Research On Image Recognition And Classification Of Urinary Sediment Based On Adaptive Segmentation Algorithm And Transfer Learning

Urine sediment image recognition and classification system is the core of the automatic urine sediment analyzer.The traditional urine sediment inspection method has complicated test steps,many interfering links,and poor repeatability.Therefore,the use of digital image processing technology to accurately and quickly recognize and classify the visible components in urine sediment images,applied to the automatic recognition and classification system of urine sediment,is of great significance for the realization of intelligent and scientific analysis of medical images.In order to solve this problem,this paper carries out targetedly preprocessing on urine sediment images;uses the adaptive segmentation algorithm which is fusion of Mean-shift clustering and Canny operator to segment urine sediment images;three deep transfer learning models are used to complete the classification and compared with three commonly used machine learning classification methods;based on the deep transfer learning model,the method of urinary sediment image classification is improved,which effectively improves the accuracy of the classification of urinary sediment visible components.Firstly,the reasons for the difficulty of segmentation of urine sediment images are described,and the characteristics of urine sediment are analyzed.Three filtering methods are used and the noise reduction effect of urine sediment image is compared.The first type of image is enhanced by using histogram equalization,the second type of image is enhanced by using combined enhancement methods.Secondly,applied two commonly used image segmentation algorithms on urine sediment images and analyzed the segmentation effect.According to the segmentation effect,the Canny operator is used as the first image segmentation algorithm.The fusion of Mean-shift clustering and Canny operator is proposed as the second type of image segmentation algorithm.The adaptive segmentation algorithm in this paper is designed and the results of segmentation experiments are analyzed.The results show that the algorithm has a good segmentation effect.Then,the flow of common machine learning algorithms to realize the classification of urine sediment visible components is analyzed.The shape and texture features of urine sediment visible components are extracted and selected for classification.The classification results of urine sediment visible components by BP,SVM and ELM is analyzed.It is found that classification based on feature extraction has the problems of low accuracy and inaccurate statistics.Finally,aiming at the problem of low classification accuracy,it is proposed to use three deep transfer learning models for classification;make image data sets of urine sediment and adjust the network structure and parameters of the deep learning transfer model,train and test the correct classification of the model rate;verified that the transfer learning model can effectively improve the classification accuracy.Aiming at the problem of inaccurate statistics.An enumeration classification method based on deep transfer learning model was proposed,which targetedly expanded the urinary sediment visible component data set;retrain and test the transfer model which with the best comprehensive performance.The model is applied to the recognition and classification system of urinary sediment visible components.Through the comparison of experimental results,it is verified that it can complete the recognition classification and quantity statistics with higher accuracy.