Research On Bladder Tumor Sensing Technology Based On Meta Learning And Ensemble Learning

With the continuous breakthrough of deep learning methods,computer-aided diag-nosis technology has been more and more developed in the field of medical image anal-ysis.As the key to computer-aided diagnosis technology,medical image analysis relies on deep learning methods based on image processing technology.The cystoscopy tumor image analysis involved in this article relies on multi-target detection methods based on deep convolutional neural networks.However,the number of cystoscopy images and la-bels cannot get rid of the problem of lack of samples,and it faces technical difficulties in bladder tumor perception in few shot scenarios.This thesis,based on deep convolutional neural networks,starting from the multi-target classification and detection of few shot scenes,focusing on the actual effect of ensemble learning and meta learning on improving the robustness of classification and detection,proposes a method based on ensemble learning and meta learning and a bladder tumor detection method based on the meta-fusion network and it basically solves the clin-ical practice of bladder tumor microscopic image analysis in the few shot scene.Aiming at the defects of the above methods relying on hand-designed convolutional neural net-works,this thesis uses neural network architecture search methods to improve the current bladder tumor perception model,and proposes a few shot classification method based on task-adaptive neural network architecture search,which automatically searches out the structure of the convolutional neural network module,which achieves better performance than manually designed networks,and thus improves the detection method of bladder tu-mor.This thesis mainly carried out the above three research work,through designing com-parative experiments,analyzing and explaining the improvement effect of each method on the image perception of bladder tumor microscopy in few shot scenes,which is mainly re-flected in the improvement of recognition accuracy and detection indicators,together to help the current breakthrough in clinical computer-aided diagnosis technology for bladder tumors.