Research And Application Of Assisted Diagnosis Of Pigmented Skin Disease Based On Transfer Learning

With the continuous change of the human living environment,there are many variations in the types of skin diseases.At present,the diagnosis and treatment of dermatological pathology is mainly based on traditional clinical visual discrimination.The accuracy of this examination method depends largely on the medical experience of dermatologists.Due to the fine-grained changes in the appearance of skin tissue damage,automatic classification and recognition of skin tissue damage using image recognition is a difficult task.With further research in deep learning,the Deep Convolutional Neural Network(DCNN)has shown that it performs general tasks and tasks with large variations in performing classification and feature vector discrimination on a large number of fine-grained objects Advantage,and apply it to the medical imaging research in the field of biomedical engineering becomes more and more important.Constructing a deep learning network model specifically for automatic classification and recognition of dermatological disease images and applying it to a computer-aided discrimination system for precision medicine,thereby indirectly helping dermatologists to predict and symptomatically damage skin cells Medication is of great significance.The main research focus of this thesis is to build a deep convolutional neural network model using transfer learning to realize the automatic classification and recognition of pigmented skin diseases images,and the design and implementation of an auxiliary diagnosis system for mobile devices of pigmented skin diseases.Through in-depth research on these two key parts,first,the data of the original skin mirroring data set ISIC2018 is pre-processed by using the related encoding method of Java file stream;then,the experiments in this article are based on the deep learning tool framework TensorFlow and use migration The learning method migrates Google's pre-trained Inception-V3 deep convolutional neural network model to the pre-processed target data set for tuning training.During the network model training process,it is completed by changing the shared weight coefficient and offset coefficient.And improve the learning effect of the target task in order to be able to extract more reasonable image feature vector values,until it is trained into a neural network model of its own target area to realize automatic classification and recognition of pigmented skin disease images.Finally,the auxiliary diagnosis system for pigmented dermatosis is established through the design pattern of Java's MVC framework,so that it can automatically call the model structure trained by the transfer learning method on the mobile device to realize the automatic classification and recognition of dermatological images.