithms. The traditional BP neural network Algorithms and Convolutional Neural Network Algorithms, etc. [0003] Among the above methods, the target classification based on convolutional neural network usually has better accuracy, but similar target classification networks involve many parameters and the operation process is more complicated, and there are often bottlenecks in classification efficiency. [0004] A feasible method to improve the classification efficiency of...

AI Technical Summary

Problems solved by technology

[0006] The present invention provides an object classification method and system based on a sparse network, which is used to solve the defects in the prior art that the pruning process is lengthy and has a high impact on classification accuracy, and realize object classification based on efficient pruning

Images