Research On Optimization Of The Structural Of Convolutional Neural Network Based On Rice Breeding Optimization Algorithm

Image classification is one of the most important applications of deep learning.Convolutional neural network is widely used in image classification,and its performance will be affected by the structure.However,the optimal structure design and search are NP hard problems,which could not be completely solved at present.In order to optimize the structure of convolutional neural network,scholars have adopted different intelligent algorithms,such as genetic algorithm,particle swarm optimization algorithm,and achieved certain results.As an evolutionary algorithm proposed in recent years,rice breeding optimization algorithm has the characteristics of fast optimization speed and strong convergence ability.However,it is easy to fall into the local optimum.This thesis tries to improve the basic rice breeding optimization algorithm and use it in the structure optimization of convolution neural network,expecting to get better network structure,and for image classification,the main work is as follows:(1)This thesis implements a rice breeding optimization algorithm for optimization of the structural of convolutional neural network.The algorithm uses non-numeric encoding for individuals.Based on the characteristics of non-numeric individuals,unconventional updates are used to ensure that individuals of different lengths can hybridize directly.By using the rice breeding optimization algorithm to find the optimal structure of the corresponding problem a convolution neural network model with higher classification accuracy is obtained.(2)A parallel hybrid algorithm was used to mix the rice breeding optimization algorithm with the particle swarm optimization algorithm.Combining the global search ability of the particle swarm optimization algorithm with the fast convergence ability of the rice breeding optimization algorithm,this thesis explores the optimization ability of the hybrid algorithm,and combines the two algorithms to optimize the structure of convolution neural network.The results of the experiment show that the hybrid improved algorithm has a better performance.(3)A convolution neural network structure optimization method based on improved rice breeding optimization algorithm was designed and applied to the classification of public datasets and remote sensing images,not only comparing with the basic algorithm,but also experimenting with different convolution neural network models under the same conditions to study the applicability of this algorithm.The experiment shows that this method is not only effective in the classification of public datasets but also in the classification of remote sensing images.Good network structure could be found on the same dataset to improve the accuracy of classification.In general,the application of rice breeding optimization algorithm in the structure optimization of convolution neural network is studied in this thesis.Aiming at the shortcomings of rice breeding optimization algorithm,an improved method is proposed,which is tested with open dataset and remote sensing image data set.The results of the experiment show that the proposed algorithm can effectively improve the classification accuracy of convolution neural network,especially when the dataset is noisy and redundant,it can find a good network structure effectively.It has better applicability than some classical convolution networks,and the hybrid algorithm has better performance than the basic algorithm.