Improved Research And Application Of AlexNet Algorithm

With the increasing popularity of artificial intelligence technology in the medical field,the system of automatic diagnosis of medical images has become a hot research direction for researchers and has attracted wide attention from the academic community.In diabetic patients,retinopathy is a widespread complication that can lead to blindness,and current screening and diagnostic methods mainly rely on experienced ophthalmologists,but this approach may lead to inaccurate results.Using deep learning and computer vision technology,a classification model(AlexNet+)for diabetic retinopathy was proposed to achieve more accurate classification monitoring.For the input image,AlexNet+ network model is classified and processed.Aiming at the problem of inaccurate network classification caused by the small number of some categories of images in Eye PACS dataset,data augmentation and data equalization are carried out to improve the classification accuracy.Secondly,considering that the original network has a small number of layers,but the network model has a large number of parameters,which leads to overfitting in the image classification network model,this paper proposes IncAlexNet network model,which adjusts the convolution layer of the traditional AlexNet network model and draws inspiration from the Goog Le Net network model.The Inception module was used to replace the ordinary convolutional layer of AlexNet to improve the feature extraction ability of the convolutional network,and the Eye Pacs dataset was input into the Inc-AlexNet network model.The experimental results show that the proposed Inc-AlexNet model has a certain improvement in the two evaluation indicators of accuracy and sensitivity.In order to solve the problem of gradient disappearance in Inc-AlexNet network model,the R6-Softsign activation function is used to replace the traditional Re LU activation function,and the RS-Inc-AlexNet network model is proposed.The proposed RS-Inc-AlexNet model algorithm is experimentally compared with the basic convolutional neural network algorithm and other retinal classification algorithms for diabetic patients on the Eye Pacs dataset.The results show that the classification accuracy of RS-Inc-AlexNet for diabetic retinopathy reaches91.32%.In addition,the AlexNet+ network model adds a dual attention mechanism module to each convolutional layer to improve the feature extraction and feature expression ability of the convolutional layer,suppress noise and redundant information,and help improve the robustness of the network model.AlexNet+ is constructed for the classification and recognition of data sets with too much detail information.The experimental results show that the classification accuracy of AlexNet+ model on Eye PACS dataset reaches 93.75%,and the AlexNet+ diabetic retinopathy classification algorithm proposed in this paper has higher accuracy.