Research On Visual Perception Technology Of Bladder Tumor Based On Deep Learning

Cancer is the second most serious disease that takes human lives in the world,and malignant tumors also seriously threaten the safety of the Chinese people.One of the methods of cancer treatment is to remove the tumor through surgery.During the bladder tumor resection,the bleeding of the urethral wall will produce blood noise,which will block the vision and affect the success rate of the operation.Aiming at the problem of blood noise affecting vision during urethroscopy,this article uses deep learning to recover clear urethral images from blood noise.The main tasks are as follows:(1)The training image restoration model lacks datasets.This article uses the tra-ditional method and the deep learning-based method to generate datasets.Berlin noise generation algorithm is choosen from traditional methods that can simulate the natural noise,and an improved network of Generative Adversarial Network(GAN)is choosen from deep learning methods.Compared with the original GAN,this thesis adds the de-tails of the image to the generated network to ensure the correlation between the image and the noise,and the loss function is easier to converge during training.(2)The restoration algorithm in this thesis involves two aspects of image and video.In terms of restoring clear images,this thesis designs a cascaded residual learning struc-ture based on vector quantized variational autoencoder(VQ-VAE),which achieves better visual effects than VQ-VAE,and is also better in evaluation indicators.In terms of recov-ering clear videos,this thesis adds a long and short-term memory(LSTM)network based on the residual VQ-VAE structure of image recovery,and good experimental results are obtained.(3)In order to speed up the transformation of research results into application prod-ucts,this thesis designs a blood noise video recovery software system,and the system is designed separately in modules.