Image classification method based on hyper-parameter optimization CNN

Abstract

The invention provides an image classification method based on hyper-parameter optimization CNN, and belongs to the technical field of image recognition. According to the method, structural parametersof a convolution layer C1 and a pooling layer P1 are selected as hyper-parameters of the method according to structural characteristics of a CNN architecture. The value range of the hyper-parametersis limited to be (Xl, Xu). After that,, the hyper-parameter of the CNN is optimized by adopting two periodic variation PSO algorithms of global variation and local variation. Therefore, the optimization of the hyper-parameter by the traditional PSO algorithm is prevented from staying at local optimum. The image classification performance which is more competitive than the traditional PSO algorithmis obtained. Compared with the prior art, the efficiency and the cost of deep learning CNN hyper-parameter optimization are obviously improved. The image classification potential of the CNN architecture is brought into play to the maximum extent. Hardware resources and calculation cost during image classification of the CNN are saved. The method has certain application value in engineering practice.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an image classification method based on hyperparameter optimization CNN. Background technique [0002] Image classification technology has been relatively mature, and CNN architectures suitable for different scene classifications emerge in an endless stream, but complex CNN structures often consume hardware resources and computing costs. Before CNN is used for image classification training, some internal parameters of CNN need to be set in advance. These parameters are called hyperparameters. Selecting a set of optimal hyperparameters can maximize the performance of CNN image classification without changing the structure of CNN. Therefore, it is particularly important in engineering practice to select suitable hyperparameters to fully release the image classification performance of the CNN architecture. [0003] Research on hyperparameter optimization methods for image...

AI Technical Summary

Problems solved by technology

Hyperparameter optimization based on particle swarm optimization, its search efficiency far exceeds other hyperparameter optimization algorithms such as grid search and random search, speeds up the search time of hyperparameter optimization, and solves the problem of low optimization efficiency and time-consuming traditional hyperparameter optimization And other issues

However, the particle swarm algorithm has the problem that it is easy to fall into the local optimum, which will cause the hyperparameter optimization to stay at the local optimum instead of the global optimal set of hyperparameters, which to a certain extent makes it impossible to search for the optimal CNN performance. A set of hyperparameters, so that CNN image classification cannot achieve the most ideal results

Images