Design And Implementation Of Intelligent Recognition System Of Wireless Capsule Endoscopic Lesion Image

Gastrointestinal disease is the most common disease among system diseases.Gastrointestinal cancer is a serious gastrointestinal disease.Common gastrointestinal cancers such as esophageal cancer,gastric cancer and colorectal cancer are all related to bleeding,polyps,ulcers and other gastrointestinal tract.Infection-related,there are approximately 2.8 million new gastrointestinal cancer cases and 1.8 million deaths each year.Early inspection and treatment can prevent gastrointestinal cancers.WCE(Wireless capsule endoscope)can inspect the gastrointestinal tract of patients in a painless and noninvasive way,and further diagnose possible gastrointestinal cancers by observing the patient's gastrointestinal infection.WCE examination can reduce the number of deaths due to gastrointestinal cancer and is of great significance for improving the healthcare system.However,the diagnosis and analysis of WCE images requires a lot of energy from the doctor and the accuracy of the results depends on the doctor's experience and professionalism.Therefore,this subject designs and implements a WCE lesion image recognition system based on deep learning to help doctors improve the accuracy of diagnosing WCE images,and provide objective and high-precision diagnosis basis for doctors' further diagnosis and treatment.With the continuous development of deep learning,methods for intelligently detecting and recognizing WCE lesion images have gradually shifted from traditional machine learning to methods based on deep learning technology.The method of detecting and recognizing WCE images based on deep learning can independently extract features and realize end-to-end learning,which provides doctors with high-precision diagnosis basis.This subject has designed and completed a WCE lesion image recognition system based on deep learning.The main research contents of this subject are as follows:(1)Research on multi-classification of WCE imagesThis part of the research realizes the intelligent detection of common WCE image lesions(bleeds,polyps,ulcers),and builds the Res Net50-CBAM multi-lesion image classification model through the migration learning strategy combined with the attention mechanism,and realizes the accuracy of the WCE multiple lesion images classification.The experimental results show that the average detection accuracy of each type of lesion image can reach 97.6%.(2)Research on polyp segmentation of WCE imageThis part of the research realizes the WCE image polyp segmentation based on deep learning.According to the characteristics of WCE polyp images,this research proposes a WCE polyp lesion segmentation model UNet-Res Net50 that combines migration learning strategy and UNet model.The Res Net50 model pre-trained by the Image Net database is used to replace the encoding part of the UNet model,and the UNet-Res Net50 model is implemented The precise segmentation of WCE polyp lesions is achieved.In the experimental part,by comparing the current excellent image segmentation methods(Seg Net,FCN,PSPNet)and its use of different networks as the encoding part of the WCE data set segmentation experimental results,the results show that the average segmentation of the UNet-Res Net50 model MIOU can reach 97.1%.(3)Research and development of prototype WCE lesion imaging systemThis research will combine the two most excellent models in WCE image multi-classification research and WCE image polyp segmentation research to develop and design a system prototype that intelligently assists in identifying WCE lesion images,providing high-precision diagnosis results and a user-friendly control interface,Improve the efficiency and accuracy of doctors' diagnosis of patients' WCE data.