Detailed description of the invention
In conjunction with following application scenarios, the invention will be further described.
Application scenarios 1
Seeing Fig. 1, Fig. 2, a kind of in-vivo information acquiring apparatus of an embodiment of this application scene, including cellIdentification module and data obtaining module, described cell recognition module is used for determining that biological species, described data obtaining module include:
Information acquiring section, it obtains organism internal information；
Electric power source, it is for providing electric power to above-mentioned information acquiring section；
Magnetic Sensor portion, its detection, from the magnetic signal of outside input, exports control corresponding with the detection state of this magnetic signalSignal processed；
Umber of pulse count section, the umber of pulse of the pulse signal from above-mentioned Magnetic Sensor portion is counted by it；
Umber of pulse judging part, its judge the umber of pulse counted to get by above-mentioned umber of pulse count section be whether predetermined number withOn；
Power cut control portion, it is being judged as have input the feelings of the pulse of more than predetermined number by above-mentioned umber of pulse judging partUnder condition, above-mentioned electric power source is provided to the electric power that above-mentioned information acquiring section is carried out and switches to dissengaged positions from offer state.
Preferably, described data obtaining module also includes being spaced test section, and this interval test section detection passes from above-mentioned magneticThe output gap of the pulse signal in sensor portion,
Described umber of pulse count section at the output gap detected by above-mentioned interval test section not less than base set in advanceIn the case of quasi-interval, the output number of above-mentioned pulse signal is updated.
Data can be updated by this preferred embodiment in time.
Preferably, described interval test section is made up of enumerator.
It is more accurate that this preferred embodiment obtains information.
Preferably, described cell recognition module 1 includes that Methods of Segmentation On Cell Images unit 11, feature extraction unit 12, classification are knownOther unit 13；Described Methods of Segmentation On Cell Images unit 11 is for distinguishing the back of the body in the cell image gathered by cell image acquisition moduleScape, nucleus and Cytoplasm；Described feature extraction unit 12 is for extracting the textural characteristics of cell image；Described classificationRecognition unit 13 is for utilizing grader to realize cell image Classification and Identification according to textural characteristics.
This preferred embodiment constructs the unit structure of cell recognition module 1.
Preferably, described Methods of Segmentation On Cell Images unit 11 includes that image changes subelement, noise remove subelement, coarse segmentationSubelement, nuclear centers demarcate subelement, Accurate Segmentation subelement, particularly as follows:
(1) image conversion subelement, for being converted into gray level image by the cell image of collection；
(2) noise remove subelement, for gray level image is carried out denoising, including:
For pixel, (x y), chooses its neighborhood S of 3 × 3x,y(2N+1) the neighborhood L of × (2N+1)x,y, N is for being more thanInteger equal to 2；
First whether be that boundary point judges to pixel, set threshold value T, T ∈ [13,26], calculate pixel (x, y)With its neighborhood Sx,yIn the gray scale difference value of each pixel, and compare with threshold value T, if gray scale difference value is more than the number of threshold value TMore than or equal to 6, then (x, y) is boundary point to pixel, and otherwise, (x y) is non-boundary point to pixel；
If (x, y) is boundary point, then carry out following noise reduction process:
In formula, h (x, y) be after noise reduction pixel ((x y) is noise reduction preceding pixel point (x, ash y) to q for x, gray value y)Angle value, σ is pixel (x, y) neighborhood Lx,yInterior gray value mark is poor, q (i, j) ∈ [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] representNeighborhood Lx,yInterior gray value fall within interval [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] point, k represents neighborhood Lx,yInterior gray value falls withinInterval [q (x, y)-1.5 σ, q (x, y)+1.5 σ] the quantity of point；
If (x, y) is non-boundary point, then carry out following noise reduction process:
In formula, (x y) is pixel (x, gray value y), q (i, j) representative image midpoint (i, j) ash at place after noise reduction to hAngle value, (i j) is neighborhood L to wx,yInterior point (i, j) corresponding Gauss weight；
(3) coarse segmentation subelement, for slightly drawing the background in the cell image after denoising, Cytoplasm, nucleusPoint, particularly as follows:
By each pixel (x, y) represents with four dimensional feature vectors:
In formula, (x y) represents (x, gray value y), h to have(x y) represents its neighborhood Sx,yGray average, hmed(x, y) generationTable its neighborhood Sx,yGray scale intermediate value, hsta(x y) represents its neighborhood Sx,yGray variance；
K-means clustering procedure is used to be divided into background, Cytoplasm, nucleus three class；
(4) nuclear centers demarcates subelement, for demarcating nuclear centers:
Nucleus approximate region is obtained, if nuclear area comprises n point: (x by coarse segmentation subelement1,y1),…,(xn,yn), this region is carried out intensity-weighted demarcation and geometric center is demarcated, take its meansigma methods as nuclear centers (xz,yz):
(5) Accurate Segmentation subelement, for carrying out Accurate Segmentation to nucleus, Cytoplasm；
Build from nuclear centers (xz, yz) arrive nucleus and Cytoplasm boundary point (xp, yp) directed line segmentDistanceRepresent and round downwards；
Carry out sampling along line segment with unit length and can obtain dispIndividual point (x1, y1) ...,If sampled pointCoordinate be not integer, its gray value is obtained by surrounding pixel linear interpolation；
Point (xi, yi) place is along the gray scale difference of line segment direction:
hd(xi, yi)=h (xi-1, yi-1)-h(xi, yi)
Definition gray scale difference inhibition function:
Point (xi, yi) place is along the gradient gra (x of line segment directioni, yi):
Choose the maximum value point of gradient as nucleus and cytoplasmic precise edge.
This preferred embodiment arranges noise remove subelement, and effective integration center pixel closes on the space of neighborhood territory pixelProperty and grey similarity carry out noise reduction process, flat site in the picture, in neighborhood, grey scale pixel value is more or less the same, useGaussian filter is weighted filtering to gray value, and at the borderline region that change is violent, row bound keeps filtering, beneficially imageThe holding at edge；Use K mean cluster to extract nucleus and Cytoplasm coarse contour, can effectively remove the interference of noise；Arrange thinSubelement is demarcated at karyon center, it is simple to follow-up be accurately positioned nucleus and Cytoplasm profile；Accurate Segmentation subelement fillsDivide and make use of directional information, overcome the inflammatory cell interference to edge graph, it is possible to accurately extract nucleus and Cytoplasm limitEdge.
Preferably, the described textural characteristics to cell image extracts, including:
(1) the Gray co-occurrence matrix of cell image, described comprehensive ash is asked for based on the gray level co-occurrence matrixes method improvedDegree co-occurrence matrix embodies cell textural characteristics in different directions:
Be located at 0 °, 45 °, 90 °, gray level co-occurrence matrixes on 135 ° of four directions be respectively h (x, y, d, 0 °), h (x, y, d,45 °), h (x, y, d, 90 °), h (x, y, d, 135 °), corresponding matrix element project is X1、X2、X3、X4, then Gray is altogetherThe computing formula of raw matrix is:
H (x, y, d)=w1h(x,y,d,0°)+w2h(x,y,d,45°)+w3h(x,y,d,90°)+w4h(x,y,d,135°)
Gray co-occurrence matrix element number is:
In formula, d represents distance, and the span of d is [2,4], wiFor weight coefficient, i=1,2,3,4, it is by four sidesThe contrast level parameter that the gray level co-occurrence matrixes on each direction in is corresponding calculates, if the gray level co-occurrence matrixes on four directionCorresponding contrast level parameter is respectively Di, average isI=1,2,3,4, then weight coefficient wiComputing formula be:
(2) four textural characteristics parameters needed for utilizing described Gray co-occurrence matrix and matrix element project to obtain:Contrast, variance and, energy and average；
(3) described four textural characteristics parameters are normalized, the normalized textural characteristics value of final acquisition.
This preferred embodiment, based on the gray level co-occurrence matrixes method improved, uses the mode arranging weight coefficient to ask for cytological mapThe Gray co-occurrence matrix of picture, and then extract cell textural characteristics on appointment four direction, solve owing to outside is doneDisturb the textural characteristics ginseng of the cell that (cause such as lighting angle when cell image gathers impact, the flowing interference etc. of gas) causesNumerical value has the problem of bigger difference in different directions, improves the precision of cell image texture feature extraction；Selected contrast,Variance and, energy and four textural characteristics of average, eliminate the characteristic parameter of redundancy and repetition；To described four textural characteristics ginsengNumber is normalized, and the Classification and Identification facilitating follow-up cell image processes.
In this application scenarios, setting threshold value T=13, d=2, image denoising effect improves 5% relatively, cell imageThe extraction accuracy of feature improves 8%.
Application scenarios 2
Seeing Fig. 1, Fig. 2, a kind of in-vivo information acquiring apparatus of an embodiment of this application scene, including cellIdentification module and data obtaining module, described cell recognition module is used for determining that biological species, described data obtaining module include:
Information acquiring section, it obtains organism internal information；
Electric power source, it is for providing electric power to above-mentioned information acquiring section；
Magnetic Sensor portion, its detection, from the magnetic signal of outside input, exports control corresponding with the detection state of this magnetic signalSignal processed；
Umber of pulse count section, the umber of pulse of the pulse signal from above-mentioned Magnetic Sensor portion is counted by it；
Umber of pulse judging part, its judge the umber of pulse counted to get by above-mentioned umber of pulse count section be whether predetermined number withOn；
Power cut control portion, it is being judged as have input the feelings of the pulse of more than predetermined number by above-mentioned umber of pulse judging partUnder condition, above-mentioned electric power source is provided to the electric power that above-mentioned information acquiring section is carried out and switches to dissengaged positions from offer state.
Preferably, described data obtaining module also includes being spaced test section, and this interval test section detection passes from above-mentioned magneticThe output gap of the pulse signal in sensor portion,
Described umber of pulse count section at the output gap detected by above-mentioned interval test section not less than base set in advanceIn the case of quasi-interval, the output number of above-mentioned pulse signal is updated.
Data can be updated by this preferred embodiment in time.
Preferably, described interval test section is made up of enumerator.
It is more accurate that this preferred embodiment obtains information.
Preferably, described cell recognition module 1 includes that Methods of Segmentation On Cell Images unit 11, feature extraction unit 12, classification are knownOther unit 13；Described Methods of Segmentation On Cell Images unit 11 is for distinguishing the back of the body in the cell image gathered by cell image acquisition moduleScape, nucleus and Cytoplasm；Described feature extraction unit 12 is for extracting the textural characteristics of cell image；Described classificationRecognition unit 13 is for utilizing grader to realize cell image Classification and Identification according to textural characteristics.
This preferred embodiment constructs the unit structure of cell recognition module 1.
Preferably, described Methods of Segmentation On Cell Images unit 11 includes that image changes subelement, noise remove subelement, coarse segmentationSubelement, nuclear centers demarcate subelement, Accurate Segmentation subelement, particularly as follows:
(1) image conversion subelement, for being converted into gray level image by the cell image of collection；
(2) noise remove subelement, for gray level image is carried out denoising, including:
For pixel, (x y), chooses its neighborhood S of 3 × 3x,y(2N+1) the neighborhood L of × (2N+1)x,y, N is for being more thanInteger equal to 2；
First whether be that boundary point judges to pixel, set threshold value T, T ∈ [13,26], calculate pixel (x, y)With its neighborhood Sx,yIn the gray scale difference value of each pixel, and compare with threshold value T, if gray scale difference value is more than the number of threshold value TMore than or equal to 6, then (x, y) is boundary point to pixel, and otherwise, (x y) is non-boundary point to pixel；
If (x, y) is boundary point, then carry out following noise reduction process:
In formula, h (x, y) be after noise reduction pixel ((x y) is noise reduction preceding pixel point (x, ash y) to q for x, gray value y)Angle value, σ is pixel (x, y) neighborhood Lx,yInterior gray value mark is poor, q (i, j) ∈ [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] representNeighborhood Lx,yInterior gray value fall within interval [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] point, k represents neighborhood lx,yInterior gray value falls withinInterval [q (x, y)-1.5 σ, q (x, y)+1.5 σ] the quantity of point；
If (x, y) is non-boundary point, then carry out following noise reduction process:
In formula, (x y) is pixel (x, gray value y), q (i, j) representative image midpoint (i, j) ash at place after noise reduction to hAngle value, (i j) is neighborhood L to wx,yInterior point (i, j) corresponding Gauss weight；
(3) coarse segmentation subelement, for slightly drawing the background in the cell image after denoising, Cytoplasm, nucleusPoint, particularly as follows:
By each pixel (x, y) represents with four dimensional feature vectors:
In formula, (x y) represents (x, gray value y), h to have(x y) represents its neighborhood Sx,yGray average, hmed(x, y) generationTable its neighborhood Sx,yGray scale intermediate value, hsta(x y) represents its neighborhood Sx,yGray variance；
K-means clustering procedure is used to be divided into background, Cytoplasm, nucleus three class；
(4) nuclear centers demarcates subelement, for demarcating nuclear centers:
Nucleus approximate region is obtained, if nuclear area comprises n point: (x by coarse segmentation subelement1,y1),…,(xn,yn), this region is carried out intensity-weighted demarcation and geometric center is demarcated, take its meansigma methods as nuclear centers (xz,yz):
(5) Accurate Segmentation subelement, for carrying out Accurate Segmentation to nucleus, Cytoplasm；
Build from nuclear centers (xz, yz) arrive nucleus and Cytoplasm boundary point (xp, yp) directed line segmentDistanceRepresent and round downwards；
Carry out sampling along line segment with unit length and can obtain dispIndividual point (x1, y1) ...,If samplingThe coordinate of point is not integer, and its gray value is obtained by surrounding pixel linear interpolation；
Point (xi, yi) place is along the gray scale difference of line segment direction:
hd(xi, yi)=h (xi-1, yi-1)-h(xi, yi)
Definition gray scale difference inhibition function:
Point (xi,yi) place is along the gradient gra (x of line segment directioni,yi):
Choose the maximum value point of gradient as nucleus and cytoplasmic precise edge.
This preferred embodiment arranges noise remove subelement, and effective integration center pixel closes on the space of neighborhood territory pixelProperty and grey similarity carry out noise reduction process, flat site in the picture, in neighborhood, grey scale pixel value is more or less the same, useGaussian filter is weighted filtering to gray value, and at the borderline region that change is violent, row bound keeps filtering, beneficially imageThe holding at edge；Use K mean cluster to extract nucleus and Cytoplasm coarse contour, can effectively remove the interference of noise；Arrange thinSubelement is demarcated at karyon center, it is simple to follow-up be accurately positioned nucleus and Cytoplasm profile；Accurate Segmentation subelement fillsDivide and make use of directional information, overcome the inflammatory cell interference to edge graph, it is possible to accurately extract nucleus and Cytoplasm limitEdge.
Preferably, the described textural characteristics to cell image extracts, including:
(1) the Gray co-occurrence matrix of cell image, described comprehensive ash is asked for based on the gray level co-occurrence matrixes method improvedDegree co-occurrence matrix embodies cell textural characteristics in different directions:
Be located at 0 °, 45 °, 90 °, gray level co-occurrence matrixes on 135 ° of four directions be respectively h (x, y, d, 0 °), h (x, y, d,45 °), h (x, y, d, 90 °), h (x, y, d, 135 °), corresponding matrix element project is X1、X2、X3、X4, then Gray is altogetherThe computing formula of raw matrix is:
H (x, y, d)=w1h(x,y,d,0°)+w2h(x,y,d,45°)+w3h(x,y,d,90°)+w4h(x,y,d,135°)
Gray co-occurrence matrix element number is:
In formula, d represents distance, and the span of d is [2,4], wiFor weight coefficient, i=1,2,3,4, it is by four sidesThe contrast level parameter that the gray level co-occurrence matrixes on each direction in is corresponding calculates, if the gray level co-occurrence matrixes on four directionCorresponding contrast level parameter is respectively Di, average isI=1,2,3,4, then weight coefficient wiComputing formula be:
(2) four textural characteristics parameters needed for utilizing described Gray co-occurrence matrix and matrix element project to obtain:Contrast, variance and, energy and average；
(3) described four textural characteristics parameters are normalized, the normalized textural characteristics value of final acquisition.
This preferred embodiment, based on the gray level co-occurrence matrixes method improved, uses the mode arranging weight coefficient to ask for cytological mapThe Gray co-occurrence matrix of picture, and then extract cell textural characteristics on appointment four direction, solve owing to outside is doneDisturb the textural characteristics ginseng of the cell that (cause such as lighting angle when cell image gathers impact, the flowing interference etc. of gas) causesNumerical value has the problem of bigger difference in different directions, improves the precision of cell image texture feature extraction；Selected contrast,Variance and, energy and four textural characteristics of average, eliminate the characteristic parameter of redundancy and repetition；To described four textural characteristics ginsengNumber is normalized, and the Classification and Identification facilitating follow-up cell image processes.
In this application scenarios, setting threshold value T=15, d=2, image denoising effect improves 6% relatively, cell imageThe extraction accuracy of feature improves 8%.
Application scenarios 3
Seeing Fig. 1, Fig. 2, a kind of in-vivo information acquiring apparatus of an embodiment of this application scene, including cellIdentification module and data obtaining module, described cell recognition module is used for determining that biological species, described data obtaining module include:
Information acquiring section, it obtains organism internal information；
Electric power source, it is for providing electric power to above-mentioned information acquiring section；
Magnetic Sensor portion, its detection, from the magnetic signal of outside input, exports control corresponding with the detection state of this magnetic signalSignal processed；
Umber of pulse count section, the umber of pulse of the pulse signal from above-mentioned Magnetic Sensor portion is counted by it；
Umber of pulse judging part, its judge the umber of pulse counted to get by above-mentioned umber of pulse count section be whether predetermined number withOn；
Power cut control portion, it is being judged as have input the feelings of the pulse of more than predetermined number by above-mentioned umber of pulse judging partUnder condition, above-mentioned electric power source is provided to the electric power that above-mentioned information acquiring section is carried out and switches to dissengaged positions from offer state.
Preferably, described data obtaining module also includes being spaced test section, and this interval test section detection passes from above-mentioned magneticThe output gap of the pulse signal in sensor portion,
Described umber of pulse count section at the output gap detected by above-mentioned interval test section not less than base set in advanceIn the case of quasi-interval, the output number of above-mentioned pulse signal is updated.
Data can be updated by this preferred embodiment in time.
Preferably, described interval test section is made up of enumerator.
It is more accurate that this preferred embodiment obtains information.
Preferably, described cell recognition module 1 includes that Methods of Segmentation On Cell Images unit 11, feature extraction unit 12, classification are knownOther unit 13；Described Methods of Segmentation On Cell Images unit 11 is for distinguishing the back of the body in the cell image gathered by cell image acquisition moduleScape, nucleus and Cytoplasm；Described feature extraction unit 12 is for extracting the textural characteristics of cell image；Described classificationRecognition unit 13 is for utilizing grader to realize cell image Classification and Identification according to textural characteristics.
This preferred embodiment constructs the unit structure of cell recognition module 1.
Preferably, described Methods of Segmentation On Cell Images unit 11 includes that image changes subelement, noise remove subelement, coarse segmentationSubelement, nuclear centers demarcate subelement, Accurate Segmentation subelement, particularly as follows:
(1) image conversion subelement, for being converted into gray level image by the cell image of collection；
(2) noise remove subelement, for gray level image is carried out denoising, including:
For pixel, (x y), chooses its neighborhood S of 3 × 3x,y(2N+1) the neighborhood L of × (2N+1)x,y, N is for being more thanInteger equal to 2；
First whether be that boundary point judges to pixel, set threshold value T, T ∈ [13,26], calculate pixel (x, y)With its neighborhood Sx,yIn the gray scale difference value of each pixel, and compare with threshold value T, if gray scale difference value is more than the number of threshold value TMore than or equal to 6, then (x, y) is boundary point to pixel, and otherwise, (x y) is non-boundary point to pixel；
If (x, y) is boundary point, then carry out following noise reduction process:
In formula, h (x, y) be after noise reduction pixel ((x y) is noise reduction preceding pixel point (x, ash y) to q for x, gray value y)Angle value, σ is pixel (x, y) neighborhood Lx,yInterior gray value mark is poor, q (i, j) ∈ [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] representNeighborhood Lx,yInterior gray value fall within interval [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] point, k represents neighborhood Lx,yInterior gray value falls withinInterval
[q (and x, y)-1.5 σ, q (x, y)+1.5 σ] the quantity of point；
If (x, y) is non-boundary point, then carry out following noise reduction process:
In formula, (x y) is pixel (x, gray value y), q (i, j) representative image midpoint (i, j) ash at place after noise reduction to hAngle value, (i j) is neighborhood L to wx,yInterior point (i, j) corresponding Gauss weight；
(3) coarse segmentation subelement, for slightly drawing the background in the cell image after denoising, Cytoplasm, nucleusPoint, particularly as follows:
By each pixel (x, y) represents with four dimensional feature vectors:
In formula, (x y) represents (x, gray value y), h to have(x y) represents its neighborhood Sx,yGray average, hmed(x, y) generationTable its neighborhood Sx,yGray scale intermediate value, hsta(x y) represents its neighborhood Sx,yGray variance；
K-means clustering procedure is used to be divided into background, Cytoplasm, nucleus three class；
(4) nuclear centers demarcates subelement, for demarcating nuclear centers:
Nucleus approximate region is obtained, if nuclear area comprises n point: (x by coarse segmentation subelement1,y1),…,(xn,yn), this region is carried out intensity-weighted demarcation and geometric center is demarcated, take its meansigma methods as nuclear centers (xz,yz):
(5) Accurate Segmentation subelement, for carrying out Accurate Segmentation to nucleus, Cytoplasm；
Build from nuclear centers (xz,yz) arrive nucleus and Cytoplasm boundary point (xp,yp) directed line segmentDistanceRepresent and round downwards；
Carry out sampling along line segment with unit length and can obtain dispIndividual point (x1,y1) ...,If samplingThe coordinate of point is not integer, and its gray value is obtained by surrounding pixel linear interpolation；
Point (xi,yi) place is along the gray scale difference of line segment direction:
hd(xi,yi)=h (xi-1,yi-1)-h(xi,yi)
Definition gray scale difference inhibition function:
Point (xi,yi) place is along the gradient gra (x of line segment directioni,yi):
Choose the maximum value point of gradient as nucleus and cytoplasmic precise edge.
This preferred embodiment arranges noise remove subelement, and effective integration center pixel closes on the space of neighborhood territory pixelProperty and grey similarity carry out noise reduction process, flat site in the picture, in neighborhood, grey scale pixel value is more or less the same, useGaussian filter is weighted filtering to gray value, and at the borderline region that change is violent, row bound keeps filtering, beneficially imageThe holding at edge；Use K mean cluster to extract nucleus and Cytoplasm coarse contour, can effectively remove the interference of noise；Arrange thinSubelement is demarcated at karyon center, it is simple to follow-up be accurately positioned nucleus and Cytoplasm profile；Accurate Segmentation subelement fillsDivide and make use of directional information, overcome the inflammatory cell interference to edge graph, it is possible to accurately extract nucleus and Cytoplasm limitEdge.
Preferably, the described textural characteristics to cell image extracts, including:
(1) the Gray co-occurrence matrix of cell image, described comprehensive ash is asked for based on the gray level co-occurrence matrixes method improvedDegree co-occurrence matrix embodies cell textural characteristics in different directions:
Be located at 0 °, 45 °, 90 °, gray level co-occurrence matrixes on 135 ° of four directions be respectively h (x, y, d, 0 °), h (x, y, d,45 °), h (x, y, d, 90 °), h (x, y, d, 135 °), corresponding matrix element project is X1、X2、X3、X4, then Gray is altogetherThe computing formula of raw matrix is:
H (x, y, d)=w1h(x,y,d,0°)+w2h(x,y,d,45°)+w3h(x,y,d,90°)+w4h(x,y,d,135°)
Gray co-occurrence matrix element number is:
In formula, d represents distance, and the span of d is [2,4], wiFor weight coefficient, i=1,2,3,4, it is by four sidesThe contrast level parameter that the gray level co-occurrence matrixes on each direction in is corresponding calculates, if the gray level co-occurrence matrixes on four directionCorresponding contrast level parameter is respectively Di, average isI=1,2,3,4, then weight coefficient wiComputing formula be:
(2) four textural characteristics parameters needed for utilizing described Gray co-occurrence matrix and matrix element project to obtain:Contrast, variance and, energy and average；
(3) described four textural characteristics parameters are normalized, the normalized textural characteristics value of final acquisition.
This preferred embodiment, based on the gray level co-occurrence matrixes method improved, uses the mode arranging weight coefficient to ask for cytological mapThe Gray co-occurrence matrix of picture, and then extract cell textural characteristics on appointment four direction, solve owing to outside is doneDisturb the textural characteristics ginseng of the cell that (cause such as lighting angle when cell image gathers impact, the flowing interference etc. of gas) causesNumerical value has the problem of bigger difference in different directions, improves the precision of cell image texture feature extraction；Selected contrast,Variance and, energy and four textural characteristics of average, eliminate the characteristic parameter of redundancy and repetition；To described four textural characteristics ginsengNumber is normalized, and the Classification and Identification facilitating follow-up cell image processes.
In this application scenarios, setting threshold value T=18, d=3, image denoising effect improves 7% relatively, cell imageThe extraction accuracy of feature improves 7%.
Application scenarios 4
Seeing Fig. 1, Fig. 2, a kind of in-vivo information acquiring apparatus of an embodiment of this application scene, including cellIdentification module and data obtaining module, described cell recognition module is used for determining that biological species, described data obtaining module include:
Information acquiring section, it obtains organism internal information；
Electric power source, it is for providing electric power to above-mentioned information acquiring section；
Magnetic Sensor portion, its detection, from the magnetic signal of outside input, exports control corresponding with the detection state of this magnetic signalSignal processed；
Umber of pulse count section, the umber of pulse of the pulse signal from above-mentioned Magnetic Sensor portion is counted by it；
Umber of pulse judging part, its judge the umber of pulse counted to get by above-mentioned umber of pulse count section be whether predetermined number withOn；
Power cut control portion, it is being judged as have input the feelings of the pulse of more than predetermined number by above-mentioned umber of pulse judging partUnder condition, above-mentioned electric power source is provided to the electric power that above-mentioned information acquiring section is carried out and switches to dissengaged positions from offer state.
Preferably, described data obtaining module also includes being spaced test section, and this interval test section detection passes from above-mentioned magneticThe output gap of the pulse signal in sensor portion,
Described umber of pulse count section at the output gap detected by above-mentioned interval test section not less than base set in advanceIn the case of quasi-interval, the output number of above-mentioned pulse signal is updated.
Data can be updated by this preferred embodiment in time.
Preferably, described interval test section is made up of enumerator.
It is more accurate that this preferred embodiment obtains information.
Preferably, described cell recognition module 1 includes that Methods of Segmentation On Cell Images unit 11, feature extraction unit 12, classification are knownOther unit 13；Described Methods of Segmentation On Cell Images unit 11 is for distinguishing the back of the body in the cell image gathered by cell image acquisition moduleScape, nucleus and Cytoplasm；Described feature extraction unit 12 is for extracting the textural characteristics of cell image；Described classificationRecognition unit 13 is for utilizing grader to realize cell image Classification and Identification according to textural characteristics.
This preferred embodiment constructs the unit structure of cell recognition module 1.
Preferably, described Methods of Segmentation On Cell Images unit 11 includes that image changes subelement, noise remove subelement, coarse segmentationSubelement, nuclear centers demarcate subelement, Accurate Segmentation subelement, particularly as follows:
(1) image conversion subelement, for being converted into gray level image by the cell image of collection；
(2) noise remove subelement, for gray level image is carried out denoising, including:
For pixel, (x y), chooses its neighborhood S of 3 × 3x,y(2N+1) the neighborhood L of × (2N+1)x,y, N is for being more thanInteger equal to 2；
First whether be that boundary point judges to pixel, set threshold value T, T ∈ [13,26], calculate pixel (x, y)With its neighborhood Sx,yIn the gray scale difference value of each pixel, and compare with threshold value T, if gray scale difference value is more than the number of threshold value TMore than or equal to 6, then (x, y) is boundary point to pixel, and otherwise, (x y) is non-boundary point to pixel；
If (x, y) is boundary point, then carry out following noise reduction process:
In formula, h (x, y) be after noise reduction pixel ((x y) is noise reduction preceding pixel point (x, ash y) to q for x, gray value y)Angle value, σ is pixel (x, y) neighborhood Lx,yInterior gray value mark is poor, q (i, j) ∈ [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] representNeighborhood Lx,yInterior gray value fall within interval [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] point, k represents neighborhood Lx,yInterior gray value falls withinInterval
[q (and x, y)-1.5 σ, q (x, y)+1.5 σ] the quantity of point；
If (x, y) is non-boundary point, then carry out following noise reduction process:
In formula, (x y) is pixel (x, gray value y), q (i, j) representative image midpoint (i, j) ash at place after noise reduction to hAngle value, (i j) is neighborhood L to wx,yInterior point (i, j) corresponding Gauss weight；
(3) coarse segmentation subelement, for slightly drawing the background in the cell image after denoising, Cytoplasm, nucleusPoint, particularly as follows:
By each pixel (x, y) represents with four dimensional feature vectors:
In formula, (x y) represents (x, gray value y), h to have(x y) represents its neighborhood Sx,yGray average, hmed(x, y) generationTable its neighborhood Sx,yGray scale intermediate value, hsta(x y) represents its neighborhood Sx,yGray variance；
K-means clustering procedure is used to be divided into background, Cytoplasm, nucleus three class；
(4) nuclear centers demarcates subelement, for demarcating nuclear centers:
Nucleus approximate region is obtained, if nuclear area comprises n point: (x by coarse segmentation subelement1,y1),…,(xn,yn), this region is carried out intensity-weighted demarcation and geometric center is demarcated, take its meansigma methods as nuclear centers (xz,yz):
(5) Accurate Segmentation subelement, for carrying out Accurate Segmentation to nucleus, Cytoplasm；
Build from nuclear centers (xz,yz) arrive nucleus and Cytoplasm boundary point (xp,yp) directed line segmentDistanceRepresent and round downwards；
Carry out sampling along line segment with unit length and can obtain dispIndividual point (x1,y1) ...,If samplingThe coordinate of point is not integer, and its gray value is obtained by surrounding pixel linear interpolation；
Point (xi,yi) place is along the gray scale difference of line segment direction:
hd(xi,yi)=h (xi-1,yi-1)-h(xi,yi)
Definition gray scale difference inhibition function:
Point (xi,yi) place is along the gradient gra (x of line segment directioni,yi):
Choose the maximum value point of gradient as nucleus and cytoplasmic precise edge.
This preferred embodiment arranges noise remove subelement, and effective integration center pixel closes on the space of neighborhood territory pixelProperty and grey similarity carry out noise reduction process, flat site in the picture, in neighborhood, grey scale pixel value is more or less the same, useGaussian filter is weighted filtering to gray value, and at the borderline region that change is violent, row bound keeps filtering, beneficially imageThe holding at edge；Use K mean cluster to extract nucleus and Cytoplasm coarse contour, can effectively remove the interference of noise；Arrange thinSubelement is demarcated at karyon center, it is simple to follow-up be accurately positioned nucleus and Cytoplasm profile；Accurate Segmentation subelement fillsDivide and make use of directional information, overcome the inflammatory cell interference to edge graph, it is possible to accurately extract nucleus and Cytoplasm limitEdge.
Preferably, the described textural characteristics to cell image extracts, including:
(1) the Gray co-occurrence matrix of cell image, described comprehensive ash is asked for based on the gray level co-occurrence matrixes method improvedDegree co-occurrence matrix embodies cell textural characteristics in different directions:
Be located at 0 °, 45 °, 90 °, gray level co-occurrence matrixes on 135 ° of four directions be respectively h (x, y, d, 0 °), h (x, y, d,45 °), h (x, y, d, 90 °), h (x, y, d, 135 °), corresponding matrix element project is X1、X2、X3、X4, then Gray is altogetherThe computing formula of raw matrix is:
H (x, y, d)=w1h(x,y,d,0°)+w2h(x,y,d,45°)+w3h(x,y,d,90°)+w4h(x,y,d,135°)
Gray co-occurrence matrix element number is:
In formula, d represents distance, and the span of d is [2,4], wiFor weight coefficient, i=1,2,3,4, it is by four sidesThe contrast level parameter that the gray level co-occurrence matrixes on each direction in is corresponding calculates, if the gray level co-occurrence matrixes on four directionCorresponding contrast level parameter is respectively Di, average isI=1,2,3,4, then weight coefficient wiComputing formula be:
(2) four textural characteristics parameters needed for utilizing described Gray co-occurrence matrix and matrix element project to obtain:Contrast, variance and, energy and average；
(3) described four textural characteristics parameters are normalized, the normalized textural characteristics value of final acquisition.
This preferred embodiment, based on the gray level co-occurrence matrixes method improved, uses the mode arranging weight coefficient to ask for cytological mapThe Gray co-occurrence matrix of picture, and then extract cell textural characteristics on appointment four direction, solve owing to outside is doneDisturb the textural characteristics ginseng of the cell that (cause such as lighting angle when cell image gathers impact, the flowing interference etc. of gas) causesNumerical value has the problem of bigger difference in different directions, improves the precision of cell image texture feature extraction；Selected contrast,Variance and, energy and four textural characteristics of average, eliminate the characteristic parameter of redundancy and repetition；To described four textural characteristics ginsengNumber is normalized, and the Classification and Identification facilitating follow-up cell image processes.
In this application scenarios, setting threshold value T=20, d=4, image denoising effect improves 8% relatively, cell imageThe extraction accuracy of feature improves 6%.
Application scenarios 5
Seeing Fig. 1, Fig. 2, a kind of in-vivo information acquiring apparatus of an embodiment of this application scene, including cellIdentification module and data obtaining module, described cell recognition module is used for determining that biological species, described data obtaining module include:
Information acquiring section, it obtains organism internal information；
Electric power source, it is for providing electric power to above-mentioned information acquiring section；
Magnetic Sensor portion, its detection, from the magnetic signal of outside input, exports control corresponding with the detection state of this magnetic signalSignal processed；
Umber of pulse count section, the umber of pulse of the pulse signal from above-mentioned Magnetic Sensor portion is counted by it；
Umber of pulse judging part, its judge the umber of pulse counted to get by above-mentioned umber of pulse count section be whether predetermined number withOn；
Power cut control portion, it is being judged as have input the feelings of the pulse of more than predetermined number by above-mentioned umber of pulse judging partUnder condition, above-mentioned electric power source is provided to the electric power that above-mentioned information acquiring section is carried out and switches to dissengaged positions from offer state.
Preferably, described data obtaining module also includes being spaced test section, and this interval test section detection passes from above-mentioned magneticThe output gap of the pulse signal in sensor portion,
Described umber of pulse count section at the output gap detected by above-mentioned interval test section not less than base set in advanceIn the case of quasi-interval, the output number of above-mentioned pulse signal is updated.
Data can be updated by this preferred embodiment in time.
Preferably, described interval test section is made up of enumerator.
It is more accurate that this preferred embodiment obtains information.
Preferably, described cell recognition module 1 includes that Methods of Segmentation On Cell Images unit 11, feature extraction unit 12, classification are knownOther unit 13；Described Methods of Segmentation On Cell Images unit 11 is for distinguishing the back of the body in the cell image gathered by cell image acquisition moduleScape, nucleus and Cytoplasm；Described feature extraction unit 12 is for extracting the textural characteristics of cell image；Described classificationRecognition unit 13 is for utilizing grader to realize cell image Classification and Identification according to textural characteristics.
This preferred embodiment constructs the unit structure of cell recognition module 1.
Preferably, described Methods of Segmentation On Cell Images unit 11 includes that image changes subelement, noise remove subelement, coarse segmentationSubelement, nuclear centers demarcate subelement, Accurate Segmentation subelement, particularly as follows:
(1) image conversion subelement, for being converted into gray level image by the cell image of collection；
(2) noise remove subelement, for gray level image is carried out denoising, including:
For pixel, (x y), chooses its neighborhood S of 3 × 3x,y(2N+1) the neighborhood L of × (2N+1)x,y, N is for being more thanInteger equal to 2；
First whether be that boundary point judges to pixel, set threshold value T, T ∈ [13,26], calculate pixel (x, y)With its neighborhood Sx,yIn the gray scale difference value of each pixel, and compare with threshold value T, if gray scale difference value is more than the number of threshold value TMore than or equal to 6, then (x, y) is boundary point to pixel, and otherwise, (x y) is non-boundary point to pixel；
If (x, y) is boundary point, then carry out following noise reduction process:
In formula, h (x, y) be after noise reduction pixel ((x y) is noise reduction preceding pixel point (x, ash y) to q for x, gray value y)Angle value, σ is pixel (x, y) neighborhood Lx,yInterior gray value mark is poor, q (i, j) ∈ [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] representNeighborhood Lx,yInterior gray value fall within interval [q (and x, y)-1.5 σ, q (x, y)+1.5 σ] point, k represents neighborhood Lx,yInterior gray value falls withinInterval
[q (and x, y)-1.5 σ, q (x, y)+1.5 σ] the quantity of point；
If (x, y) is non-boundary point, then carry out following noise reduction process:
In formula, (x y) is pixel (x, gray value y), q (i, j) representative image midpoint (i, j) ash at place after noise reduction to hAngle value, w (i) it is neighborhood lx,yInterior point (i, j) corresponding Gauss weight；
(3) coarse segmentation subelement, for slightly drawing the background in the cell image after denoising, Cytoplasm, nucleusPoint, particularly as follows:
By each pixel (x, y) represents with four dimensional feature vectors:
In formula, (x y) represents (x, gray value y), h to have(x y) represents its neighborhood Sx,yGray average, hmed(x, y) generationTable its neighborhood Sx,yGray scale intermediate value, hsta(x y) represents its neighborhood Sx,yGray variance；
K-means clustering procedure is used to be divided into background, Cytoplasm, nucleus three class；
(4) nuclear centers demarcates subelement, for demarcating nuclear centers:
Nucleus approximate region is obtained, if nuclear area comprises n point: (x by coarse segmentation subelement1,y1),…,(xn,yn), this region is carried out intensity-weighted demarcation and geometric center is demarcated, take its meansigma methods as nuclear centers (xz,yz):
(5) Accurate Segmentation subelement, for carrying out Accurate Segmentation to nucleus, Cytoplasm；
Build from nuclear centers (xz,yz) arrive nucleus and Cytoplasm boundary point (xp,yp) directed line segmentDistanceRepresent and round downwards；
Carry out sampling along line segment with unit length and can obtain dispIndividual point (x1,y1) ...,If samplingThe coordinate of point is not integer, and its gray value is obtained by surrounding pixel linear interpolation；
Point (xi,yi) place is along the gray scale difference of line segment direction:
hd(xi,yi)=h (xi-1,yi-1)-h(xi,yi)
Definition gray scale difference inhibition function:
Point (xi,yi) place is along the gradient gra (x of line segment directioni,yi):
Choose the maximum value point of gradient as nucleus and cytoplasmic precise edge.
This preferred embodiment arranges noise remove subelement, and effective integration center pixel closes on the space of neighborhood territory pixelProperty and grey similarity carry out noise reduction process, flat site in the picture, in neighborhood, grey scale pixel value is more or less the same, useGaussian filter is weighted filtering to gray value, and at the borderline region that change is violent, row bound keeps filtering, beneficially imageThe holding at edge；Use K mean cluster to extract nucleus and Cytoplasm coarse contour, can effectively remove the interference of noise；Arrange thinSubelement is demarcated at karyon center, it is simple to follow-up be accurately positioned nucleus and Cytoplasm profile；Accurate Segmentation subelement fillsDivide and make use of directional information, overcome the inflammatory cell interference to edge graph, it is possible to accurately extract nucleus and Cytoplasm limitEdge.
Preferably, the described textural characteristics to cell image extracts, including:
(1) the Gray co-occurrence matrix of cell image, described comprehensive ash is asked for based on the gray level co-occurrence matrixes method improvedDegree co-occurrence matrix embodies cell textural characteristics in different directions:
Be located at 0 °, 45 °, 90 °, gray level co-occurrence matrixes on 135 ° of four directions be respectively h (x, y, d, 0 °), h (x, y, d,45 °), h (x, y, d, 90 °), h (x, y, d, 135 °), corresponding matrix element project is X1、X2、X3、X4, then Gray is altogetherThe computing formula of raw matrix is:
H (x, y, d)=w1h(x,y,d,0°)+w2h(x,y,d,45°)+w3h(x,y,d,90°)+w4h(x,y,d,135°)
Gray co-occurrence matrix element number is:
In formula, d represents distance, and the span of d is [2,4], wiFor weight coefficient, i=1,2,3,4, it is by four sidesThe contrast level parameter that the gray level co-occurrence matrixes on each direction in is corresponding calculates, if the gray level co-occurrence matrixes on four directionCorresponding contrast level parameter is respectively Di, average isI=1,2,3,4, then weight coefficient wiComputing formula be:
(2) four textural characteristics parameters needed for utilizing described Gray co-occurrence matrix and matrix element project to obtain:Contrast, variance and, energy and average；
(3) described four textural characteristics parameters are normalized, the normalized textural characteristics value of final acquisition.
This preferred embodiment, based on the gray level co-occurrence matrixes method improved, uses the mode arranging weight coefficient to ask for cytological mapThe Gray co-occurrence matrix of picture, and then extract cell textural characteristics on appointment four direction, solve owing to outside is doneDisturb the textural characteristics ginseng of the cell that (cause such as lighting angle when cell image gathers impact, the flowing interference etc. of gas) causesNumerical value has the problem of bigger difference in different directions, improves the precision of cell image texture feature extraction；Selected contrast,Variance and, energy and four textural characteristics of average, eliminate the characteristic parameter of redundancy and repetition；To described four textural characteristics ginsengNumber is normalized, and the Classification and Identification facilitating follow-up cell image processes.
In this application scenarios, setting threshold value T=26, d=2, image denoising effect improves 7.5% relatively, cytological mapAs the extraction accuracy of feature improves 8%.
Last it should be noted that, above example is only in order to illustrate technical scheme, rather than the present invention is protectedProtecting the restriction of scope, although having made to explain to the present invention with reference to preferred embodiment, those of ordinary skill in the art shouldWork as understanding, technical scheme can be modified or equivalent, without deviating from the reality of technical solution of the present inventionMatter and scope.