| The application of convolutional neural network(CNN)in the field of image recognition has become increasingly mature,with the advent of the Internet of Things era,the amount of data is growing exponentially.The huge training data requires a lot of computing resources and training time.The traditional centralized methods are not suitable for solving these large-scale optimization problems.On the one hand,the centralized framework is subject to performance limitations,such as high communication and computing requirements,single points of failure,and limited flexibility and scalability.On the other hand,the cost of transmitting data collected in a distributed manner to a central node is too high and may cause sensitive information leakage.Therefore,it is inevitable to study distributed optimization algorithms to solve large-scale optimization tasks.In addition,the application of distributed optimization algorithms in the field of image recognition has gradually become a focus of attention.Based on this,this paper uses the CNN model to extract the image features of poisonous mushrooms,and then designs a distributed optimization algorithm to optimize the logistic regression model and applies it to the research of poisonous mushroom image classification.Finally,it verifies the advantages of the distributed optimization algorithm in the recognition of poisonous mushroom images.The main research work of this paper is as follows:(1)Establishing a poisonous mushroom image data set for machine learning training and testing.Analyzing the advantages and characteristics of CNN over traditional neural networks in machine learning.Analyzing and selecting the relevant elements involved in the training of the convolutional neural network.By using transfer learning and learning from the Letnet network model structure,the model is tested and trained with the poisonous mushroom image data set.Finally we keep the model as a feature extractor.(2)Combining the advantages of CNN and logistic regression models,a hybrid classification model is integrated.The hybrid model uses CNN as a trainable feature extractor,and uses a logistic regression model optimized by a distributed algorithm as a trainable classifier to classify poisonous mushroom images.This thesis designs a new accelerated distributed event trigger algorithm(referred to as A-DETA)to optimize the logistic regression model,and strictly proves its precise convergence.The simulation results verify the feasibility of the A-DETA algorithm and the correctness of the theoretical analysis.(3)Applying the hybrid model to classify poisonous mushroom images and verifying the feasibility of the hybrid model.Comparing the A-DETA algorithm with other optimization algorithms,without reducing the test accuracy,it is finally verified that the hybrid model has advantages in terms of communication times and convergence rate. |
