CN113111796B - Detection and identification method for automatically refining and marking categories of geminizing spores - Google Patents
Detection and identification method for automatically refining and marking categories of geminizing spores Download PDFInfo
- Publication number
- CN113111796B CN113111796B CN202110416482.9A CN202110416482A CN113111796B CN 113111796 B CN113111796 B CN 113111796B CN 202110416482 A CN202110416482 A CN 202110416482A CN 113111796 B CN113111796 B CN 113111796B
- Authority
- CN
- China
- Prior art keywords
- blastospores
- blastospore
- fine classification
- num
- classification category
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 83
- 238000001514 detection method Methods 0.000 title claims abstract description 55
- 238000007670 refining Methods 0.000 title claims description 5
- 238000002372 labelling Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 22
- 230000011218 segmentation Effects 0.000 claims description 22
- 230000034303 cell budding Effects 0.000 claims description 16
- 101100269850 Caenorhabditis elegans mask-1 gene Proteins 0.000 claims description 12
- 230000003044 adaptive effect Effects 0.000 claims description 11
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 238000012850 discrimination method Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 34
- 238000005457 optimization Methods 0.000 description 10
- 230000008901 benefit Effects 0.000 description 4
- 230000000052 comparative effect Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 3
- 210000005000 reproductive tract Anatomy 0.000 description 3
- 241000936934 Kiritimatiellaeota Species 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001000 micrograph Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 230000001717 pathogenic effect Effects 0.000 description 2
- 241000894006 Bacteria Species 0.000 description 1
- 206010046914 Vaginal infection Diseases 0.000 description 1
- 201000008100 Vaginitis Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000005002 female reproductive tract Anatomy 0.000 description 1
- 230000002538 fungal effect Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011206 morphological examination Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 239000013076 target substance Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/695—Preprocessing, e.g. image segmentation
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Geometry (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a detection and identification method for automatically thinning and marking categories of geminiferous spores, which comprises the following steps: acquiring an original image and an expert annotation document; formulating a fine classification rule of the blastospores; designing a self-adaptive mask-based rapid automatic classification method aiming at a fine classification rule formulated by the blastospores; intercepting all marked geminized spore interested areas and storing the intercepted geminized spore interested areas; automatically classifying the intercepted and stored bud spore interested areas according to a formulated fine classification rule and an automatic classification method; constructing a training set with small intra-class difference and large inter-class difference; training a fine classification AI target detection model for detecting the blastospores; and summarizing the blastospores of the fine classification type detected by using the AI target detection model as a final detection result. The method effectively improves the detection rate of the blastospores and reduces the false detection rate of the blastospores; in addition, the method of self-adaptive mask code fast automatic classification is adopted, so that the time and labor cost for manual re-labeling are greatly reduced.
Description
Technical Field
The invention relates to the technical field of intelligent detection and identification of blastospores in microscopic images of gynecological vaginal microecology, in particular to an intelligent detection and identification method aiming at automatic and rapid thinning and labeling categories and divide and conquer of the blastospores in the gynecological vaginal microecology.
Background
Blastospores are a common pathogenic mold in the female reproductive tract, and the spores germinate to form two sporulous pathogenic molds. At present, morphological examination under a microscope is the gold standard for microecological diagnosis of the gynecological genital tract, so that morphological detection of blastospores and improvement of the detection rate of the blastospores play an important role in diagnosing the fungal vaginitis of women.
A traditional machine vision method for detecting the blastospores generally adopts a series of methods such as adaptive threshold segmentation, morphological processing, contour detection, contour geometric shape fitting, screening through geometric characteristics such as perimeter, area and the like of the geometric shapes, the method can detect the blastospores with simple and scattered background distribution, but the shapes of the blastospores are various, and if the blastospores are stacked and distributed or the background is complex, or the blastospores and other types in the microecology such as bacteria and the like are stacked in a cross mode, the geometrical characteristics such as the perimeter, the area, the shape and the like of the blastospores can be influenced by the conditions, so that the traditional machine vision method is easy to cause missed detection of the blastospores.
The artificial labeling based on the blastospores and the deep learning method based on the convolutional neural network can improve the detection rate of the blastospores, greatly overcomes the defects of the traditional machine vision method, and has the following problems: generally, the artificial marking of the blastospores is only marked as a category label, however, as the shapes of the blastospores are very various, namely the intra-category difference is relatively large, and the shapes of the blastospores are relatively small, the intra-category difference of the blastospores is greatly influenced when the blastospores are stacked or the blastospores and other categories in the reproductive tract microecology are stacked in a crossed manner, so that the intra-category difference of the blastospores is enlarged, the trained model can easily cause missing detection, and based on the reasons, the intelligent detection and identification method for automatically thinning and marking the categories is provided, and the detection rate of the blastospores is effectively improved.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects in the prior art, and provide a fine classification category rule of the blastospores according to the form and the background characteristics of the blastospores, wherein the fine classification category has no biomedical significance, but the fine classification has large inter-class difference and small intra-class difference, and a more effective training set can be manufactured according to the rule, so that the trained fine classification target detection model can effectively improve the detection rate of the blastospores and reduce the false detection rate of the blastospores; in addition, the automatic classification method based on the self-adaptive mask is adopted, the fine classification categories of the blastospores are quickly and automatically modified in batches, the time cost and the labor cost of manual re-labeling are greatly reduced, the efficiency is effectively improved, and the economic benefit of enterprises is improved.
The invention is realized by the following technical scheme:
a detection and identification method for automatically refining and labeling categories of geminized spores comprises the following steps:
1) acquiring an original image with blastospores and an expert annotation document containing blastospore annotation information;
2) formulating eight fine classification category rules of the blastospores according to the shapes and the background characteristics of the blastospores;
3) aiming at the fine classification rule formulated in the step 2), designing a self-adaptive mask based rapid discrimination automatic classification method;
4) marking a document by using the original image and the expert in the step 1), intercepting and storing the interested areas of all marked geminizing spores;
5) automatically classifying the bud spore images intercepted and stored in the step 4) according to the fine classification category rules in the step 2) and the automatic classification method in the step 3), and sequentially storing the bud spore images into corresponding folders with the names of the fine classification categories;
6) constructing a training set with small intra-class difference and large inter-class difference;
7) training an AI target detection model for detecting the finely classified bud spores based on the training set constructed in the step 6);
8) detecting the fine classified blastospores by using the fine classified AI target detection model trained in the step 7), and summarizing the fine classified blastospores as a final detected blastospore result.
As optimization, the fast discriminant automatic classification method based on the adaptive mask in the step 3) includes:
3.1) converting the region of interest of the blastospores from a color space to a gray space;
3.2) carrying out binarization segmentation on the gray level image of the blastospore by using an automatic threshold segmentation method, wherein the area of the blastospore is used as a foreground to obtain a binarization image;
3.3) carrying out contour detection on the binary image obtained in the step 3.2), calculating the area of the contour, and selecting the contour with the largest area;
3.4) fitting the minimum circumscribed upright rectangle and the minimum circumscribed rotating rectangle with the maximum outline of the area so as to generate six self-adaptive masks with the same size as the area of interest of the blastospore;
3.5) calculating the area of the minimum external upright rectangle and the height bounngRecthWide boundingRefw(ii) a Calculating the area of the minimum circumscribed rotating rectangle and calculating the area ratio of the minimum circumscribed rotating rectangle to the minimum circumscribed upright rectanglearea(ii) a Calculating boundingRecwAnd boundingRechMin (boundingRect) is the minimum of the twow,boundingRecth) Calculating boundingRefwAnd boundingRechThe maximum value of the two is max (boundingRect)w,boundingRecth) Calculating the ratio
3.7) respectively performing AND operation on the binary image obtained in the step 3.2) and each self-adaptive mask;
3.8) size relationship of height and width, area ratio of the smallest circumscribed upright rectangleareaAnd a threshold valueThe ratio of (1)lengthAnd a threshold valueAnd 3.7) and the operation result are combined to obtain a distinguishing method of each fine classification category, and the fine classification category to which the blastospores belong is judged by the distinguishing method.
As an optimization, the fine classification rule of the blastospores formulated in the step 2) is as follows: eight fine classification categories are established according to the morphology of the blastospores and the background characteristics of the blastospores.
As an optimization, the step 4) intercepts and stores the regions of interest of all the marked blastospores, specifically: setting the name of a microecological microscopic image with budding spores as 'name', storing position coordinate information (xmin, ymin, xmax and ymax) of the budding spores in the image in a corresponding annotation document of the image, intercepting all the budding spores in the image according to the position coordinate information, and storing an area of the intercepted budding spores as an image file with the name 'name _ xmin _ ymin _ xmax _ ymax'; and 5) automatically classifying, and quickly obtaining three information of the blastospores: the image from which the blastospores are derived, the position coordinates of the blastospores and the fine classification category of the blastospores; writing the obtained three information of each blastospore into a new annotation document, and storing the three information so as to construct a training set with small intra-class difference and large inter-class difference; the step 7) training a fine classification AI target detection model for detecting the blastospores based on the convolutional neural network by using a backbone network and combining a deep learning target detection framework method based on the constructed training set of the blastospores; and 8) detecting eight fine classification types of blastospores by using the trained fine classification AI target detection model of the blastospores, and summarizing the fine classification types of blastospores as the final detected result of the blastospores.
As an optimization, the eight fine classification categories of the blastospores in the step 2) are specifically:
fine classification category 1: if two spores of the blastospores are respectively positioned on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and no other target object exists in the upper right corner and the lower left corner in the minimum circumscribed upright rectangle, the fine classification category label of the blastospores meeting the conditions is set as 'tl _ br _ blastospore';
fine classification category 2: if two spores of the blastospores are respectively positioned on the upper right area and the lower left area in the minimum circumscribed upright rectangle, and no other target object is positioned at the upper left corner and the lower right corner in the minimum circumscribed upright rectangle, the fine classification category label of the blastospores meeting the conditions is set as 'tr _ bl _ blastospore';
fine classification category 3: if two spores of a blastospore are arranged up and down in the vertical direction within its smallest circumscribed upright rectangle, a fine classification category label of a blastospore satisfying such a case is set to "vertical _ blastospore";
fine classification category 4: if two spores of blastospores are arranged right and left in the horizontal direction within the smallest circumscribed upright rectangle thereof, a fine classification category label of the blastospores satisfying such a case is set to "horizontal _ blastospore";
fine classification category 5: if two spores of a blastospore are respectively on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and the lower left corner in the minimum circumscribed upright rectangle has other objects and the upper right corner has no other objects, the subdivided category label of the blastospore which satisfies such a condition is assumed to be "tl _ br _ down _ blastoxysore";
fine classification category 6: if two spores of the blastospores are respectively arranged on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and the upper right corner in the minimum circumscribed upright rectangle is provided with other objects, and the lower left corner is not provided with other objects, the fine classification category label of the blastospores meeting the conditions is set as 'tl _ br _ up _ blastospore';
fine classification category 7: if two spores of the blastospores are respectively positioned on the upper right area and the lower left area in the minimum circumscribed upright rectangle, and the upper left corner in the minimum circumscribed upright rectangle is provided with other objects and the lower right corner is provided with no other objects, the fine classification category label of the blastospores meeting the conditions is set as 'tr _ bl _ up _ blastospore';
fine classification category 8: if two spores of a blastospore are respectively on the upper right region and the lower left region within the minimum circumscribed upright rectangle, and there are other objects in the lower right corner and no other objects in the lower left corner within the minimum circumscribed upright rectangle, the classification category label of the blastospore satisfying such a condition is set to "tr _ bl _ down _ blastoxysore".
As optimization, the width of the smallest external upright rectangle is set as bounngselectwHigh is boundingRefhCalculating the area of the minimum external vertical rectangle as bounngrectarea, the area of the minimum external rotation rectangle as rotatedRectarea, and calculating the area ratio
Calculating boundingRecwAnd boundingRechMin (boundingRect) is the minimum of the twow,boundingRecth) Calculating boundingRefwAnd boundingRechThe maximum value of the two is denoted as maxLength max (bounngselect)w,boundingRecth) Calculating the aspect ratio
Setting area ratio thresholdAnd aspect ratio thresholdPreferably, the following components: area ratio threshold Aspect ratio threshold
As an optimization, the six adaptive masks specifically include:
setting the coordinate of the upper left corner point of the smallest external upright rectangle bounngdirector as (x, y), and the width w of the upright rectangle bounngdirector is equal to the bounngdirectorwHigh h equal to boundingRecth;
Mask 1: coordinates of the upper left corner point are (x, y) and width isGao WeiHas a pixel value of 255 and has a coordinate of the upper left corner point of And has a width ofGao WeiHas a pixel value of255, remaining pixel values 0;
mask 2: coordinates of the upper left corner point areAnd has a width ofGao WeiHas a pixel value of 255 and a coordinate of the upper left corner point ofAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 3: coordinates of the upper left corner point are (x, y) and width isGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 4: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 5: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 6: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0. Preferably, the parameter s is set to 4.
As optimization, performing an and operation on the binarized image obtained in step 3.2) and the mask 1, and calculating the number of foreground pixel points of a result after the and operation, and recording the number as Num1;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 2, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num2;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 3, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num3;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 4, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num4;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 5, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num5;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 6, and calculating a foreground image of a result after the AND operationThe number of prime points is recorded as: num6。
As optimization, the discrimination method for eight fine classification categories of the blastospores is as follows:
the method for discriminating the fine classification category 1 comprises the following steps:
if it is usedNum5=0,Num6If 0, the blastospore belongs to the fine classification category 1, namely labeled "tl _ br _ blastospore";
the method for discriminating the fine classification category 2 comprises:
if it is usedNum1<Num2,Num3=0,Num4If 0, the blastospore belongs to the fine classification category 2, labeled "tr _ bl _ blastospore";
the method for discriminating the fine classification category 3 includes:
if it is notminLength=boundingRectw,maxLength=boundingRecth,Then the blastospore belongs to the fine classification category 3, labeled "vertical _ blastposore";
the method for discriminating the fine classification category 4 comprises the following steps:
if it is notminLength=boundingRecth、,maxLength=boundingRectw,The blastospores belong to the fine classification category 4, labeled "horizontal _ blastospore";
the method for discriminating the fine classification category 5 includes:
Num1>Num2,Num5=0,Num6>0, the blastospore belongs to the fine classification category 5, i.e., labeled as
“tl_br_down_blastospore”;
The method for discriminating the fine classification category 6 includes:
Num1>Num2,Num5>0,Num6If 0, the blastospore belongs to the fine classification category 6, labeled "tl _ br _ up _ blastospore";
the method for discriminating the fine classification category 7 includes:
Num1<Num2,Num3>0,Num4If 0, the blastospore belongs to the fine classification category 7, labeled "tr _ bl _ up _ blastospore";
the method for discriminating the fine classification category 8 includes:
if it is notNum1<Num2,Num3=0,Num4>0, the blastospore belongs to the fine classification category 8, labeled "tr _ bl _ down _ blastoxysore".
As optimization, the backbone network in step 7) includes but is not limited to one of VGG-16, VGG-19, ResNet-50, ResNet-101, inclusion V3, mobilenetv2, DarkNet19 and DarkNet 53.
As optimization, the deep learning target detection framework method in the step 7) includes but is not limited to one of SSD, fast-RCNN and YOLO.
The invention has the beneficial effects that:
the method is ingenious in conception, eight fine classification categories of the blastospores are formulated according to the shapes and the background characteristics of the blastospores, the training set of the blastospores with one label with large intra-class difference is converted into a more effective training set of the eight fine classification blastospores with large inter-class difference and small intra-class difference, the problem that the blastospores are stacked in a cross mode or are stacked in a cross mode with other types of targets is effectively solved, the detection rate of the blastospores is effectively improved, and meanwhile the false detection rate of the blastospores is reduced; in addition, by adopting the automatic classification method based on the self-adaptive mask, the fine classification categories of the blastospores are quickly and automatically modified in batches, so that the time cost and the labor cost for manual re-labeling are greatly reduced, the efficiency is effectively improved, and the economic benefit of an enterprise is improved; has better practical application value and popularization value.
Drawings
The following describes a method for detecting and identifying the automatic thinning and labeling type of the geminiferous spores with reference to the accompanying drawings:
FIG. 1 is a schematic flow chart of a method for detecting and identifying the automatically refined marking category of geminiferous spores;
FIG. 2 is a schematic flow chart of the automatic fine classification method of blastospores according to the present invention;
FIG. 3 is a schematic illustration of six masks;
FIG. 4(a) is an exemplary diagram of blastospores belonging to subfraction class 1;
FIG. 4(b) is an exemplary plot of the threshold segmentation result of FIG. 4(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 4(c) is an exemplary diagram of six masks for 4 (a);
FIG. 4(d) is a diagram of an example of the first diagram of FIG. 4(b) and the result of the six mask ANDing operations of FIG. 4 (c);
FIG. 5(a) is an exemplary diagram of blastospores belonging to subfraction class 2;
FIG. 5(b) is an exemplary plot of the threshold segmentation result of FIG. 5(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 5(c) is an exemplary diagram of six masks for 5 (a);
FIG. 5(d) is a diagram of an example of the first diagram of FIG. 5(b) and the result of the six mask ANDing operations of FIG. 5 (c);
FIG. 6(a) is an exemplary diagram of blastospores belonging to subfraction class 3;
FIG. 6(b) is an exemplary plot of the threshold segmentation result of FIG. 6(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 7(a) is an exemplary diagram of blastospores belonging to subfraction class 4;
FIG. 7(b) is an exemplary plot of the threshold segmentation result of FIG. 7(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 8(a) is an exemplary diagram of blastospores belonging to subfraction class 5;
FIG. 8(b) is an exemplary plot of the threshold segmentation result of FIG. 8(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 8(c) is an exemplary diagram of six masks for 8 (a);
FIG. 8(d) is an exemplary graph of the first graph of FIG. 8(b) and the result of the six mask ANDing operations of FIG. 8 (c);
FIG. 9(a) is an exemplary diagram of blastospores belonging to subfraction class 6;
FIG. 9(b) is an exemplary plot of the threshold segmentation result of FIG. 9(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 9(c) is an exemplary diagram of six masks for 9 (a);
FIG. 9(d) is a diagram of an example of the first diagram of FIG. 9(b) and the result of the six mask ANDing operations of FIG. 9 (c);
FIG. 10(a) is an exemplary diagram of blastospores belonging to subfraction 7;
FIG. 10(b) is an exemplary graph of the threshold segmentation result of FIG. 9(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 10(c) is an exemplary diagram of six masks for 10 (a);
FIG. 10(d) is a diagram of an example of the first diagram of FIG. 10(b) and the result of the six mask ANDing operations of FIG. 10 (c);
FIG. 11(a) is an exemplary diagram of blastospores belonging to fine classification 8;
FIG. 11(b) is an exemplary graph of the threshold segmentation result of FIG. 11(a), and its minimum bounding upright rectangle and minimum bounding rotated rectangle;
FIG. 11(c) is an exemplary diagram of six masks for 11 (a);
FIG. 11(d) is a diagram of an example of the first diagram of FIG. 11(b) and the result of the sequential AND operation of the six masks of FIG. 11 (c);
FIG. 12(a) is an exemplary image of a microecological micrograph of a budding spore from the gynecological tract;
FIG. 12(b) is an exemplary plot of blastospores labeled by the expert in FIG. 12 (a);
FIG. 12(c) is an exemplary view of modifying the label of FIG. 12(b) for identifying blastospores to a fine classification label;
FIG. 13(a) is an exemplary image of a microecological micrograph of a budding spore from the gynecological tract;
FIG. 13(b) is an exemplary plot of blastospores labeled by the expert in FIG. 13 (a);
FIG. 13(c) is an exemplary view of modifying the label of FIG. 13(b) for identifying blastospores to a fine classification label;
FIG. 14(a) is an exemplary diagram of a few blastospores with cross-stacking between blastospores;
FIG. 14(b) is an exemplary plot of AI models with only one blastospore tag against the blastospores detected in FIG. 14 (a);
FIG. 14(c) is a diagram illustrating an example of eight fine classification blastospores detected in accordance with the present invention for 14 (a);
FIG. 15(a) is an exemplary diagram in which blastospores are large and the blastospores are stacked with each other while crossing each other;
FIG. 15(b) is an exemplary plot of the AI model with only one blastospore tag against the blastospores detected in FIG. 15 (a);
FIG. 15(c) is a diagram illustrating an example of eight fine classification blastospores detected in accordance with the present invention for 15 (a);
FIG. 16(a) is an exemplary diagram of a large number of blastospores stacked with other objects crossing each other;
FIG. 16(b) is an exemplary plot of the AI model with only one blastospore tag against the blastospores detected in FIG. 16 (a);
FIG. 16(c) is an exemplary graph of eight fine-classified blastospores detected in the present invention for 16 (a).
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and examples.
Referring to fig. 1, a method for detecting and identifying the automatic thinning and labeling type of geminispores includes the following steps:
the method comprises the following steps: acquiring the collected gynecological genital tract microecological microscopic original image with the blastospore and a corresponding labeling document which is labeled by experts in related fields and contains the blastospore labeling information. Assuming that the category label of the budding spores labeled in the labeling document is "blastospore", the labeling document comprises position coordinates (xmin, ymin, xmax, ymax) of all budding spores in the corresponding image, wherein xmin represents the abscissa of the upper left corner point of the budding spore interesting region, ymin represents the ordinate of the upper left corner point of the budding spore interesting region, xmax represents the abscissa of the lower right corner of the budding spore interesting region, and ymax represents the ordinate of the lower right corner of the budding spore interesting region.
Step two: establishing a fine classification rule of the blastospores: the fine classification is carried out according to the morphology of the geminizing spores and the background characteristics of the geminizing spores, and is divided into eight fine classification categories which have no biomedical significance, wherein the eight fine classification categories are as follows:
fine classification category 1: if two spores of blastospores are respectively positioned on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and neither the upper right nor the lower left area in the minimum circumscribed upright rectangle has other objects, it is assumed that the fine classification category label of the blastospores satisfying such a condition is "tl _ br _ blastospore", such as the blastospores shown in fig. 4 (a);
fine classification category 2: if two spores of blastospores are respectively located on the upper right region and the lower left region in the minimum circumscribed upright rectangle, and neither of the upper left corner and the lower right corner in the minimum circumscribed upright rectangle has other objects, it is assumed that the fine classification category label of the blastospores satisfying such a condition is "tr _ bl _ blastospore", such as the blastospores shown in fig. 5 (a);
fine classification category 3: if two spores of blastospores are arranged up and down in the vertical direction within its smallest circumscribed upright rectangle, it is assumed that the fine classification category label of the blastospores satisfying such a case is "vertical _ blastospore" blastospores as shown in fig. 6 (a);
fine classification category 4: if two spores of blastospores are arranged right and left in the horizontal direction within the smallest circumscribed upright rectangle thereof, it is assumed that the fine classification category label of the blastospores satisfying such a case is "horizontal _ blastospore", the blastospores shown in fig. 7 (a);
fine classification category 5: if two spores of a blastospore are respectively on the upper left region and the lower right region within the minimum circumscribed upright rectangle, and there are other objects in the lower left corner and no other objects in the upper right corner within the minimum circumscribed upright rectangle, it is assumed that the classification type label of the blastospore satisfying such a case is "tl _ br _ down _ blastosore" as shown in fig. 8 (a);
fine classification category 6: if two spores of blastospores are respectively positioned on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and the upper right area in the minimum circumscribed upright rectangle has other objects and the lower left area has no other objects, it is assumed that the fine classification category label of the blastospores satisfying such a condition is "tl _ br _ up _ blastospore", such as the blastospores shown in fig. 9 (a);
fine classification category 7: if two spores of blastospores are respectively on the upper right region and the lower left region within the minimum circumscribed upright rectangle, and there are other objects in the upper left corner and no other objects in the lower right corner within the minimum circumscribed upright rectangle, it is assumed that the classification category label of the blastospores satisfying such a case is "tr _ bl _ up _ blastospore", such as the blastospores shown in fig. 10 (a);
fine classification category 8: if two spores of a blastospore are on the upper right and lower left regions, respectively, within its smallest circumscribed upright rectangle with other targets in the lower right corner and no other targets in the lower left corner, it is assumed that a fine classification category label of a blastospore satisfying such a condition is "tr _ bl _ down _ blastoxyspore" as shown in fig. 11 (a).
Step three: and designing a self-adaptive mask-based rapid discrimination automatic classification method aiming at the eight fine classification categories in the step two, and automatically classifying the blastospores. Comparing the step one, in which the expert marks the blastospores as only one category label, because the step two has no biomedical significance for the fine category of the blastospores, the expert in the field cannot mark the blastospores according to the eight fine category labels, if the category labels of the blastospores are manually modified according to the blastospores marked by the expert and the eight fine category labels of the step two, not only is the time cost and the labor cost of marking increased, but also because the fine categories are too many, the manual marking is very easy to cause errors again, therefore, the method for quickly judging and automatically classifying the blastospores marked by the expert based on the self-adaptive mask code is adopted to finely classify the blastospores marked by the expert, the flow of the whole method is shown in fig. 2, and the specific method comprises the following steps:
the first step is as follows: the region of interest of the blastospores is converted from color space to gray space.
The second step is that: and (3) performing binary segmentation on the gray level image of the blastospore by using an automatic threshold segmentation method, such as OTSU automatic threshold segmentation, wherein the area of the blastospore is taken as a foreground, namely the gray level value of the blastospore area is 255, and obtaining the binary image.
The third step: and (4) carrying out contour detection on the binary image obtained in the second step, calculating the area of the contour, and selecting the contour with the largest contour area.
The fourth step: and fitting the minimum circumscribed upright rectangle bounngdark and the minimum circumscribed rotating rectangle rotadrect of the outline with the maximum area in the third step, and further generating six self-adaptive masks with the same size as the area of interest of the blastospore.
The fifth step: setting the width of the smallest external upright rectangle bounngRact as bounngRactwHigh is boundingRefhCalculating the area of the minimum external vertical rectangle as bounngRecrectArea, the area of the minimum external rotating rectangle as rotatedRecrectArea, and calculating the proportionCalculating boundingRecwAnd boundingRechThe minimum value of these two is minLength min (boundingRec)w,boundingRecth) Calculating boundingRefwAnd boundingRechThe maximum value of the two is denoted as maxLength max (bounngselect)w,boundingRecth) Calculating the aspect ratio Setting the coordinate of the upper left corner point of the smallest external upright rectangle bounngdirector as (x, y), and the width w of the upright rectangle bounngdirector is equal to the bounngdirectorwHigh h ═ boundingRecth。
The seventh step: and (3) performing AND operation on the binary image obtained in the second step and six self-adaptive masks with the same size as the binary image, setting the coordinates of the upper left corner point of the smallest external upright rectangle bounngdirector in the fourth step as (x, y), and setting the width w of the upright rectangle equal to bounngdirectorwHigh h equal to boundingRecthThe shapes of the six adaptive masks are roughly as shown in fig. 3, and are as follows:
mask 1: the coordinate of the upper left corner point is (x, y) and a width ofGao WeiHas a pixel value of 255 and a coordinate of the upper left corner point of And has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 2: coordinates of the upper left corner point areAnd has a width ofGao WeiHas a pixel value of 255 and has a coordinate of the upper left corner point ofAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 3: coordinates of the upper left corner point are (x, y) and width isGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 4: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 5: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 6: coordinates of point at the upper left corner areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0. Preferably, the parameter s is set to 4.
And the binary image and the mask 1 in the second step are subjected to AND operation, and the pixel points of the foreground of the result after the AND operation are calculatedNumber, denoted as Num1;
And the binary image and the mask 2 in the second step are subjected to AND operation, the number of pixel points of the foreground of the result after the AND operation is calculated and recorded as Num2;
And the binary image and the mask 3 in the second step are subjected to AND operation, the number of pixel points of the foreground of the result after the AND operation is calculated and recorded as Num3;
And the binary image and the mask 4 in the second step are subjected to AND operation, the number of pixel points of the foreground of the result after the AND operation is calculated and recorded as Num4;
And (5) performing AND operation on the binary image and the mask 5, calculating the number of pixel points of the foreground of the AND operation result, and recording the number as Num5;
And the binary image and the mask 6 in the second step are subjected to AND operation, and the number of pixel points of the foreground of the result after the AND operation is calculated and recorded as: num6;
The eighth step: ratio of area to width of the smallest circumscribed upright rectangleareaAnd a threshold valueLength to width ratio oflengthAnd a threshold valueThe classification method of each fine classification category is obtained by combining the size relationship of the spore and the operation result, and the fine classification category to which the blastospores belong is obtained by the judgment method. The specific discrimination method is as follows:
the method for discriminating the fine classification category 1 comprises the following steps:
if it is notNum1>Num2,Num5=0,Num6If 0, the blastospore belongs to subfraction class 1, namely labeled "tl _ br _ blastospore";
taking the blastospore of FIG. 4(a) as an example, the blastospore symbolAnd synthesizing the characteristics of the fine category 1 described in the step two. The three graphs in FIG. 4(b) are the results of the automatic thresholding for the OTSU of FIG. 4(a), and the minimum bounding upright rectangle and minimum bounding rotated rectangle of the outline with the largest area, the upright rectangle having an area of 2907, the rotated rectangle having an area of 2160, and the area ratioAnd satisfySix adaptive masks for fig. 4(a) are sequentially generated based on the minimum circumscribed rectangle with the maximum outline area and the size of fig. 4(a), and as shown in fig. 4(c), the first binary image of fig. 4(b) and the six masks of fig. 4(c) are sequentially and-operated, and the result is sequentially as shown in fig. 4(d), the first image of fig. 4(d) is the result of the and-operation of the binary image and the mask 1, the second image of fig. 4(d) is the result of the and-operation of the binary image and the mask 2, the fifth image of fig. 4(d) is the result of the and-operation of the binary image and the mask 5, and the sixth image of fig. 4(d) is the result of the and-operation of the binary image and the mask 6, it is obvious that Num1>Num2,Num5=0,Num6When the number is 0, the blastospores of fig. 4(a) belong to fine classification 1, i.e., the label after fine classification is "tl _ br _ blastospore";
the method for discriminating the fine classification category 2 comprises:
if it is notNum1<Num2,Num3=0,Num4If 0, the blastospore belongs to subfraction class 1, labeled "tr _ bl _ blastospore";
taking the blastospores of fig. 5(a) as an example, the blastospores conform to the characteristics of fine category 2 described in step two. The three graphs in FIG. 5(b) are the results of the automatic threshold segmentation of the OTSU in FIG. 5(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the largest outline of their areas, the upright rectangle having an area of 3404, and the rotated rectangle having an area of 34042511, area ratioSatisfy the requirement ofSix adaptive masks for fig. 5(a) are sequentially generated according to the minimum circumscribed upright rectangle of the outline with the maximum area and the size of fig. 5(a), as shown in fig. 5(c), the first binary image of fig. 5(b) and the six masks of fig. 5(c) are sequentially subjected to and operation, and the result is sequentially shown in fig. 5(d), the first image of fig. 5(d) is the result of and operation of the binary image and the mask 1, the second image of fig. 5(d) is the result of and operation of the binary image and the mask 2, the third image of fig. 5(d) is the result of and operation of the binary image and the mask 3, and the fourth image of fig. 5(d) is the result of and operation of the binary image and the mask 4, obviously, Num1<Num2,Num3=0,Num4When equal to 0, the blastospores of fig. 5(a) belong to subfraction class 2, i.e., after the subfraction class label is "tr _ bl _ blastospore";
the method for discriminating the fine classification category 3 includes:
if it is notminLength=boundingRectw,maxLength=boundingRecth,Then the blastospore belongs to fine category 3, labeled "vertical _ blastposore";
taking the blastospores of fig. 6(a) as an example, the blastospores conform to the characteristics of fine classification 3 described in step two. The three graphs in FIG. 6(b) are the results of the automatic threshold segmentation of the OTSU in FIG. 6(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the maximum outline of their areas, the area of the upright rectangle is 2790, the area of the rotated rectangle is 2664, and the area ratio is 2664Satisfy the requirement ofWide bounngselect of upright rectanglew30, high boundingRefh93, apparently minLength boundiglectw,maxLength=boundingRecth, The blastospores of fig. 6(a) belong to fine category 3, i.e., labeled "vertical _ blastospore" after fine category;
the method for discriminating the fine classification category 4 comprises the following steps:
if it is notminLength=boundingRecth,maxLength=boundingRectw,The blastospores belong to fine category 4, labeled "horizontal _ blastospore";
taking the blastospores of fig. 7(a) as an example, the blastospores conform to the characteristics of fine classification 4 described in step two. The three graphs in FIG. 7(b) are the results of the automatic threshold segmentation of the OTSU of FIG. 7(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the maximum outline of their areas, the upright rectangle having an area of 2916, the rotated rectangle having an area of 2451 and the area ratio ofSatisfy the requirement ofWide bounngselect of upright rectanglew81, high bounngrecth36, apparently minLength boundiglecth,maxLength=boundingRectw, The blastospores of fig. 7(a) belong to fine category 4, i.e., labeled "horizontal _ blastospore" after fine category;
the method for discriminating the fine classification category 5 includes:
Num1>Num2,Num5=0,Num6>0, the blastospore belongs to the fine category 5, namely the label is 'tl _ br _ down _ blastospore';
taking the blastospores of fig. 8(a) as an example, the blastospores conform to the characteristics of fine classification 5 described in step two. The three graphs in FIG. 8(b) are the results of the automatic thresholding for the OTSU of FIG. 8(a), and their minimum bounding upright rectangle and minimum bounding rotated rectangle with the largest outline in area, the upright rectangle having an area of 3432, the rotated rectangle having an area of 3315, and the area ratio ofSatisfy the requirement ofWide baoudingRec of upright rectangletw is 52, high bounngselecth66, minLength is apparently boundingetw,maxLength=boundingRecth, The six adaptive masks for fig. 8(a) are sequentially generated according to the minimum circumscribed vertical rectangle of the outline with the largest area and the size of fig. 8(a), and as shown in fig. 8(c), the binarized map of the first one of fig. 8(b)The image and the six masks of fig. 8(c) are and-operated in sequence, and the results are sequentially shown in fig. 8(d), the first diagram of fig. 8(d) is the result of the and-operation of the binarized image and the mask 1, the second diagram of fig. 8(d) is the result of the and-operation of the binarized image and the mask 2, the fifth diagram of fig. 8(d) is the result of the and-operation of the binarized image and the mask 5, and the sixth diagram of fig. 8(d) is the result of the and-operation of the binarized image and the mask 6, and Num is obviously1>Num2,Num5=0,Num6>0, the blastospores of fig. 8(a) belong to fine classification 5, i.e., after fine classification, the label is "tl _ br _ down _ blastospore";
the method for discriminating the fine classification category 6 includes:
Num1>Num2,Num5>0,Num6If equal to 0, the blastospore belongs to the fine category 6, labeled "tl _ br _ up _ blastospore";
taking the blastospores of fig. 9(a) as an example, the blastospores conform to the characteristics of fine classification 6 described in step two. The three graphs in FIG. 9(b) are the results of the automatic threshold segmentation of the OTSU of FIG. 9(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the largest outline of their areas, the upright rectangle having an area of 4402, the rotated rectangle having an area of 4270, and the area ratio of the rotated rectanglesSatisfy the requirement ofWide bounngselect of upright rectanglew71, high bounngrecth62, apparently minLength boundiglecth,maxLength=boundingRectw, The six adaptive masks for fig. 9(a) are sequentially generated according to the minimum circumscribed upright rectangle of the outline with the maximum area and the size of fig. 9(a), as shown in fig. 9(c), the first binary image of fig. 9(b) and the six masks of fig. 9(c) are sequentially subjected to and operation, and the result is sequentially shown in fig. 9(d), the first image of fig. 9(d) is the result of and operation of the binary image and the mask 1, the second image of fig. 9(d) is the result of and operation of the binary image and the mask 2, the fifth image of fig. 9(d) is the result of and operation of the binary image and the mask 5, and the sixth image of fig. 9(d) is the result of and operation of the binary image and the mask 6, obviously, Num1>Num2,Num5>0,Num6When the number of spores is 0, the blastospores of fig. 9(a) belong to the fine classification 6, i.e., the label after the fine classification is "tl _ br _ up _ blastospore"
The method for discriminating the fine classification category 7 includes:
Num1<Num2,Num3>0,Num4If 0, the blastospore belongs to fine classification 7, labeled "tr _ bl _ up _ blastospore";
taking the blastospores of fig. 10(a) as an example, the blastospores conform to the characteristics of fine classification 7 described in step two. The three graphs in FIG. 10(b) are the results after the automatic threshold segmentation of the OTSU in FIG. 10(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the largest outline of their areas, the upright rectangle having an area of 4096, the rotated rectangle having an area of 3969, and the area ratio Satisfy the requirement ofWide boundingRef of upright rectanglew=64High boundingRefh64, apparently minLength boundiglecth,maxLength=boundingRectw, The six adaptive masks for fig. 10(a) are sequentially generated based on the minimum circumscribed rectangle with the maximum outline area and the size of fig. 10(a), and as shown in fig. 10(c), the first binary image of fig. 10(b) and the six masks of fig. 10(c) are sequentially and-operated, and the result is sequentially as shown in fig. 10(d), the first image of fig. 10(d) is the result of the and-operation of the binary image and the mask 1, the second image of fig. 10(d) is the result of the and-operation of the binary image and the mask 2, the third image of fig. 10(d) is the result of the and-operation of the binary image and the mask 3, and the fourth image of fig. 10(d) is the result of the and-operation of the binary image and the mask 4, it is obvious that Num1<Num2,Num3>0,Num4When the value is 0, the blastospores in fig. 9(a) belong to fine classification 7, i.e., the label after fine classification is "tr _ bl _ up _ blastospore";
the method for discriminating the fine classification category 8 includes:
Num1<Num2,Num3=0,Num4>0, the blastospore belongs to the fine classification 8, namely, the label is 'tr _ bl _ down _ blastospore';
taking the blastospores of fig. 11(a) as an example, the blastospores conform to the characteristics of fine classification 8 described in step two. The three graphs in FIG. 11(b) are the results after the OTSU automatic threshold segmentation in FIG. 11(a), and the minimum bounding upright rectangle and the minimum bounding rotated rectangle with the largest outline of their areas, the upright rectangle having an area of 4686, the rotated rectangle having an area of 4550, and the area ratio Satisfy the requirement ofWide bounngselect of upright rectanglew66, high bounngrecth71, apparently minLength boundgractw,maxLength=boundingRecth, The six adaptive masks for fig. 11(a) are sequentially generated based on the minimum circumscribed rectangle with the maximum outline of the area and the size of fig. 11(a), and as shown in fig. 11(c), the first graph of fig. 11(b) is the result of the and operation of the binarized image with the mask 1, the second graph of fig. 11(d) is the result of the and operation of the binarized image with the mask 2, the third graph of fig. 11(d) is the result of the and operation of the binarized image with the mask 3, and the fourth graph of fig. 11(d) is the result of the and operation of the binarized image with the mask 4, it is obvious that Num1<Num2,Num3=0,Num4>0, the blastospores of FIG. 11(a) belong to fine classification 8, i.e., after fine classification the label is "tr _ bl _ down _ blastospore";
step four: and (4) intercepting all marked interesting regions of the blastospores by utilizing the microecological microscopic image with the blastospores in the step one and the corresponding expert marking document containing the blastospore marking information, and storing the interesting regions in a magnetic disk. Assuming that the name of the microecological microscopic image with the blastospores is 'name', the corresponding annotation document of the image stores the position coordinate information (xmin, ymin, xmax, ymax) of the blastospores in the image, intercepts all the blastospores in the image according to the position coordinate information, and stores the intercepted area of the blastospores as an image file with the name 'name _ xmin _ ymin _ xmax _ ymax'.
Step five: and automatically classifying the bud spore images intercepted and stored in the fourth step according to the eight fine classification categories in the second step and the automatic classification method in the third step, and sequentially storing the bud spore images into corresponding folders with the eight fine classification categories as names. Three information of the blastospores are effectively and rapidly obtained through the steps: the method is more efficient and easier to modify than manual re-labeling, time cost and labor cost of manual re-labeling are greatly reduced, and economic benefits of enterprises are improved.
Fig. 12(a) and 13(a) are diagrams illustrating the results of classifying the 11 blastospores artificially marked in fig. 12(b) and the 12 blastospores artificially marked in fig. 13(b) finely by the automatic classification method of step three.
The results of the automatic fine classification of blastospores in FIG. 12(b) are shown in FIG. 12(c), and the number of each fine classification is shown in the following table:
fine category classification | Number of |
Subdivision class 1, label "tl _ br _ blastospore" | 3 |
Subdivision class 2, labeled tr _ bl _ blastscore " | 1 |
Subclass 3, labeled "vertical _ blastospore" | 2 |
Subdivision class 4, label "horizontal _ blastscore" | 4 |
|
1 |
Subdivision class 6, label "tl _ br _ up _ blastscore" | 0 |
Subdivision class 7, labeled "tr _ bl _ up _ blastscore" | 0 |
Subclass 8, labeled "tr _ bl _ down _ blastospore" | 0 |
The results of the automatic fine classification of blastospores in FIG. 13(b) are shown in FIG. 13(c), and the number of each fine classification is shown in the following table:
sub-category of sub-category | Number of |
Subdivision class 1, labeled "tl _ br _ blastoxysore" | 2 |
Subclass 2, labeled "tr _ bl _ blastospore" | 2 |
Subclass 3, labeled "vertical _ blastospore" | 1 |
Subdivision class 4, label "horizontal _ blastscore" | 3 |
|
0 |
Subdivision class 6, label "tl _ br _ up _ blastospore" | 1 |
Subdivision class 7, labeled "tr _ bl _ up _ blastscore" | 2 |
Subclass 8, labeled "tr _ bl _ down _ blastospore" | 1 |
Step six: and obtaining three information of each blastospore according to the fifth step, writing the three information into a new annotation document, and storing the annotation document. The training set of the geminispores related to the eight fine classification categories is constructed in the step, the training set is large in inter-category difference and small in intra-category difference, and an AI target detection model can be trained more conveniently.
Step seven: and training a fine classification AI target detection model for detecting eight fine classification blastospores by adopting the combination of ResNet-50+ SSD based on the training set of the blastospores constructed in the sixth step.
Step eight: and (4) detecting eight kinds of subdivided blastospores by using the subdivided AI target detection model of the blastospores trained in the step seven, and then summarizing the results of the eight kinds of subdivided blastospores as the final result of detecting the blastospores.
Comparative example: and (3) only using one label for all the blastospores, setting the label type labels of the blastospores to be 'blastospore', and training an AI target detection model by using the same network architecture of the step seven.
In order to more clearly embody the specific enhancement effect of the invention, three types of common complex scenes are selected:
the first scenario is as follows: the number of blastospores is small, and the blastospores are stacked in a cross manner, as shown in fig. 14 (a); the results of comparing the comparative examples with the invention are shown in table 1:
as can be seen from table 1, 4 blastospores were detected for fig. 14(a) using the AI object detection model trained using only one class label, as shown in fig. 14(b), 4 blastospores that were cross-stacked were missed, as shown in region of interest 1 of fig. 14(b), whereas 7 blastospores were detected for fig. 14(a) according to the present invention, as shown in fig. 14(c), wherein 1 "tl _ br _ blastscore", 0 "tr _ bl _ blastscore", 2 "vertical _ blastscore", 2 "horizontal _ blastscore", 0 "tl _ br _ downblastscore", 0 "tl _ br _ upblast _ blastscore", 1 "trblbl _ up _ blastblast", as shown in the following table
Fine category classification | Number of finely divided blastospores detected |
Subdivision class 1 labeled "tl _ br _ blastscore | 1 |
Subdivision class 2 labeled "tr _ bl _ |
0 |
Fine Classification 3 labeled "vertical _ blastscore | 2 |
Subfraction class 4 labeled "horizontal _ blastscore | 2 |
|
0 |
Fine category 6 labeled "tl _ br _ up _ blastscore | 0 |
Subdivided class 7 labeled "tr _ bl _ up _ blastscore | 1 |
Fine classification 8 labeled "tr _ bl _ down _ blastscore | 1 |
That is, 7 blastospores were detected in total, while the blastospores stacked together with the missed cross-colonies of interest 1 of FIG. 14(b) were detected, as shown in region of interest 1 of FIG. 14 (c).
The second scenario is as follows: there were many blastospores and there were scattered distributions and cross-stacked distributions, as shown in fig. 15(a), and the results of comparing the comparative example with the present invention are shown in table 2:
as can be seen from table 2, 12 blastospores were detected in fig. 15(a) using the AI object detection model trained using only one class label for the blastospores, as shown in fig. 15(b), and the blastospores were missed, e.g., the blastospores cross-stacked with other objects in the regions of interest 1 and 2 in fig. 15(b) were missed, whereas the present invention detected more 16 blastospores in fig. 15(a), and detected 2 "tl _ br _ blastscore", 5 "tr _ bl _ blastscore", 4 "vertical _ blastscore", 4 "horizontal _ blastscore", 0 "tl _ br _ down _ blastscore", 1 "tl _ br _ up _ blastscore", 0 "trjblup _ blastup _ score", as follows:
fine category classification | Number of finely divided blastospores detected |
Subdivision class 1 labeled "tl _ br _ blastscore | 2 |
Subdivision class 2 labeled "tr _ bl _ |
5 |
Sub-category 3 labeled "vertical _ blastospore | 4 |
Sub-category 4 labeled "horizontal _ blastscore | 4 |
|
0 |
Fine category 6 labeled "tl _ br _ up _ blastscore | 1 |
Subdivided class 7 labeled "tr _ bl _ up _ |
0 |
Fine classification 8 labeled "tr _ bl _ down _ |
0 |
That is, 16 blastospores were detected in total, and simultaneously, the cross-stacked blastospores that were missed in the regions of interest 1 and 2 in fig. 15(b) were detected, as shown in the regions of interest 1 and 2 in fig. 15 (c).
The third scenario is as follows: the number of blastospores was large, and the number of blastospores was distributed in a cross-stacked manner with hyphae of other target substances, as shown in fig. 16 (a). The results of comparing the comparative examples with the present invention are shown in Table 3:
as can be seen from table 3, for blastospores, 5 blastospores were detected in fig. 16(a) using an AI object detection model trained using only one class label, as shown in fig. 16(b), blastospores were missed, e.g., blastospores stacked with other object hyphae in region of interest 1 in fig. 16(b) were missed, whereas the present invention detected 9 more blastospores in fig. 16(c), 0 "tl _ br _ blastscore", 0 "tr _ bl _ score", 5 "vertical _ blast", 2 "horizontal _ blast", 0 "tl _ br _ down _ blast", 1 "tl _ br _ up _ blast", 0 "tr _ up _ blast", and 1 "tr _ up _ blast", as follows:
fine category classification | Number of finely divided blastospores detected |
Subdivision class 1 labeled "tl _ br _ |
0 |
Subdivision class 2 labeled "tr _ bl _ |
0 |
Fine Classification 3 labeled "vertical _ |
5 |
Subfraction class 4 labeled "horizontal _ blastscore | 2 |
|
0 |
Sub-category 6 labeled "tl _ br _ up _ blastospore | 1 |
Subdivided class 7 labeled "tr _ bl _ up _ |
0 |
Fine classification 8 labeled "tr _ bl _ down _ blastscore | 1 |
That is, 9 blastospores were detected in total, and simultaneously, blastospores stacked with other objects crossing each other were detected as the missed detection of interest 1 in fig. 16(b), as shown in interest 1 in fig. 16 (c).
Although the present invention has been described with reference to the preferred embodiments, it is not limited thereto. Various changes and modifications within the scope of the present invention may be made by those skilled in the art without departing from the scope of the present invention, and therefore the scope of the present invention should be determined by the appended claims.
Claims (8)
1. An intelligent detection and identification method for automatically refining and labeling categories of geminized spores is characterized by comprising the following steps:
1) acquiring an original image with blastospores and an expert annotation document containing blastospore annotation information;
2) formulating a fine classification rule of the blastospores according to the shapes and the background characteristics of the blastospores;
3) aiming at the fine classification rule formulated in the step 2), designing a self-adaptive mask based rapid discrimination automatic classification method;
4) marking a document by using the original image and the expert in the step 1), intercepting and storing the interested areas of all marked geminizing spores;
5) automatically classifying the bud spore images intercepted and stored in the step 4) according to the fine classification category rules in the step 2) and the automatic classification method in the step 3), and sequentially storing the bud spore images into corresponding folders with the names of the fine classification categories;
6) constructing a training set with small intra-class difference and large inter-class difference;
7) training an AI target detection model for detecting the finely classified bud spores based on the training set constructed in the step 6);
8) detecting the fine classified blastospores by using the fine classified AI target detection model trained in the step 7), and summarizing the fine classified blastospores as a final detected blastospore result;
the fast automatic classification method based on the self-adaptive mask in the step 3) comprises the following steps:
3.1) converting the region of interest of the blastospores from a color space to a gray space;
3.2) carrying out binarization segmentation on the gray level image of the blastospore by using an automatic threshold segmentation method, wherein the area of the blastospore is used as a foreground to obtain a binarization image;
3.3) carrying out contour detection on the binary image obtained in the step 3.2), calculating the area of the contour, and selecting the contour with the largest area;
3.4) fitting the minimum circumscribed upright rectangle and the minimum circumscribed rotating rectangle with the maximum outline of the area so as to generate six self-adaptive masks with the same size as the area of interest of the blastospore;
3.5) calculating the area of the minimum external upright rectangle and the height bounngRecthWide boundingRefw(ii) a Calculating the area of the minimum circumscribed rotating rectangle and calculating the area ratio of the minimum circumscribed rotating rectangle to the minimum circumscribed upright rectanglearea(ii) a Calculating boundingRecwAnd boundingRechMin (boundingRec) is the minimum of the twow,boundingRecth) Calculating boundingRefwAnd boundingRefhThe maximum value of the two is denoted as maxLength max (bounngselect)w,boundingRecth) Calculating the ratio
3.7) respectively carrying out AND operation on the binary image obtained in the step 3.2) and six self-adaptive masks;
3.8) size relationship of height and width, area ratio of the smallest circumscribed upright rectangleareaAnd a threshold valueThe ratio of (1)lengthAnd a threshold valueThe size relationship of the spore is combined with the operation result of 3.7) to obtain a distinguishing method of each fine classification category, and the fine classification category to which the blastospores belong is judged through the distinguishing method; the six adaptive masks specifically include:
setting the coordinate of the upper left corner point of the smallest external upright rectangle bounngdirector as (x, y), and the width w of the upright rectangle bounngdirector is equal to the bounngdirectorwHigh h ═ boundingRecth;
Mask 1: coordinates of the upper left corner point are (x, y) and width isGao WeiHas a pixel value of 255 and has a coordinate of the upper left corner point ofAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 2: coordinates of the upper left corner point areAnd has a width ofGao WeiHas a pixel value of 255 and a coordinate of the upper left corner point ofAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 3: coordinates of the upper left corner point are (x, y) and width isGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 4: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
mask 5: coordinates of the upper left corner point areAnd has a width ofGao WeiThe pixel value in the region of (1) is 255, and the remaining pixel values are 0;
2. The intelligent detection and identification method for automatically refining and labeling categories of blastospores as claimed in claim 1, wherein said step 4) intercepts all regions of interest of labeled blastospores and saves them specifically as follows: setting the name of a microecological microscopic image with budding spores as 'name', storing position coordinate information (xmin, ymin, xmax, ymax) of the budding spores in the image in a corresponding annotation document of the image, intercepting all the budding spore interested areas in the image according to the coordinate information, and storing the intercepted budding spore areas as an image file with the name 'name _ xmin _ ymin _ xmax _ ymax'; and 5) automatically classifying, and quickly obtaining three information of the blastospores: which image the blastospore comes from, the position coordinates of the blastospore, and the fine classification category of the blastospore; writing the obtained three information of each blastospore into a new annotation document, and storing the three information so as to construct a training set with small intra-class difference and large inter-class difference; the step 7) training a subdivided AI target detection model of the blastospores based on the convolutional neural network by using a backbone network and combining a deep learning target detection framework method based on the constructed training set of the blastospores; and 8) detecting eight fine classification types of blastospores by using the trained fine classification AI target detection model of the blastospores, and summarizing the fine classification types of blastospores as the final detected result of the blastospores.
3. The intelligent detection and identification method for the automatic refinement and labeling of the geminiferous spores, as claimed in claim 2, wherein the fine classification categories are eight, specifically:
fine classification category 1: if two spores of the blastospores are respectively positioned on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and no other target object is positioned at the upper right corner and the lower left corner in the minimum circumscribed upright rectangle, the fine classification category label of the blastospores meeting the conditions is set as 'tl _ br _ blastscore';
fine classification category 2: if two spores of the blastospores are respectively positioned on the upper right area and the lower left area in the minimum circumscribed upright rectangle, and no other target object is positioned at the upper left corner and the lower right corner in the minimum circumscribed upright rectangle, the fine classification category label of the blastospores meeting the conditions is set as 'tr _ bl _ blastospore';
fine classification category 3: if two spores of a blastospore are arranged up and down in the vertical direction within its smallest circumscribed upright rectangle, a fine classification category label of a blastospore satisfying such a case is set to "vertical _ blastospore";
fine classification category 4: if two spores of blastospores are arranged left and right in the horizontal direction in the smallest circumscribed upright rectangle, a fine classification category label of the blastospores satisfying such a condition is set as "horizontal _ blastospore";
fine classification category 5: if two spores of a blastospore are respectively on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and the lower left corner in the minimum circumscribed upright rectangle has other objects and the upper right corner has no other objects, the subdivided category label of the blastospore which satisfies such a condition is assumed to be "tl _ br _ down _ blastoxysore";
fine classification category 6: if two spores of the blastospores are respectively arranged on the upper left area and the lower right area in the minimum circumscribed upright rectangle, and the upper right corner in the minimum circumscribed upright rectangle has other targets but the lower left corner has no other targets, the fine classification category label of the blastospores meeting the conditions is set as 'tl _ br _ up _ blastospore';
fine classification category 7: if two spores of the blastospores are respectively positioned on the upper right area and the lower left area in the minimum circumscribed upright rectangle, and other objects are positioned at the upper left corner and no other objects are positioned at the lower right corner in the minimum circumscribed upright rectangle, the fine classification category label of the blastospores meeting the conditions is set as 'tr _ bl _ up _ blastospore';
fine classification category 8: if two spores of a blastospore are respectively on the upper right region and the lower left region within the minimum circumscribed upright rectangle, and there are other objects in the lower right corner and no other objects in the lower left corner within the minimum circumscribed upright rectangle, the classification category label of the blastospore satisfying such a condition is set to "tr _ bl _ down _ blastoxysore".
4. The intelligent detection and identification method for the automatic refinement and labeling of the geminiferous spores as claimed in claim 3, characterized in that:
setting the width of the minimum external upright rectangle as bounngRectwHigh is boundingRefhThe area of the minimum circumscribed upright rectangle is calculated to be boundingRectArea, the area of the minimum circumscribed rectangle is rotadRectArea, and the area ratio is calculated
Calculating boundingRecwAnd boundingRechMin (boundingRec) is the minimum of the twow,boundingRecth) Calculating boundingRefwAnd boundingRechThe maximum value of the two is denoted as maxLength max (bounngselect)w,boundingRecth) Calculating the aspect ratio
5. The method for intelligently detecting and identifying the automatic thinning and labeling categories of the geminiferous spores as claimed in claim 4, wherein a parameter s is set to 4.
6. The intelligent detection and identification method for the automatic refinement and labeling of the geminiferous spores as claimed in claim 5, wherein the intelligent detection and identification method comprises the following steps:
performing AND operation on the binary image obtained in the step 3.2) and the mask 1, calculating the number of foreground pixel points of the result after the AND operation, and recording the number as Num1;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 2, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num2;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 3, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num3;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 4, calculating the number of foreground pixel points of a result after the AND operation, and recording the number as Num4;
Performing AND operation on the binary image obtained in the step 3.2) and the mask 5, and calculating the number of foreground pixel points of the result after the AND operation, and recording the number as Num5;
And computing the binary image obtained in the step 3.2) with the mask 6, computing the number of foreground pixel points of the result after the and computing, and recording as: num6。
7. The intelligent detection and identification method for the automatic refined labeling category of the geminispores as claimed in claim 6, wherein the discrimination method for the eight fine classification categories of the geminispores is as follows:
the method for discriminating the fine classification category 1 comprises the following steps:
if it is notNum1>Num2,Num5=0,Num6If 0, the blastospore belongs to the fine classification category 1, namely labeled "tl _ br _ blastospore";
the method for discriminating the fine classification category 2 comprises:
if it is notNum1<Num2,Num3=0,Num4If 0, the blastospore belongs to the fine classification category 2, labeled "tr _ bl _ blastospore";
the method for discriminating the fine classification category 3 includes:
if it is notminLength=boundingRectw,maxLength=boundingRecth,Then the blastospore belongs to the fine classification category 3, labeled "vertical _ blastposore";
the method for discriminating the fine classification category 4 comprises the following steps:
if it is notminLength=boundingRecth,maxLength=boundingRectw,The blastospores belong to the fine classification category 4, labeled "horizontal _ blastospore";
the method for discriminating the fine classification category 5 includes:
Num1>Num2,Num5=0,Num6>0, then the blastospore belongs to the fine classification category 5, namely labeled "tl _ br _ down _ blastospore";
the method for discriminating the fine classification category 6 includes:
Num1>Num2,Num5>0,Num6If 0, the blastospore belongs to the fine classification category 6, labeled "tl _ br _ up _ blastospore";
the method for discriminating the fine classification category 7 includes:
Num1<Num2,Num3>0,Num4If 0, the blastospore belongs to the fine classification category 7, labeled "tr _ bl _ up _ blastospore";
the method for discriminating the fine classification category 8 includes:
8. The intelligent detection and identification method for the automatic refinement and labeling of the types of the geminispores as claimed in any one of claims 2 to 7, characterized in that: the backbone network in the step 7) comprises one of VGG-16, VGG-19, ResNet-50, ResNet-101, Inception V3, mobilenetv2, DarkNet19 and DarkNet 53; the deep learning target detection framework method in the step 7) comprises one of SSD, Faster-RCNN and YOLO.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110416482.9A CN113111796B (en) | 2021-04-19 | 2021-04-19 | Detection and identification method for automatically refining and marking categories of geminizing spores |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110416482.9A CN113111796B (en) | 2021-04-19 | 2021-04-19 | Detection and identification method for automatically refining and marking categories of geminizing spores |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113111796A CN113111796A (en) | 2021-07-13 |
CN113111796B true CN113111796B (en) | 2022-06-24 |
Family
ID=76718452
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110416482.9A Active CN113111796B (en) | 2021-04-19 | 2021-04-19 | Detection and identification method for automatically refining and marking categories of geminizing spores |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113111796B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113673396A (en) * | 2021-08-10 | 2021-11-19 | 佛山市海天(高明)调味食品有限公司 | Spore germination rate calculation method and device and storage medium |
CN114445408B (en) * | 2022-04-11 | 2022-06-24 | 山东仕达思生物产业有限公司 | Improved circulation-oriented filtering algorithm-based pathogen detection promoting method, equipment and storage medium |
CN116758024B (en) * | 2023-06-13 | 2024-02-23 | 山东省农业科学院 | Peanut seed direction identification method |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101493886A (en) * | 2009-02-24 | 2009-07-29 | 武汉兰丁医学高科技有限公司 | Karyoplast categorization and identification method in case of unsoundness of characteristic parameter |
US9900409B2 (en) * | 2012-02-07 | 2018-02-20 | Fabulous Inventions Ab | Classification engine for data packet classification |
CN108337172A (en) * | 2018-01-30 | 2018-07-27 | 长沙理工大学 | Extensive OpenFlow flow table classification storage architecture and acceleration lookup method |
CN108830149A (en) * | 2018-05-07 | 2018-11-16 | 深圳市恒扬数据股份有限公司 | A kind of detection method and terminal device of target bacteria |
CN108846312A (en) * | 2018-05-07 | 2018-11-20 | 深圳市恒扬数据股份有限公司 | A kind of recognition methods, device and the terminal device of the effective zone of action of bacterium |
CN110458808A (en) * | 2019-07-10 | 2019-11-15 | 山东仕达思生物产业有限公司 | Female genital tract pathogen recognition methods based on morphology Yu YOLO algorithm |
CN110929678A (en) * | 2019-12-04 | 2020-03-27 | 山东省计算中心(国家超级计算济南中心) | Method for detecting candida vulva vagina spores |
CN112257704A (en) * | 2020-09-15 | 2021-01-22 | 深圳视见医疗科技有限公司 | Cervical fluid-based cell digital image classification method based on deep learning detection model |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10776606B2 (en) * | 2013-09-22 | 2020-09-15 | The Regents Of The University Of California | Methods for delineating cellular regions and classifying regions of histopathology and microanatomy |
-
2021
- 2021-04-19 CN CN202110416482.9A patent/CN113111796B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101493886A (en) * | 2009-02-24 | 2009-07-29 | 武汉兰丁医学高科技有限公司 | Karyoplast categorization and identification method in case of unsoundness of characteristic parameter |
US9900409B2 (en) * | 2012-02-07 | 2018-02-20 | Fabulous Inventions Ab | Classification engine for data packet classification |
CN108337172A (en) * | 2018-01-30 | 2018-07-27 | 长沙理工大学 | Extensive OpenFlow flow table classification storage architecture and acceleration lookup method |
CN108830149A (en) * | 2018-05-07 | 2018-11-16 | 深圳市恒扬数据股份有限公司 | A kind of detection method and terminal device of target bacteria |
CN108846312A (en) * | 2018-05-07 | 2018-11-20 | 深圳市恒扬数据股份有限公司 | A kind of recognition methods, device and the terminal device of the effective zone of action of bacterium |
CN110458808A (en) * | 2019-07-10 | 2019-11-15 | 山东仕达思生物产业有限公司 | Female genital tract pathogen recognition methods based on morphology Yu YOLO algorithm |
CN110929678A (en) * | 2019-12-04 | 2020-03-27 | 山东省计算中心(国家超级计算济南中心) | Method for detecting candida vulva vagina spores |
CN112257704A (en) * | 2020-09-15 | 2021-01-22 | 深圳视见医疗科技有限公司 | Cervical fluid-based cell digital image classification method based on deep learning detection model |
Non-Patent Citations (2)
Title |
---|
国凯平.基于自适应的多掩码抽样算法.《电子技术》.2017, * |
王丹萍.白粉病孢子图像的识别方法研究.《中国电子学会第十八届青年学术年会论文集》.2013, * |
Also Published As
Publication number | Publication date |
---|---|
CN113111796A (en) | 2021-07-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113111796B (en) | Detection and identification method for automatically refining and marking categories of geminizing spores | |
CN108108761B (en) | Rapid traffic signal lamp detection method based on deep feature learning | |
CN110111297A (en) | A kind of injection-molded item surface image defect identification method based on transfer learning | |
CN115082683A (en) | Injection molding defect detection method based on image processing | |
CN105574531A (en) | Intersection point feature extraction based digital identification method | |
Ferrari et al. | Bacterial colony counting by convolutional neural networks | |
CN111402224B (en) | Target identification method for power equipment | |
CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
CN108846831B (en) | Band steel surface defect classification method based on combination of statistical characteristics and image characteristics | |
CN114067114B (en) | Area nuclear segmentation counting method based on algae cell morphology | |
CN110298376A (en) | A kind of bank money image classification method based on improvement B-CNN | |
CN115100191B (en) | Metal casting defect identification method based on industrial detection | |
CN112508857B (en) | Aluminum product surface defect detection method based on improved Cascade R-CNN | |
CN113256624A (en) | Continuous casting round billet defect detection method and device, electronic equipment and readable storage medium | |
CN106874942A (en) | A kind of object module fast construction method semantic based on regular expression | |
CN113033417B (en) | Detection and identification method and equipment for spores in gynecological microecology and storage medium | |
CN113252614A (en) | Transparency detection method based on machine vision | |
Li et al. | The research on traffic sign recognition based on deep learning | |
CN112381140A (en) | Abrasive particle image machine learning identification method based on new characteristic parameters | |
CN112686872A (en) | Wood counting method based on deep learning | |
CN112580659A (en) | Ore identification method based on machine vision | |
US11521427B1 (en) | Ear detection method with deep learning pairwise model based on contextual information | |
CN109615630A (en) | Semi-continuous casting alusil alloy Analysis on Microstructure method based on image processing techniques | |
CN112950636B (en) | R3 Det-based rotating frame target detection method for female vaginal pathogens | |
CN111626150B (en) | Commodity identification method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A detection and recognition method for automatically refining and labeling categories of spores Granted publication date: 20220624 Pledgee: Bank of China Limited Jinan Huaiyin sub branch Pledgor: SHANDONG STARS BIOINDUSTRY CO.,LTD. Registration number: Y2024980006615 |