Research On Automatic Calibration System Of Cold-water Meter Based On Machine Vision

The percentage of white blood cell content in the blood can often diagnose a certain type of disease.Therefore,studying the content of white blood cells in the blood has a great auxiliary effect on the clinical guidance of doctors.But at the same time,the distinction between the morphological characteristics of various types of white blood cells is not obvious.The statistics of the five types of white blood cells in the peripheral blood cells can help quickly determine the type of disease of the patient and help doctors take effective measures to treat The patient makes a diagnosis.Therefore,most hospitals use manual microscopy to improve detection accuracy.However,this method is inefficient,consumes manpower and material resources,and requires doctors’ experience.It is not conducive to rapid diagnosis of patients.With the rapid progress of neural networks in the field of deep learning,image target detection tasks and target recognition tasks have made great progress.The update iterations of computer hardware technology and the advancement of CCD imaging technology have enabled the preservation of medical images,which is conducive to the combination of deep learning technology and the medical field,and the research on medical images.Therefore,this article discusses the detection and recognition of white blood cell images by deep learning detection and recognition algorithms,and verifies them on the self-organized data set and the ALL-IDB public data set.Based on the one-stage deep learning detection framework YOLO model,this paper proposes a CBAM-YOLO white blood cell detection algorithm that blends attention mechanism,which only detects white blood cell images.First,the improved K-Means algorithm clusters the sample anchor boxes,and outputs clustering priors of three different sizes for each layer of CBAM-YOLO;then,after adding spatial attention to the residual structure of the network model Mechanism and channel attention mechanism to improve the network model’s ability to locate the position of white blood cells.Finally,the feature map uses the feature pyramid structure to judge and output the target coordinate position.In this way,the ability of the model algorithm in this paper to locate white blood cells is greatly improved,while ensuring the detection and reasoning time of the system.Aiming at the field of white blood cell classification,this paper uses multiple network structures to identify and verify the five types of white blood cells.Firstly,the data is analyzed,and the data set is expanded through enhancement operations such as translation,rotation,contrast,and mirroring;secondly,the network model input In the case of fixed resolution,this article refers to the global average pooling operation so that the network model can accept inputs of different sizes,and discusses the different fitting capabilities of the neural network model of the white blood cell image at different sizes.Finally,in view of the possible over-fitting results of the network model,this paper applies the Loss function of the smooth label,and at the same time adopts the Dropout strategy to punish the neural network model to prevent the learning rate from falling into a saddle point or the model from over-fitting.In the end,this paper adopts the residual network model to use the migration learning strategy to identify and train white blood cells,and finally complete the automatic identification of five types of white blood cells.This paper uses the deep learning detection network and the recognition network to detect and recognize white blood cells separately,confirming the feasibility of combining deep learning technology with medical images,and at the same time integrate the two algorithms into an independent system through hardware equipment,to a large extent This improves the speed of white blood cell diagnosis,reduces the cost of medical equipment,and reduces the burden on doctors.