Research On Automatic Detection Of Urine Sediments Based On Deep Learning

In recent years,with the development of computer technology and modern evidence-based medicine,computer-assisted diagnosis and therapy has become an indispensable part of clinical diagnosis.More and more artificial intelligence technology has entered into the clinical practice,by providing effective technical support for clinical diagnosis,data analysis and sharing.With the development of computer vision technology and the improvement of image processing technology,many clinical tests have been lifted from traditional manual mode to modern automatic mode.Under this background,this dissertation focuses on the automatic detection of urine sediments,i.e.formed elements in urine,based on deep learning.The main research work and innovations are as follows:1.Collect and make a Uri Sed 2019 data set for the detection of urinary sediment sediment and systematically analyze and summarize the image characteristics of the data set.2.Aiming at the characteristics of small scale and high similarity of the image of sediment,this paper studies the target detection method based on deep learning,and proposes a target detection method based on effective receptive field,which uses the effective receptive field in convolutional neural network to distinguish the target into visible target and invisible target,so as to achieve the matching of positive and negative samples.By applying the above methods to the regional recommendation network,a new target detection network V-FCN is proposed.The results show that compared with the traditional object detection network,V-FCN has higher detection accuracy,and has significant advantages in the effectiveness and fairness of object proposal.Based on the object detection network of this paper,a set of recognition and detection system of upper computer and a website for recognition of visible components of urinary sediment are developed.