Research On Classification Of Poisonous Wild Vegetables Based On Convolutional Neural Networ

This article mainly studies the convolutionAl neurAl network model under deep learning to achieve classification and recognition of toxic wild vegetables.The main content is as follows:Firstly,three different convolutionAl neurAl network models were proposed for the classification and recognition of toxic wild vegetables,and a wild vegetable dataset was created for model training.And three models were trained with different parameter configurations,and an optimAl experimentAl configuration parameter was selected based on the experimentAl results.In order to better improve the recognition performance of the three models,different improvements were made based on the network structure of the three models.A method of changing the size of convolutionAl kernels and adding fully connected layers is proposed for the single network structure of the LeNet-5 model,which can improve the depth and classification performance of the model.For the AlexNet model,it is proposed to replace the LRN layer with the BN layer,use the PReLu function,and replace the originAl fully connected layer with globAl average pooling.The experiment found that the recognition accuracy of both models has been improved,but the improved recognition effect is not significantly different from the unmodified VGG16 model.Therefore,further improvement of the recognition effect of VGG16 is considered.Because the VGG16 model has a deeper network level and fewer toxic potherb datasets,the transfer learning method is used to improve the VGG16 model.On this basis,the recognition performance of the model was further improved by thawing the convolutionAl layer and changing the top-level structure.FinAlly,the VGG16 model with the best recognition performance after improvement was selected,and a classification and recognition system for toxic wild vegetables was developed.