CN110473166A - A kind of urinary formed element recognition methods based on improvement Alexnet model - Google Patents
A kind of urinary formed element recognition methods based on improvement Alexnet model Download PDFInfo
- Publication number
- CN110473166A CN110473166A CN201910614832.5A CN201910614832A CN110473166A CN 110473166 A CN110473166 A CN 110473166A CN 201910614832 A CN201910614832 A CN 201910614832A CN 110473166 A CN110473166 A CN 110473166A
- Authority
- CN
- China
- Prior art keywords
- formed element
- network model
- urinary formed
- image
- alexnet
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000002485 urinary effect Effects 0.000 title claims abstract description 88
- 238000000034 method Methods 0.000 title claims abstract description 31
- 230000006872 improvement Effects 0.000 title claims description 3
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000012360 testing method Methods 0.000 claims abstract description 45
- 210000002700 urine Anatomy 0.000 claims abstract description 44
- 239000013049 sediment Substances 0.000 claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 11
- 230000002708 enhancing effect Effects 0.000 claims description 10
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 241000894006 Bacteria Species 0.000 claims description 6
- 210000004027 cell Anatomy 0.000 claims description 6
- 229920000297 Rayon Polymers 0.000 claims description 5
- 241000235342 Saccharomycetes Species 0.000 claims description 5
- QXDMQSPYEZFLGF-UHFFFAOYSA-L calcium oxalate Chemical compound [Ca+2].[O-]C(=O)C([O-])=O QXDMQSPYEZFLGF-UHFFFAOYSA-L 0.000 claims description 5
- 239000013078 crystal Substances 0.000 claims description 5
- 210000003743 erythrocyte Anatomy 0.000 claims description 5
- 239000000463 material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 5
- 230000005284 excitation Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000007788 liquid Substances 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 6
- 230000000116 mitigating effect Effects 0.000 abstract description 3
- 239000004615 ingredient Substances 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 241000208340 Araliaceae Species 0.000 description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 235000008434 ginseng Nutrition 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000007170 pathology Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005119 centrifugation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000010827 pathological analysis Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000009394 selective breeding Methods 0.000 description 1
- 230000009885 systemic effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
- G01N15/10—Investigating individual particles
- G01N15/14—Electro-optical investigation, e.g. flow cytometers
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
- G01N33/483—Physical analysis of biological material
- G01N33/487—Physical analysis of biological material of liquid biological material
- G01N33/493—Physical analysis of biological material of liquid biological material urine
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/60—Rotation of a whole image or part thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G01N15/01—
-
- G01N15/149—
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume, or surface-area of porous materials
- G01N15/10—Investigating individual particles
- G01N15/14—Electro-optical investigation, e.g. flow cytometers
- G01N2015/1486—Counting the particles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Abstract
The present invention relates to field of medical image processing, and in particular to a kind of based on the urinary formed element recognition methods for improving Alexnet model.Step 1: acquisition and expansion image data set construct urine sediment image training set and test set；Step 2: urinary formed element of the building based on Alexnet network model identifies network model；Step 3: setting urinary formed element identifies the training parameter of network model；Step 4: urinary formed element of the training based on Alexnet network model identifies network model；Step 5: urinary formed element of the test based on Alexnet network model identifies network model；The present invention is improved on the basis of Alexnet network model, reduce network training parameter amount, characteristics of image can be automatically extracted, has the characteristics that discrimination is high, recognition time is fast, generalization ability is strong, it is with important application prospects for assisted medical diagnosis, mitigation doctor's burden.
Description
Technical field
The present invention relates to field of medical image processing, and in particular to a kind of tangible based on the urine for improving Alexnet model
Ingredient recognition methods.
Background technique
The visible component in concentrated urine obtained by centrifugation is referred to as arena, and urinary formed element inspection is
One of hospital's routine inspection project can help clinician to understand the variation at each position of urinary system, for assisting uropoiesis
Level diagnosis, antidiastole and the Index for diagnosis of systemic disease play an important role.Microexamination is visible tangible in urine
Ingredient type is very more, and some of ingredients have specific pathology sense, such as bacterium, red blood cell, leucocyte have very
Important diagnostic value.Therefore the identification of urinary formed element and Accurate classification mitigate doctor's burden for assisting pathological diagnosis
Play a significant role.
To figure after being observation under the microscope due to conventional urinary formed element detection method or being imaged by photographic device
Piece carries out observation and manual identified, and the heavy workload of this method is cumbersome, and is easy by technical staff's level difference
The problem of influencing, be easy to causeing missing inspection erroneous detection, while the medical image sharply increased is also that manual identified increases difficulty.With
The development of computer technology has had already appeared a variety of methods for carrying out automated analysis to urine sediment image, but existing
Method all also needs manual extraction characteristics of image, and artificial selection classifier is classified.These automatic methods are worked
Journey is complicated, and the influence vulnerable to technical staff, it is therefore desirable to which more effective urinary formed element recognition methods reduces pathology
The workload of doctor.
Currently, having caused as deep learning is in the application of every field and having carried out data using the technology in more areas
The upsurge of analysis, deep learning also result in attention in field of medical image processing.It is introduced in urinary formed element identification field
Deep learning method is trained arena picture using convolutional neural networks, can automatically extract the important spy in image
Sign, and obtain accurate recognition result.
Summary of the invention
The purpose of the present invention is to provide it is a kind of based on improve Alexnet model urinary formed element recognition methods, with
Realization automatically extracts characteristics of image, with the convolution number of plies is less, parameter amount is relatively fewer, discrimination is high, recognition speed is fast, extensive
The strong feature of ability, it is with important application prospects for assisted medical diagnosis, mitigation doctor's burden.
The embodiment of the present invention provide it is a kind of based on improve Alexnet model urinary formed element recognition methods step such as
Under:
Step 1: acquisition and expansion image data set construct urine sediment image training set and test set；
The initial pictures that arena is acquired according to microscope, are labeled the urinary formed element in initial pictures, right
The image category for wherein marking negligible amounts carries out data enhancing processing, obtains urine sediment image data set；Later in proportion with
Machine chooses the image after mark, and building obtains urine sediment image training set and urine sediment image test set；Wherein, the urine has
It is formed point including bacterium, saccharomycete, calcium oxalate crystal, hyalina, viscose rayon, red blood cell, sperm, squamous cell, white
Cell, leucocyte roll into a ball totally ten class；
Step 2: urinary formed element of the building based on Alexnet network model identifies network model；
Step 3: setting urinary formed element identifies the training parameter of network model；
Step 4: urinary formed element of the training based on Alexnet network model identifies network model；
According to the training parameter being arranged in step 3, the urine sediment image training set obtained using step 1 is tangible to urine
Ingredient identification network model is trained, and after iteration to maximum number of iterations, obtains the urinary formed element identification of training completion
Network model；
Step 5: urinary formed element of the test based on Alexnet network model identifies network model；
The urine sediment image test set obtained using step 1 tests urinary formed element identification network model, obtains
The recognition result and overall accuracy of urinary formed element into urine sediment image test set.
The invention also includes structure features some in this way:
1, the step 1, in which:
The hyalina urinary formed element data volume is less, needs to carry out data enhancing, specifically includes two methods
Are as follows: rotation transformation: Random-Rotation image certain angle changes the direction of picture material；It is turning-over changed: along horizontally or vertically
Direction flipped image；
The image randomly selected after mark in proportion, i.e., to the data set of ten kinds of urinary formed element classifications with 4:1
Ratio, the data set after randomly selecting mark constructs urine sediment image training set and urine sediment image test set, arena
Training set of images and training set label are for training urinary formed element to identify network model, urine sediment image test set and test
Collection label is used to test the performance for the urinary formed element identification network model that training obtains；
2, step 2 specifically:
Urinary formed element identifies that convolutional neural networks model used in network model is according to existing Alexnet network
What model improved, comprising: the input layer for improving Alexnet network model is 171 × 171 × 1, first convolutional layer
Use 96 step-lengths for 39 × 9 filters, then the maximum pond by using 3 × 3 convolution kernel and step-length as 2
Calculated result is transferred to next layer by layer, and total five convolutional layers and three pond layers are applied in combination；It is three later
A full articulamentum, wherein the number of nodes of the last one full articulamentum is 10, respectively corresponds ten kinds of urinary formed elements, finally leads to
It crosses softmax classifier and identifying processing is carried out to ten kinds of different urinary formed elements, export to obtain in picture in output layer
The recognition result of urinary formed element.
3, the Alexnet network model one is divided into eight layers, wherein first five layer is convolutional layer, and latter three layers are to connect entirely
Layer contains excitation function ReLU and local acknowledgement's normalization LRN processing in each convolutional layer, passes through drop again later
Sample pool processing.
4, step 3 specifically: setting urinary formed element identifies that the size of the input picture of network model is 171 × 171
× 1 gray level image, the small lot size on single GPU are 64, and initial learning rate is 0.001.
5, the step 4, in which:
During the training identifies network model based on the urinary formed element of Alexnet network model, by small quantities of
Stochastic gradient descent algorithm iteration optimization loss function is measured, loss function is made to reach minimum, to adjust the convolution weight ginseng of network
Number.
The beneficial effects of the present invention are:
1. the invention proposes a kind of based on the urinary formed element recognition methods for improving Alexnet model, for urine
Visible component identification mission, by construct convolutional neural networks model, urine sediment image can be carried out automatically feature extraction and
Identification；
2. the present invention is on arena test set to bacterium, saccharomycete, calcium oxalate crystal, hyalina, viscose rayon, red thin
The recognition result that born of the same parents, sperm, squamous cell, leucocyte, leucocyte roll into a ball 10 class visible components is assessed, and training obtains
Improved model based on Alexnet network obtains 96.72% accuracy rate on test set, in Core i7-7700K CPU,
Under the hardware condition of NVIDIA 2080Ti GPU and on MATLAB platform, the recognition time of every urine sediment image is only needed
4.83ms；
3. the characteristics of present invention differs larger and be gray level image for arena dimension of picture size, the present invention exists
It is improved on the basis of Alexnet network model, reduces network training parameter amount, characteristics of image can be automatically extracted,
Have the characteristics that discrimination is high, recognition time is fast, generalization ability is strong, for assisted medical diagnosis, mitigates doctor's burden with weight
The application prospect wanted.
Detailed description of the invention
Fig. 1 is a kind of flow chart based on the urinary formed element recognition methods for improving Alexnet model；
Fig. 2 is the schematic diagram that urinary formed element of the invention identifies network model；
Fig. 3 obscures square with what urine sediment image test set test urinary formed element identification network model obtained for the present invention
Battle array；
Fig. 4 is the normalization that the present invention tests that urinary formed element identification network model is obtained with urine sediment image test set
Confusion matrix.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing to the present invention
It is described further:
A kind of urinary formed element recognition methods based on Alexnet network model, comprising the following steps:
Step 1: acquisition and expansion image data set construct urine sediment image training set and test set；
Arena microscope photograph is acquired, after being labeled to the urinary formed element in image, to negligible amounts
Image category carries out data enhancing, and randomly selects the image after mark in proportion to construct training set and test set；
The urinary formed element includes bacterium, saccharomycete, calcium oxalate crystal, hyalina, viscose rayon, red blood cell, essence
Son, squamous cell, leucocyte, leucocyte roll into a ball totally ten class；
Step 2: urinary formed element of the building based on Alexnet network model identifies network；
The convolutional neural networks model is the network model improved on the basis of existing Alexnet network,
Improve AlexNet network model input layer be 171 × 171 × 1, first convolutional layer use 96 step-lengths for 39 × 9 filter
Wave device, then the maximum pond of progress is turned to output and is transmitted to next layer, and there are five convolutional layers and three pond layers to be applied in combination altogether.
It is three full articulamentums later, wherein the number of nodes of the last one full articulamentum is 10, corresponding ten kinds of urinary formed elements, then
The identification to ten kinds of different urinary formed elements is completed by softmax classifier, is exported in output layer to urine in picture
The recognition result of visible component；
Step 3: the training parameter of the network is set；
The gray level image that the size of input picture is 171 × 171 × 1, the small lot size on single GPU are 64, initially
Learning rate is 0.001；
Step 4: urinary formed element identification model of the training based on Alexnet network model；
The urine sediment image data set obtained using step 1, in conjunction with the training parameter being arranged in step 3, training urine has
A point identification network is formed, obtains network model after iteration to maximum number of iterations；
Step 5: the urine sediment image test set obtained using step 1 tests gained model, and it is each to obtain test set
The recognition result and overall accuracy of urinary formed element in a image.
Further, hyalina image less for data volume in the step 1 has carried out data enhancing, including following
Method:
Rotation transformation: Random-Rotation image certain angle changes the direction of picture material；
It is turning-over changed: along horizontal or vertical direction flipped image；
Further, in the urinary formed element identification network development process in training based on Alexnet network model, by small quantities of
Stochastic gradient descent algorithm iteration optimization loss function is measured, loss function is made to reach minimum, to adjust the convolution weight ginseng of network
Number
Data enhancing, including following methods have been carried out for the less hyalina image of data volume in the step 1: rotation
Transformation is changed: Random-Rotation image certain angle changes the direction of picture material；It is turning-over changed: along horizontal or vertical direction
Flipped image；
The convolutional neural networks model of the step 2 is improved on the basis of existing Alexnet network,
The former input layer of Alexnet network is 227 × 227 × 3, but since arena dimension of picture is generally smaller and is gray level image,
Therefore the input layer for improving legacy network is 171 × 171 × 1, is allowed to be more suitable for arena data set；
The Alexnet improved model of the step 2, first convolutional layer use 96 step-lengths for 39 × 9 filters, make
It obtains the characteristic pattern that convolution operation obtains and is more suitable for the characteristic information of input picture, while reducing the training parameter of network model
Amount；
The Alexnet improved model of the step 2, the number of nodes of the last one full articulamentum section are 10, corresponding ten kinds of urine
Liquid visible component；
Described image recognition methods is applied to medical image recognition.
The embodiment of the present invention is provided in conjunction with specific value:
As shown in Figure 1, a kind of urinary formed element recognition methods based on Alexnet network model of the present invention is main
Include the following steps:
Step 1: acquisition and expansion image data set construct urine sediment image training set and test set；
Arena MIcrosope image is acquired, the urinary formed element in image is labeled, obtained urine sediment image
Sum is 267788, wherein sharing ten kinds of different urinary formed element classifications: bacterium, saccharomycete, calcium oxalate crystal, transparent
Cast, viscose rayon, red blood cell, sperm, squamous cell, leucocyte, leucocyte group.
There is a problem of that quantity differs larger between each category dataset of urinary formed element, wherein hyalina is tangible
The minimum number of ingredient, only 862.Occur causing that accuracy is not high or model is excessively quasi- since data volume is insufficient in order to prevent
The problem of conjunction, has carried out data enhancing to the least hyalina data set of quantity, hyalina data set has been extended to
4312.
Data enhancing is carried out to hyalina image, mainly uses following methods:
Rotation transformation: Random-Rotation image certain angle changes the direction of picture material；
It is turning-over changed: along horizontal or vertical direction flipped image；
Later, the data to the data set of ten kinds of urinary formed element classifications all with the ratio of 4:1, after randomly selecting mark
Collect to construct training set and test set, training set and training set label are for training network model, test set and test set label
For testing the performance for the network model that training obtains.
Step 2: urinary formed element of the building based on Alexnet network model identifies network model；
AlexNet network model one is divided into eight layers, and wherein first five layer is convolutional layer, and latter three layers are full articulamentums, every
Excitation function ReLU and local acknowledgement's normalization (LRN) processing are all contained in one convolutional layer, then using down-sampled
(pool processing).Other deep learning network models are compared, Alexnet network model is more suitable for the identification of urine sediment image, it
It is less with the convolution number of plies, parameter amount is relatively fewer, be able to carry out the characteristics of quick and precisely identifying.Since arena picture has
Size is generally smaller, and majority is in 16 pixels between 200 pixels, and the characteristics of be gray level image, and Alexnet network is defeated
Enter layer it is desirable that 227 × 227 sizes color image.In order to adapt to arena dimension of picture attribute, network model is reduced
Training parameter, convolutional neural networks model of the present invention are improved on the basis of existing Alexnet model.
It is urinary formed element identification network architecture of the present invention as shown in Figure 2.Improve AlexNet network mould
The input layer of type is 171 × 171 × 1, is allowed to be more suitable for arena data set.First convolutional layer uses 96 step-lengths for 3
9 × 9 filters so that the characteristic pattern that convolution operation obtains is more suitable for the characteristic information of input picture, while reducing network
The training parameter amount of model.Then by maximum pond layer, and calculated result output is transmitted to next layer, what pond layer used
It is 3 × 3 convolution kernel, step-length 2.There are five convolutional layers and three pond layers to be applied in combination altogether for the model, is three complete later
Articulamentum, the full articulamentum number of nodes of third is 10, respectively corresponds ten kinds of urinary formed elements, classifies eventually by softmax
Device completes the identification to ten kinds of different urinary formed elements, in output layer output to the identification knot of urinary formed element in picture
Fruit.
Step 3: the training parameter of the network is set；
The gray level image that input picture is 171 × 171 × 1, is trained, batch input picture number on single GPU
It is 64, initial learning rate is 0.001；
Step 4: urinary formed element identification model of the training based on Alexnet network model；
The urine sediment image training set and training set label obtained using step 1, in conjunction with the training ginseng being arranged in step 3
Number, urinary formed element of the training based on Alexnet network model identify network.
It is preceding to biography by minimizing using back propagation learning algorithm and stochastic gradient descent method in training pattern
The loss function value size broadcast carrys out inverse iteration and updates network weight, when the loss function value convergence of model, deconditioning
Network；
Step 5: the urine sediment image test set obtained using step 1 carries out the Alexnet improved model that step 4 obtains
Test, obtains the urinary formed element recognition result and overall accuracy in each image of test set；
It is as shown in Figure 3 and shown in Fig. 4 test result of the model on test set, obscures from confusion matrix and normalization
Matrix can be seen that the model and all obtain preferable discrimination in ten kinds of urinary formed element classifications, minimum
87.21%, up to 98.77%.Overall accuracy of the model on entire test set has reached 96.72%.
To sum up, the invention discloses a kind of urinary formed element recognition methods based on Alexnet network model, firstly,
Acquisition and EDS extended data set carry out data enhancing for the classification of negligible amounts in image pattern, and divide training set and test
Collection；Then, according to arena samples pictures size attribute, the urinary formed element identification based on Alexnet network model is constructed
Network, and network parameter is set and is trained；Finally, being tested using test set trained network model, obtain
96.72% accuracy rate.Improved model proposed by the present invention based on Alexnet network can automatically extract characteristics of image, and
Ten kinds of urinary formed elements are identified with have the characteristics that discrimination is high, recognition time is fast, generalization ability is strong, for auxiliary
Medical diagnosis, mitigation doctor's burden are with important application prospects.
Claims (6)
1. a kind of based on the urinary formed element recognition methods for improving Alexnet model, it is characterised in that: steps are as follows:
Step 1: acquisition and expansion image data set construct urine sediment image training set and test set；
The initial pictures that arena is acquired according to microscope, are labeled the urinary formed element in initial pictures, to wherein
The image category for marking negligible amounts carries out data enhancing processing, obtains urine sediment image data set；Random choosing in proportion later
Image after taking mark, building obtain urine sediment image training set and urine sediment image test set；Wherein, the urine to be formed
Point including bacterium, saccharomycete, calcium oxalate crystal, hyalina, viscose rayon, red blood cell, sperm, squamous cell, leucocyte,
Leucocyte group's totally ten class；
Step 2: urinary formed element of the building based on Alexnet network model identifies network model；
Step 3: setting urinary formed element identifies the training parameter of network model；
Step 4: urinary formed element of the training based on Alexnet network model identifies network model；
According to the training parameter being arranged in step 3, the urine sediment image training set obtained using step 1 is to urinary formed element
Identification network model is trained, and after iteration to maximum number of iterations, obtains the urinary formed element identification network of training completion
Model；
Step 5: urinary formed element of the test based on Alexnet network model identifies network model；
The urine sediment image test set obtained using step 1 tests urinary formed element identification network model, is urinated
The recognition result and overall accuracy of sediment image measurement concentration urinary formed element.
2. according to claim 1 a kind of based on the urinary formed element recognition methods for improving Alexnet model, feature
It is, the step 1, in which:
The hyalina urinary formed element data volume is less, needs to carry out data enhancing, specifically includes two methods are as follows: rotation
Transformation is changed: Random-Rotation image certain angle changes the direction of picture material；It is turning-over changed: along horizontal or vertical direction
Flipped image；
The image randomly selected after mark in proportion, i.e., to the data set of ten kinds of urinary formed element classifications with the ratio of 4:1
Example randomly selects the data set after marking to construct urine sediment image training set and urine sediment image test set, urine sediment image
Training set and training set label are for training urinary formed element to identify network model, urine sediment image test set and test set mark
Sign the performance for testing the urinary formed element identification network model that training obtains.
3. a kind of urinary formed element recognition methods based on improvement Alexnet model according to claim 1 or 2,
It is characterized in that: step 2 specifically:
Urinary formed element identifies that convolutional neural networks model used in network model is according to existing Alexnet network model
It improves, comprising: the input layer for improving Alexnet network model is 171 × 171 × 1, and first convolutional layer uses
9 × 9 filters that 96 step-lengths are 3 will then by the maximum pond layer for using 3 × 3 convolution kernel and step-length as 2
Calculated result is transferred to next layer as output, and total five convolutional layers and three pond layers are applied in combination；It is three complete later
Articulamentum, wherein the number of nodes of the last one full articulamentum is 10, respectively corresponds ten kinds of urinary formed elements, finally by
Softmax classifier carries out identifying processing to ten kinds of different urinary formed elements, exports to obtain in output layer and urinate in picture
The recognition result of liquid visible component.
4. according to claim 3 a kind of based on the urinary formed element recognition methods for improving Alexnet model, feature
Be: the Alexnet network model one is divided into eight layers, wherein first five layer is convolutional layer, and latter three layers are full articulamentum, In
Excitation function ReLU and local acknowledgement's normalization LRN processing are contained in each convolutional layer, later again by down-sampled
Pool processing.
5. according to claim 4 a kind of based on the urinary formed element recognition methods for improving Alexnet model, feature
It is: step 3 specifically: setting urinary formed element identifies that the size of the input picture of network model is 171 × 171 × 1
Gray level image, the small lot size on single GPU are 64, and initial learning rate is 0.001.
6. according to claim 5 a kind of based on the urinary formed element recognition methods for improving Alexnet model, feature
It is: the step 4, in which:
The training based on Alexnet network model urinary formed element identify network model during, by small lot with
Machine gradient descent algorithm iteration optimization loss function makes loss function reach minimum, to adjust the convolution weight parameter of network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614832.5A CN110473166A (en) | 2019-07-09 | 2019-07-09 | A kind of urinary formed element recognition methods based on improvement Alexnet model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910614832.5A CN110473166A (en) | 2019-07-09 | 2019-07-09 | A kind of urinary formed element recognition methods based on improvement Alexnet model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110473166A true CN110473166A (en) | 2019-11-19 |
Family
ID=68507151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910614832.5A Pending CN110473166A (en) | 2019-07-09 | 2019-07-09 | A kind of urinary formed element recognition methods based on improvement Alexnet model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473166A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178174A (en) * | 2019-12-12 | 2020-05-19 | 美康盛德医疗科技(苏州)有限公司 | Urine tangible component image identification method based on deep convolutional neural network |
| CN111583292A (en) * | 2020-05-11 | 2020-08-25 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111652326A (en) * | 2020-06-30 | 2020-09-11 | 华南农业大学 | Improved fruit maturity identification method and identification system based on MobileNet v2 network |
| CN112200726A (en) * | 2020-10-29 | 2021-01-08 | 陈根生 | Urinary sediment visible component detection method and system based on lens-free microscopic imaging |
| CN112329537A (en) * | 2020-10-10 | 2021-02-05 | 上海宏勃生物科技发展有限公司 | Method for detecting stool visible components based on yolov3 algorithm |
| CN112598620A (en) * | 2020-11-25 | 2021-04-02 | 哈尔滨工程大学 | Method for identifying transparent tubular type, pathological tubular type and mucus thread in urinary sediment |
| CN113469967A (en) * | 2021-06-28 | 2021-10-01 | 哈尔滨工程大学 | Method for identifying transparent tube type and pathological tube type in urinary sediment |
| WO2023284341A1 (en) * | 2021-07-15 | 2023-01-19 | 北京小蝇科技有限责任公司 | Deep learning-based context-sensitive detection method for urine formed element |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169956A (en) * | 2017-04-28 | 2017-09-15 | 西安工程大学 | Yarn dyed fabric defect detection method based on convolutional neural networks |
| CN107832801A (en) * | 2017-11-23 | 2018-03-23 | 桂林优利特医疗电子有限公司 | A kind of cell image classification model building method |
| CN107871098A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | Method and device for acquiring human face characteristic points |
| CN108614997A (en) * | 2018-04-04 | 2018-10-02 | 南京信息工程大学 | A kind of remote sensing images recognition methods based on improvement AlexNet |
| CN108629369A (en) * | 2018-04-19 | 2018-10-09 | 中南大学 | A kind of Visible Urine Sediment Components automatic identifying method based on Trimmed SSD |
| AU2018101313A4 (en) * | 2018-09-07 | 2018-10-11 | Gao, Jiafan Mr | Image recognition of dangerous tools based on deep learning |
| CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
| CN109460787A (en) * | 2018-10-26 | 2019-03-12 | 北京交通大学 | IDS Framework method for building up, device and data processing equipment |
| CN109740697A (en) * | 2019-03-05 | 2019-05-10 | 重庆大学 | Arena micro-image visible component recognition methods based on deep learning |
| WO2019110583A1 (en) * | 2017-12-07 | 2019-06-13 | Ventana Medical Systems, Inc. | Deep-learning systems and methods for joint cell and region classification in biological images |
-
2019
- 2019-07-09 CN CN201910614832.5A patent/CN110473166A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107871098A (en) * | 2016-09-23 | 2018-04-03 | 北京眼神科技有限公司 | Method and device for acquiring human face characteristic points |
| CN107169956A (en) * | 2017-04-28 | 2017-09-15 | 西安工程大学 | Yarn dyed fabric defect detection method based on convolutional neural networks |
| CN107832801A (en) * | 2017-11-23 | 2018-03-23 | 桂林优利特医疗电子有限公司 | A kind of cell image classification model building method |
| WO2019110583A1 (en) * | 2017-12-07 | 2019-06-13 | Ventana Medical Systems, Inc. | Deep-learning systems and methods for joint cell and region classification in biological images |
| CN108614997A (en) * | 2018-04-04 | 2018-10-02 | 南京信息工程大学 | A kind of remote sensing images recognition methods based on improvement AlexNet |
| CN108629369A (en) * | 2018-04-19 | 2018-10-09 | 中南大学 | A kind of Visible Urine Sediment Components automatic identifying method based on Trimmed SSD |
| CN109086799A (en) * | 2018-07-04 | 2018-12-25 | 江苏大学 | A kind of crop leaf disease recognition method based on improvement convolutional neural networks model AlexNet |
| AU2018101313A4 (en) * | 2018-09-07 | 2018-10-11 | Gao, Jiafan Mr | Image recognition of dangerous tools based on deep learning |
| CN109460787A (en) * | 2018-10-26 | 2019-03-12 | 北京交通大学 | IDS Framework method for building up, device and data processing equipment |
| CN109740697A (en) * | 2019-03-05 | 2019-05-10 | 重庆大学 | Arena micro-image visible component recognition methods based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 李东升: "尿沉渣细胞深度神经网络分类方法研究与实现", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》, no. 2, 15 February 2019 (2019-02-15), pages 060 - 94 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178174A (en) * | 2019-12-12 | 2020-05-19 | 美康盛德医疗科技(苏州)有限公司 | Urine tangible component image identification method based on deep convolutional neural network |
| CN111178174B (en) * | 2019-12-12 | 2023-10-31 | 美康盛德医疗科技(苏州)有限公司 | Urine formed component image identification method based on deep convolutional neural network |
| CN111583292A (en) * | 2020-05-11 | 2020-08-25 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111583292B (en) * | 2020-05-11 | 2023-07-07 | 浙江大学 | Self-adaptive image segmentation method for two-photon calcium imaging video data |
| CN111652326A (en) * | 2020-06-30 | 2020-09-11 | 华南农业大学 | Improved fruit maturity identification method and identification system based on MobileNet v2 network |
| CN111652326B (en) * | 2020-06-30 | 2023-04-28 | 华南农业大学 | Fruit maturity identification method and system based on MobileNet v2 network improvement |
| CN112329537A (en) * | 2020-10-10 | 2021-02-05 | 上海宏勃生物科技发展有限公司 | Method for detecting stool visible components based on yolov3 algorithm |
| CN112200726B (en) * | 2020-10-29 | 2023-04-07 | 陈根生 | Urinary sediment visible component detection method and system based on lensless microscopic imaging |
| CN112200726A (en) * | 2020-10-29 | 2021-01-08 | 陈根生 | Urinary sediment visible component detection method and system based on lens-free microscopic imaging |
| CN112598620A (en) * | 2020-11-25 | 2021-04-02 | 哈尔滨工程大学 | Method for identifying transparent tubular type, pathological tubular type and mucus thread in urinary sediment |
| CN113469967B (en) * | 2021-06-28 | 2022-06-21 | 哈尔滨工程大学 | Method for identifying transparent tube type and pathological tube type in urinary sediment |
| CN113469967A (en) * | 2021-06-28 | 2021-10-01 | 哈尔滨工程大学 | Method for identifying transparent tube type and pathological tube type in urinary sediment |
| WO2023284341A1 (en) * | 2021-07-15 | 2023-01-19 | 北京小蝇科技有限责任公司 | Deep learning-based context-sensitive detection method for urine formed element |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110473166A (en) | A kind of urinary formed element recognition methods based on improvement Alexnet model | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN106248559B (en) | A kind of five sorting technique of leucocyte based on deep learning | |
| CN107680678A (en) | Based on multiple dimensioned convolutional neural networks Thyroid ultrasound image tubercle auto-check system | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| US20090262993A1 (en) | Pathological tissue mapping | |
| CN108764286A (en) | The classifying identification method of characteristic point in a kind of blood-vessel image based on transfer learning | |
| CN108717869A (en) | Diabetic retinopathy diagnosis aid system based on convolutional neural networks | |
| CN110827242A (en) | Intracranial aneurysm detection method and system based on convolutional neural network | |
| CN109961059A (en) | Detect the method and system in kidney tissue of interest region | |
| CN110021425A (en) | A kind of relatively detector and its construction method and cervical cancer cell detection method | |
| CN109359569A (en) | Erythrocyte image sub-classification method based on CNN | |
| CN108629369A (en) | A kind of Visible Urine Sediment Components automatic identifying method based on Trimmed SSD | |
| CN109215040B (en) | Breast tumor segmentation method based on multi-scale weighted learning | |
| CN110265119A (en) | Bone age assessment and prediction of height model, its system and its prediction technique | |
| CN110473167A (en) | A kind of urine sediment image identifying system and method based on deep learning | |
| CN109949304B (en) | Training and acquiring method of image detection learning network, image detection device and medium | |
| CN112263217B (en) | Improved convolutional neural network-based non-melanoma skin cancer pathological image lesion area detection method | |
| CN109670489B (en) | Weak supervision type early senile macular degeneration classification method based on multi-instance learning | |
| CN112215217B (en) | Digital image recognition method and device for simulating doctor to read film | |
| CN114782350A (en) | Multi-modal feature fusion MRI brain tumor image segmentation method based on attention mechanism | |
| Yonekura et al. | Improving the generalization of disease stage classification with deep CNN for glioma histopathological images | |
| CN109671062A (en) | Ultrasound image detection method, device, electronic equipment and readable storage medium storing program for executing | |
| Cao et al. | An automatic breast cancer grading method in histopathological images based on pixel-, object-, and semantic-level features | |
| CN115526834A (en) | Immunofluorescence image detection method and device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191119 |