Research And Design Of Vegetable Recognition System Based On Convolutional Neural Networ

With the rapid development of the Social Internet of Things,the traditional convolutional neural network is stretched in such a large data training and learning.The reason is that the training of the convolutional neural network model needs to consume a lot of time and computing resources,which is not suitable for solving large-scale optimization problems.Based on this,this paper uses distributed optimization technology to design a optimization algorithm,which is combined with a logistic regressor to form a hybrid logistic regressor to build a vegetable image recognition model.Convolutional neural network is used to extract vegetable image features,and an optimization algorithm is designed and applied to the vegetable model for classification.Finally,the classification accuracy is obtained through simulation experiments to verify the advantages of the optimization algorithm in vegetable image recognition.The main research work of this paper is as follows:(1)Creating a research object image dataset for convolutional neural network training and testing.Firstly,reasonable use of image collection methods-manual shooting and crawler downloading,etc.to collect vegetable pictures.Secondly,the initial simple labeling and image cutting of the image.Then,using image processing methods such as background replacement and random cropping,a data set with a uniform number of samples of different categories is obtained.Finally,the amount of training calculations is reduced by normalizing and standardizing the images.(2)Build a convolutional neural network.First,construct a LetNet neural network including 3 convolutional layers,2 pooling layers,and 2 fully connected layers and set appropriate parameters.Secondly,select appropriate activation functions,cost functions,and optimizations according to the research content requirements.Then,based on the construction of the image classification model of the Let Net-5 network.Finally,the training is carried out to obtain the optimal model and feature extractor.(3)A hybrid classifier model is integrated by combining the advantages ofconvolutional neural networks and logistic regression models.A logistic regressor based on a distributed stochastic algorithm is trained with the feature vector trained by the neural network as input using a transfer learning approach and classifies a test set of vegetable images.A new distributed stochastic proximal optimization algorithm is focused on designing to optimize the logistic regression model,and its accuracy and convergence are rigorously demonstrated.Numerical simulation results demonstrate the feasibility of the distributed stochastic proximal optimization algorithm for image classification applications.