Research On Feature Extraction Of Laryngoscope Image Distribution And Texture Fusion And Improved AlexNet Classification Algorithm

The classification and prediction of diseases has always been a concern in the medical field.Regarding the medical image data of the laryngoscopy and endoscopy,professional physicians diagnose the throat reflux disease by observing the laryngoscopy images and medical treatment.This is undoubtedly a task with a huge amount of tasks and strict requirements.Computer-aided professional physicians have great significance in disease prediction,especially the research on computer-assisted treatment of laryngoscope image data is rare.This paper is based on laryngoscopy image data of otolaryngology head and neck surgery in a hospital in Jilin Province.The research on the classification of laryngoscope image data from the improved feature extraction method and the improved AlexNet convolutional neural network.The first part studies the feature extraction of laryngoscope images from the 2 aspects of image distribution and texture.The local binomial feature extraction method is used to extract the image texture feature,the gray histogram method is used to extract the image distribution feature,and the feature fusion is performed by the difference comparison method.The combined performance of the 5 traditional classification algorithms is compared.The results show that the combination of the distribution and texture fusion features with the random forest discriminant algorithm has the highest accuracy in the classification of laryngoscope images,reaching 96.61%,and the algorithm does not require high sample size.The second part constructs a convolutional neural network algorithm for classification of laryngoscope image data.The shallow feature and deep feature structure of AlexNet are modified to avoid the original network’s ignorance of the subtle differences in the laryngoscope images,and to enlarge the difference information as much as possible.Using this algorithm to classify and predict laryngoscope images has good results,with an accuracy rate of 99.83%,which is 27.65% higher than the accuracy of the unimproved AlexNet convolutional neural network for laryngoscope image classification,and the algorithm can adapt to large samples data.