Research On Medical Image Recognition Method Of Urine Visible Components Based On Deep Learning

Urinary sediment examination is one of the most commonly used clinical diagnostic items.It refers to the detection and analysis of the sediment under a microscope after the urine is centrifuged,which is of great significance for the detection of the genitourinary system and the monitoring of the body’s state.Urine formed component medical image recognition is the core subject of urine sediment inspection.The traditional method uses manual counting for statistics,which is time-consuming and labor-intensive and relies on the experience of the operator,which may easily lead to missed detections and false detections.In recent years,the image data of urine sediment has grown rapidly,and the use of computer vision technology to realize automatic urine sediment detection has a bright future.Urinary sediment image size is small,different categories are easy to confuse,the shape of the same category is diverse,feature extraction is difficult,it is necessary to build a deeper neural network for feature extraction,but this requires too much hardware and the recognition speed is slow,which is not conducive to applying to urine sediment in the actual scene of detection.In order to solve the above problems,based on the depth separable residual structure and superresolution image reconstruction technology,this paper proposes a lightweight urine sediment image recognition network,which achieves rapid recognition while ensuring high accuracy.The main work and innovations of this paper are as follows:1.Collected and produced the urine sediment image data set US_Data₂020 for training and testing,and filtered and enhanced the urine sediment images.2.In order to solve the problem of small size and low resolution of urine sediment images,this paper uses deep learning-based super-resolution image reconstruction technology SRGAN to perform super-resolution image reconstruction on small-size urine sediment images to enlarge the urine sediment images While retaining more details,it is helpful for medical personnel to observe and automatically recognize the neural network.3.In order to solve the problem of difficulty in extracting features of urine sediment images,this paper adopts deep learning methods and uses advanced network models such as VGGNet,Res Net,Efficient Net to recognize urine sediment images.At the same time,in order to solve the network lightweight problem,the lightweight model architectures such as Efficient Net and Mobile Net are explored,using deep separable convolution,combining the bottleneck structure and the inverse bottleneck structure,to improve the initial residual structure,and propose a new type of hourglass residual Based on this structure,the Dsr Net model used for urinary sediment image recognition in this paper is built.This paper proposes a new intelligent recognition method of urine sediment images.The experimental results show that the overall accuracy of the method for the recognition of 13 types of urine sediment images can reach 99.05%,and the model parameter is only 0.73 M.It is maintaining the network depth and recognition.The accuracy rate is light enough at the same time,which is conducive to transplantation to mobile devices,and has a good engineering application prospect.