CN105654141A - Isomap and SVM algorithm-based overlooked herded pig individual recognition method - Google Patents
Isomap and SVM algorithm-based overlooked herded pig individual recognition method Download PDFInfo
- Publication number
- CN105654141A CN105654141A CN201610006515.1A CN201610006515A CN105654141A CN 105654141 A CN105654141 A CN 105654141A CN 201610006515 A CN201610006515 A CN 201610006515A CN 105654141 A CN105654141 A CN 105654141A
- Authority
- CN
- China
- Prior art keywords
- pig
- individual
- isomap
- color
- individuality
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an Isomap and SVM algorithm-based overlooked herded pig individual recognition method and belongs to the field of machine vision and mode recognition. According to the method, firstly, the foreground detection and the target extraction are conducted for the acquired video frames of overlooked herded pigs, and then individual pigs can be obtained. Secondly, a training sample is established, and the color feature, the texture feature and the shape feature of individual pigs are extracted and combined to form feature vectors that represent individual pigs. Thirdly, the feature fusion is conducted on the above combined features based on the Isomap algorithm. On the premise that the effective recognition information is maintained maximally, the feature dimension is lowered. Finally, the training and the recognizing are conducted by a kernel function-optimized support vector machine classifier. The above method provides a novel idea for the recognition of non-excitable individual pigs, and also lays a foundation for the further research on the behavior analysis of individual pigs.
Description
Technical field
The present invention relates to machine vision technique and mode identification technology, be specifically related under a kind of vertical view state pig individual discrimination method in group support pig monitor video.
Background technology
Development along with scale pig industry and computer technology, inquire into the foreground detection overlooked in group support pig video sequence based on machine vision, pig individual behavior analysis etc. has increasingly been subject to the concern (GuoYizheng of Chinese scholars, ZhuWeixing, JiaoPengpeng, etal.Multi-objectextractionfromtopviewgroup-housedpigima gesbasedonadaptivepartitioningandmultilevelthresholdings egmentation [J] .Biosystemsengineering, 2015, 135 (5): 54-60.). the problem that wherein difficulty is maximum and crucial is the identification that pig is individual in motor process, now widely used method is ear tag RFID (MaselyneJ, SaeysW, NuffelAV.Review:quantifyinganimalfeedingbehaviourwithafo cusonpigs [J] .Physiology&Behavior, 2015, 138:37-51.), Kashiha etc. (2014) are in order to identify that in group support pig, pig is individual, back in pig coats the labelling of different colours and shape, individual (the KashihaMA of pig is identified by methods such as ellipse fittings, BahrC, OttS, etal.Automaticidentificationofmarkedpigsinapenusingimage patternrecognition [J] .ComputersandElectronicsinAgriculture, 2013, 93 (2): 111-120.). in addition, machine vision technique is utilized also not have pertinent literature to report the research of scale pig farm group support pig individual identification.
Consider machine vision algorithm successful Application (such as recognition of face, iris identification, fingerprint recognition, Gait Recognition etc.) in the identification of people, inquire into the pig individual identification based on machine vision and be possibly realized.
Summary of the invention
It is an object of the invention to pig individuality in group support pig monitor video under vertical view state is identified, propose a kind of based on Isomap (IsometricMapping for this, Isometric Maps) and the vertical view group support pig individual discrimination method of SVM (supportvectormachine, support vector machine) algorithm.
The technical solution used in the present invention is: the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm comprises the following steps:
Step 1, obtains group support pig video sequence under vertical view state, then does Image semantic classification, and mainly group support pig foreground detection is extracted with pig individual goal;Step 2, pig personal feature is extracted, and including using color moment method to extract the color characteristic that pig is individual, uses co-occurrence matrix method to extract the textural characteristics that pig is individual, using not displacement method to extract the shape facility that pig is individual, can effectively describe, thus establishing, the characteristic parameter that pig is individual; Step 3, after respectively the color characteristic in step 2, textural characteristics and shape facility being standardized, adopts Isomap algorithm to carry out merging optimization; Step 4, utilizes the support vector machine classifier optimizing kernel function that sample is trained, and each pig individuality treating knowledge is identified.
Further, described step 1 specifically includes:
First reconstruction experiment pig house, installs shooting directly over pig house and overlooks the image capturing system of video, obtain and overlook group support pig color video fragment; Then qualified frame of video is chosen, single-frame images is done histogram equalization and the segmentation of maximum entropy threshold value, segmentation result doing mathematics Morphological scale-space to pig zone of action further, is only comprised the image of foreground target, and then is extracted each pig individual goal.
Further, the described concrete grammar to the segmentation result doing mathematics Morphological scale-space of pig zone of action is:
Step 1.1, result is done opening operation by the disc structure element utilizing radius fixing, removes the isolated noise point in result, disconnects the adhesion between target and between target and background, keeps target life size to be basically unchanged;
Step 1.2, labelling connected region, target is removed less than the region of certain number of pixels, because according to pig individual characteristic, being foreground target region scarcely less than the connected region of certain area;
Step 1.3, fills bianry image, i.e. non-prospect cavity within filling part pig individuality.
Further, the detailed process of described step 2 is:
Step 2.1, based on the pig individuality color feature extracted of color moment: owing to colouring information is concentrated mainly in low-order moment, therefore only need to extract single order, second order and third moment to each color passage and add up; Each pig individuality is extracted R, B, G 9 color characteristics of tri-passages, is designated as [��R��RSR��G��GSG��B��BSB];
Step 2.2, based on the pig individuality texture feature extraction of co-occurrence matrix: co-occurrence matrix is described as be on �� direction, a pair pixel of standoff distance d is respectively provided with gray value i and the j probability occurred, is designated as p (i, j; D, ��); Selecting angle second moment, contrast, unfavourable balance to divide square, entropy these four features having descriptive power most respectively, each pig individuality is extracted 4 textural characteristics, is designated as [ASMCONIDMENT];
Step 2.3, based on the pig individual shapes feature extraction of not bending moment: (x, y), the geometric moment definition on its p+q rank is the discrete picture f of a width M �� NWherein, i �� M, j �� N, p, q are constant; OrderR=(p+q+2)/2, can derive 7 translations, convergent-divergent, rotationally-varying constant absolute inequality, can extract 7 invariant moment features shape facility as image accordingly; So, every pig individuality is extracted 7 shape facilities, is designated as [M1M2M3M4M5M6M7]��
Further, described step 3 detailed process is:
Step 3.1, in the characteristic extraction procedure that pig is individual, first color characteristic, textural characteristics and shape facility are standardized, between proper normalization to [-1+1], color characteristic, textural characteristics and shape facility combine and build the characteristic vector characterizing pig individuality;
Step 3.2, adopts Isomap algorithm to carry out Feature Fusion optimization; Isomap algorithm main process is as follows:
Step 3.2.1, it is determined that neighborhood: determine each neighborhood of a point with k nearest neighbour method in theorem in Euclid space, and calculate the distance between each point and its all neighborhood point;
Step 3.2.2, calculates shortest distance matrix: obtains neighborhood distance afterwards by step 3.2.1, then calculates any two points shortest path in Neighborhood Graph, is write the result obtained as matrix form, be designated as DN��N;
Step 3.2.3, the low-dimensional solving manifold with multi-dimentional scale conversion MDS represents Y: set HN��N=-JDJ/2, wherein J=I-eeT/ N, e=(1 ..., 1)TFor N dimensional vector, I is that N ties up unit matrix. Solve maximum f the eigenvalue �� of H1..., ��fAnd characteristic of correspondence vector u1..., uf, and remember U=[u1..., uf], then
Further, in the support vector machine classifier of described step 4, the kernel function of support vector machine is to combine radial direction base RBF kernel function and Sigmoid kernel function to be constructed as follows mixed kernel function:
K (x, y)=(1-��) Ksigmoid(x, y)+�� Krbf(x, y)
=(1-��) tanh (v (xi��xj)-r)+��exp(-��||x-y||2)
Wherein �� �� (0,1), for the weights of kernel function; Parameter C and the �� of mixed kernel function can use particle swarm optimization algorithm to obtain; Each pig individuality finally treating knowledge is identified.
The invention has the beneficial effects as follows:
Traditional Man is observed and is identified that the mode of pig individuality wastes time and energy, and affects the health of staff. Although ear tag RFID mode is without artificial Real Time Observation, but also can disturb pig normal growth to a certain extent. Proposed here a kind of method pig individuality stress not being identified by machine vision technique. Utilize Isomap algorithm that color, texture and shape three category feature are merged, both can remove the redundancy between feature, retain identification information, it is also possible to reduce intrinsic dimensionality, reduce classified counting amount. SVM is utilized can effectively to process multiclass identification problem, the mixed kernel function proposed both had considered global property, have also contemplated that local characteristics, namely the learning capacity of SVM and generalization ability are obtained for reinforcement, and the more single kernel function SVM of SVM based on mixed kernel function is more suitable for overlooking the individual identification of the many pigs of group support. Isomap algorithm and mixed kernel function SVM are combined, has given full play to respective advantage, efficiently solve non-rigid under complex background and overlook the identification problem of group support pig. This pig individuality stress not be identified by machine vision technique, both the treatment that pig is provided sufficient by poultry raiser it had been easy to, improve the welfare of pig, also provide technical foundation for later stage pig individual behavior analysis, to improving breeding scale industry automatization and Intellectualized monitoring level offer technical support.
Accompanying drawing explanation
Below in conjunction with the drawings and specific embodiments, the present invention is described in further details:
Fig. 1 is pig individual identification flow chart.
Fig. 2 is certain two field picture foreground detection result example.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described.
Fig. 1 is pig individual identification flow chart, and wherein Image semantic classification purpose is foreground detection, obtains each pig individual goal; Feature extraction comprises the color feature extracted based on color moment, based on the texture feature extraction of gray level co-occurrence matrixes, based on the Shape Feature Extraction of not bending moment; What Feature Fusion optimization adopted is Isomap algorithm; Model of cognition adopts the support vector machine classifier optimizing kernel function; May finally realize based on machine vision automatically stress not each pig individuality of identification.
Step 1: obtain group support pig video sequence under vertical view state, then do Image semantic classification, mainly group support pig foreground detection is extracted with pig individual goal.
(1) reconstruction pig house, obtains group support pig video sequence under vertical view state.
Concrete grammar is to install shooting directly over pig house to overlook the image capturing system of video, obtains and overlooks group support pig color video fragment.
(2) histogram equalization of image and the segmentation of maximum entropy threshold value.
Choose qualified frame of video, single-frame images is done histogram equalization and the segmentation of maximum entropy threshold value, the segmentation result doing mathematics Morphological scale-space to pig zone of action further, is only comprised the image of foreground target, and then extract each pig individual goal.
Wherein morphology processing specifically includes: result is done "ON" computing by the disc structure element 1) utilizing radius fixing, remove the isolated noise point in result, disconnect between target and adhesion between target and background, target life size can be kept again to be basically unchanged; 2) labelling connected region, removes target less than the region of certain number of pixels, because according to pig individual characteristic, being foreground target region scarcely less than the connected region of certain area (in region, sum of all pixels is less than 18000); 3) bianry image, i.e. non-prospect " cavity " within filling part pig individuality are filled. Fig. 2 is certain two field picture foreground detection result example, and then can extract each pig individual goal.
Step 2: pig personal feature is extracted, the color characteristic that pig is individual is extracted including using color moment method, use co-occurrence matrix method to extract the textural characteristics that pig is individual, using not displacement method to extract the shape facility that pig is individual, can effectively describe, thus establishing, the characteristic parameter that pig is individual.
(1) based on the pig individuality color feature extracted of color moment
Owing to colouring information is concentrated mainly in low-order moment, therefore only each color passage need to be extracted single order, second order and third moment and add up. If hijRepresenting that in i-th Color Channel component, gray scale is the probability of the pixel appearance of j, n is total number-of-pixels, then three low-order moment mathematic(al) representations of color moment are:
These 3 low-order moment are called average, variance and degree of skewness.
(2) based on the pig individuality texture feature extraction of co-occurrence matrix
Co-occurrence matrix is described as be on �� direction, and a pair pixel of standoff distance d is respectively provided with gray value i and the j probability occurred, is designated as p (i, j; D, ��). Redundancy is there is between 14 features of gray level co-occurrence matrixes in that considers that Haralick proposes for analyzing, here comprehensively having selected angle second moment ASM (AngularSecondMoment), contrast C ON (Contrast), unfavourable balance to divide 4 features having descriptive power most such as square IDM (InverseDifferenceMoment), entropy ENT (Entropy), its definition is as follows respectively:
(3) based on the pig individual shapes feature extraction of not bending moment
(x, y), the geometric moment definition on its p+q rank is the discrete picture f of one width M �� NWherein, i �� M, j �� N, p, q are constant. OrderR=(p+q+2)/2, can derive 7 displacements, convergent-divergent, rotationally-varying constant absolute inequality, can extract 7 invariant moment features shape facility as image accordingly. As follows respectively:
M1=��20+��02(8)
M3=(��30-3��12)2+(3��21-��03)2(10)
M4=(��30+��12)2+(��21+��03)2(11)
M5=(��30-3��12)��30+��12)[(��30+��12)2-3(��21+��03)2]+(3��21-��03)(��21+��03)[3(��30+��12)2-(��21+��03)2](12)
M6=(��20-3��02)[(��30+��12)2-(��21+��03)2]+4��11(��30+��12)2(��21+��03)(13)
M7=(3 ��12-��03)(��30+��12)[(��30+��12)2-3(��21+��03)2]+(��30-3��12)(��21+��03)[3(��30+��12)2-(��21+��03)2](14)
To sum up, each pig individuality is extracted R, B, G 9 color characteristics of tri-passages; It is extracted 4 textural characteristics based on gray level co-occurrence matrixes; It is extracted 7 shape facilities based on not bending moment, collectively forms and characterize the characteristic parameter that pig is individual.
Step 3: after respectively the color in step 2, texture and shape facility being standardized, adopts Isomap algorithm to carry out merging optimization.
First above-mentioned color, texture and shape facility are standardized, between proper normalization to [-1+1], combined to build by color characteristic, textural characteristics and shape facility and characterize the characteristic vector that pig is individual, affect classifying quality because magnitude range differs eliminating numerical value.
Then Isomap algorithm is adopted to carry out Feature Fusion optimization. Algorithm main process is as follows:
1) neighborhood is determined. Theorem in Euclid space determines each neighborhood of a point with k nearest neighbour method, and calculates the distance between each point and its all neighborhood point.
2) shortest distance matrix is calculated. By step 1) obtain neighborhood distance afterwards, then calculate any two points shortest path in Neighborhood Graph, write the result obtained as matrix form, be designated as DN��N��
3) convert MDS with multi-dimentional scale to solve the low-dimensional of manifold and represent Y. If HN��N=-JDJ/2, wherein J=I-eeT/ N, e=(1 ..., 1)TFor N dimensional vector, I is that N ties up unit matrix. Solve maximum f the eigenvalue �� of H1..., ��fAnd characteristic of correspondence vector u1..., uf, and remember U=[u1..., uf], then
Step 4: utilizing the support vector machine classifier optimizing kernel function that sample is trained, each pig individuality treating knowledge is identified.
It is provided with classified sample setWherein sampling feature vectorsNamelyBeing the G vector tieing up in real number space, (-1 ,+1}, S is classification samples number to class label y ��. Build a classification function by these samples, make it test data are correctly classified as far as possible. Classification seeks a hyperplane exactlyMake it completely separable for two class samples, whereinFor planar process vector, b is bias term.
Realistic problem is nearly all Nonlinear separability, and uses kernel function that the nonlinear input space can be mapped to the space of a linear separability. Now optimization object function is:
To new sample, classification function is:
Wherein �� is Lagrange factor, and K is kernel function, and C is penalty factor.
Kernel function support vector machine has become as a kind of instrument that the problem aspects such as classification and recurrence are popular and the most powerful that solves at present. Different kernel functions will produce different classifying qualities, two class kernel functions is combined here, is constructed as follows mixed kernel function:
K (x, y)=(1-��) Ksigmoid(x, y)+�� Krbf(x, y) (18)
=(1-��) tanh (v (xi��xj)-r)+��exp(-��||x-y||2)
Wherein �� �� (0,1), for the weights of kernel function. Parameter C and the �� of mixed kernel function can use particle swarm optimization algorithm to obtain.
Preferred embodiment:
One optimum detailed description of the invention of the present invention: reconstruction pig house, under acquisition vertical view state in group support pig video sequence process, 3m place directly over pig house (long * width * height is 3.5m*3m*1m), shooting is installed and overlooks the image capturing system of video, obtain hutch number be 7 to 12, different growing stages, the color video fragment that comprises complex scene, image resolution ratio is 1760 �� 1840 pixels. In pig personal feature extraction process, each pig individuality is extracted R, B, G 9 color characteristics of tri-passages, is designated as [��R��RSR��G��GSG��B��BSB], so can obtain the pig the most direct visual signature of individuality; Being extracted 4 textural characteristics based on gray level co-occurrence matrixes, be designated as [ASMCONIDMENT], 4 textural characteristics can fully describe the key character of pig individuality surface texturisation structure; It is extracted 7 shape facilities based on not bending moment, is designated as [M1M2M3M4M5M6M7], these 7 shape facilities effect in differentiation pig individual shapes is just more apparent, and three category features together constitute and characterize the characteristic parameter that pig is individual. Neighborhood k value 4 in Isomap algorithm, calculates 2 shortest paths in Neighborhood Graph and uses dijkstra's algorithm, and optimum dimension is 5. Mixed kernel function SVM classifier major parameter is as follows: v takes 1/5, r and takes 3, C and take 16, �� and take 1.82, �� and take 0.8.Through above-mentioned steps, finally achieve overlooking the identification that in group support pig image sequence, pig is individual.
In sum, a kind of of the present invention utilizes machine vision technique to overlooking the method that in group support pig image sequence, pig individuality is identified, and first the vertical view group support pig frame of video gathered is carried out foreground detection and Objective extraction, it is thus achieved that single pig individuality; Thereafter setting up training sample, extract pig individuality color, texture and shape facility, combination builds and characterizes the characteristic vector that pig is individual; Assemblage characteristic then utilizes Isomap algorithm do Feature Fusion, at utmost retaining, the basis effectively identifying information reduces intrinsic dimensionality; The support vector machine classifier optimizing kernel function is finally utilized to be trained and identification. The research be stress not pig individual identification provide new approaches, also lay a good foundation for exploring further group support pig individual behavior analysis etc.
In the description of this specification, specific features, structure, material or feature that the description of reference term " embodiment ", " some embodiments ", " illustrative examples ", " example ", " concrete example " or " some examples " etc. means in conjunction with this embodiment or example describe are contained at least one embodiment or the example of the present invention. In this manual, the schematic representation of above-mentioned term is not necessarily referring to identical embodiment or example. And, the specific features of description, structure, material or feature can combine in an appropriate manner in any one or more embodiments or example.
Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that: these embodiments can being carried out multiple change, amendment, replacement and modification when without departing from principles of the invention and objective, the scope of the present invention is limited by claim and equivalent thereof.
Claims (6)
1. based on the vertical view group support pig individual discrimination method of Isomap and SVM algorithm, it is characterised in that comprise the following steps:
Step 1, obtains group support pig video sequence under vertical view state, then does Image semantic classification, and mainly group support pig foreground detection is extracted with pig individual goal; Step 2, pig personal feature is extracted, and including using color moment method to extract the color characteristic that pig is individual, uses co-occurrence matrix method to extract the textural characteristics that pig is individual, using not displacement method to extract the shape facility that pig is individual, can effectively describe, thus establishing, the characteristic parameter that pig is individual; Step 3, after respectively the color characteristic in step 2, textural characteristics and shape facility being standardized, adopts Isomap algorithm to carry out merging optimization; Step 4, utilizes the support vector machine classifier optimizing kernel function that sample is trained, and each pig individuality treating knowledge is identified.
2. the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm according to claim 1, it is characterised in that: described step 1 specifically includes:
First reconstruction experiment pig house, installs shooting directly over pig house and overlooks the image capturing system of video, obtain and overlook group support pig color video fragment; Then qualified frame of video is chosen, single-frame images is done histogram equalization and maximum entropy threshold segmentation, segmentation result doing mathematics Morphological scale-space to pig zone of action further, is only comprised the image of foreground target, and then is extracted each pig individual goal.
3. the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm according to claim 2, it is characterised in that: the described concrete grammar to the segmentation result doing mathematics Morphological scale-space of pig zone of action is:
Step 1.1, result is done opening operation by the disc structure element utilizing radius fixing, removes the isolated noise point in result, disconnects the adhesion between target and between target and background, keeps target life size to be basically unchanged;
Step 1.2, labelling connected region, target is removed less than the region of certain number of pixels, because according to pig individual characteristic, being foreground target region scarcely less than the connected region of certain area;
Step 1.3, fills bianry image, i.e. non-prospect cavity within filling part pig individuality.
4. the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm according to claim 1, it is characterised in that: the detailed process of described step 2 is:
Step 2.1, based on the pig individuality color feature extracted of color moment: owing to colouring information is concentrated mainly in low-order moment, therefore only need to extract single order, second order and third moment to each color passage and add up; Each pig individuality is extracted R, B, G 9 color characteristics of tri-passages, is designated as [��R��RSR��G��GSG��R��RSB];
Step 2.2, based on the pig individuality texture feature extraction of co-occurrence matrix: co-occurrence matrix is described as be on �� direction, a pair pixel of standoff distance d is respectively provided with gray value i and the j probability occurred, is designated as p (i, j; D, ��); Selecting angle second moment, contrast, unfavourable balance to divide square, entropy these four features having descriptive power most respectively, each pig individuality is extracted 4 textural characteristics, is designated as [ASMCONIDMENT];
Step 2.3, based on the pig individual shapes feature extraction of not bending moment: (x, y), the geometric moment definition on its p+q rank is the discrete picture f of a width M �� NWherein, i �� M, j �� N, p, q are constant; OrderR=(p+q+2)/2, can derive 7 translations, convergent-divergent, rotationally-varying constant absolute inequality, can extract 7 invariant moment features shape facility as image accordingly; So, every pig individuality is extracted 7 shape facilities, is designated as [M1M2M3M4M5M6M7]��
5. the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm according to claim 1, it is characterised in that: the detailed process of described step 3 is:
Step 3.1, in the characteristic extraction procedure that pig is individual, first color characteristic, textural characteristics and shape facility are standardized, between proper normalization to [-1+1], color characteristic, textural characteristics and shape facility combine and build the characteristic vector characterizing pig individuality;
Step 3.2, adopts Isomap algorithm to carry out Feature Fusion optimization; Isomap algorithm main process is as follows:
Step 3.2.1, it is determined that neighborhood: determine each neighborhood of a point with k nearest neighbour method in theorem in Euclid space, and calculate the distance between each point and its all neighborhood point;
Step 3.2.2, calculates shortest distance matrix: obtains neighborhood distance afterwards by step 3.2.1, then calculates any two points shortest path in Neighborhood Graph, is write the result obtained as matrix form, be designated as DN��N;
Step 3.2.3, the low-dimensional solving manifold with multi-dimentional scale conversion MDS represents Y: set HN��N=-JDJ/2, wherein J=I-eeT/ N, e=(1 ..., 1)TFor N dimensional vector, I is that N ties up unit matrix, solves maximum f the eigenvalue �� of H1..., ��fAnd characteristic of correspondence vector u1..., uf, and remember U=[u1..., uf], then
6. the vertical view group support pig individual discrimination method based on Isomap and SVM algorithm according to claim 1, it is characterized in that: in the support vector machine classifier of described step 4, the kernel function of support vector machine is to combine radial direction base RBF kernel function and Sigmoid kernel function to be constructed as follows mixed kernel function:
K (x, y)=(1-��) Ksigmoid(x, y)+�� Krbf(x, y)
=(1-��) tanh (�� (xi��xj)-r)+��exp(-��||x-y||2)
Wherein �� �� (0,1), for the weights of kernel function;Parameter C and the �� of mixed kernel function can use particle swarm optimization algorithm to obtain; Each pig individuality finally treating knowledge is identified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610006515.1A CN105654141A (en) | 2016-01-06 | 2016-01-06 | Isomap and SVM algorithm-based overlooked herded pig individual recognition method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610006515.1A CN105654141A (en) | 2016-01-06 | 2016-01-06 | Isomap and SVM algorithm-based overlooked herded pig individual recognition method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105654141A true CN105654141A (en) | 2016-06-08 |
Family
ID=56491509
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610006515.1A Pending CN105654141A (en) | 2016-01-06 | 2016-01-06 | Isomap and SVM algorithm-based overlooked herded pig individual recognition method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105654141A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778784A (en) * | 2016-12-20 | 2017-05-31 | 江苏大学 | Pig individual identification and drinking behavior analysis method based on machine vision |
CN107027650A (en) * | 2017-03-21 | 2017-08-11 | 中国农业大学 | A kind of boar abnormal state detection method and device based on PSO SVM |
CN108090426A (en) * | 2017-12-07 | 2018-05-29 | 江苏大学 | A kind of group rearing pig personal identification method based on machine vision |
CN108108700A (en) * | 2017-12-27 | 2018-06-01 | 江苏大学 | A kind of characteristic area recognition methods of the pig based on peg conversion |
CN108846326A (en) * | 2018-05-23 | 2018-11-20 | 盐城工学院 | The recognition methods of pig posture, device and electronic equipment |
CN109410208A (en) * | 2018-11-14 | 2019-03-01 | 成都极致智造科技有限公司 | The machine learning identification of Wear Mechanism of Abrasive Belt and process parameter optimizing method |
CN110008881A (en) * | 2019-03-28 | 2019-07-12 | 北京农业信息技术研究中心 | The recognition methods of the milk cow behavior of multiple mobile object and device |
CN110145695A (en) * | 2019-06-03 | 2019-08-20 | 大连理工大学 | A kind of hot duct leakage detection method based on the fusion of the depth confidence network information |
CN110222664A (en) * | 2019-06-13 | 2019-09-10 | 河南牧业经济学院 | A kind of feeding monitoring system of intelligent pigsty based on the analysis of video activity |
CN110490161A (en) * | 2019-08-23 | 2019-11-22 | 安徽农业大学 | A kind of captive animal behavior analysis method based on deep learning |
CN111242035A (en) * | 2020-01-14 | 2020-06-05 | 江苏大学 | Local feature-based group-breeding pig identity identification method |
CN111340061A (en) * | 2020-01-18 | 2020-06-26 | 中国人民解放军国防科技大学 | Multi-mode data fusion and classification method based on SVM model parameter optimization |
CN113033349A (en) * | 2021-03-11 | 2021-06-25 | 北京文安智能技术股份有限公司 | Overlook image selection method for pedestrian re-identification, storage medium and electronic device |
CN113283405A (en) * | 2021-07-22 | 2021-08-20 | 第六镜科技(北京)有限公司 | Mask detection method and device, computer equipment and storage medium |
CN118089899A (en) * | 2024-04-19 | 2024-05-28 | 中农吉牧(吉林)农业发展有限公司 | Intelligent weighing system of intelligent cattle farm |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104252709A (en) * | 2014-07-14 | 2014-12-31 | 江苏大学 | Multiple-target foreground detection method for look-down group-housed pigs in look-down state under complicated background |
CN104504704A (en) * | 2014-12-24 | 2015-04-08 | 江苏大学 | Multiple overlooked herded pig target extraction method capable of self-adaptive multi-threshold segmentation of blocks |
-
2016
- 2016-01-06 CN CN201610006515.1A patent/CN105654141A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104252709A (en) * | 2014-07-14 | 2014-12-31 | 江苏大学 | Multiple-target foreground detection method for look-down group-housed pigs in look-down state under complicated background |
CN104504704A (en) * | 2014-12-24 | 2015-04-08 | 江苏大学 | Multiple overlooked herded pig target extraction method capable of self-adaptive multi-threshold segmentation of blocks |
Non-Patent Citations (4)
Title |
---|
YIZHENG GUO ET AL: "Foreground detection of group-housed pigs based on the combination of Mixture of Gaussians using prediction mechanism and threshold segmentation", 《BIOSYSTEMS ENGINEERING》 * |
何利: "基于图像纹理颜色特征的猪体识别研究与实现", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
刘倩等: "基于Isomap的人脸图像特征融合识别算法", 《科技通报》 * |
郭依正: "基于多特征融合的医学图像识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778784A (en) * | 2016-12-20 | 2017-05-31 | 江苏大学 | Pig individual identification and drinking behavior analysis method based on machine vision |
CN106778784B (en) * | 2016-12-20 | 2020-06-09 | 江苏大学 | Pig individual identification and drinking behavior analysis method based on machine vision |
CN107027650A (en) * | 2017-03-21 | 2017-08-11 | 中国农业大学 | A kind of boar abnormal state detection method and device based on PSO SVM |
CN108090426A (en) * | 2017-12-07 | 2018-05-29 | 江苏大学 | A kind of group rearing pig personal identification method based on machine vision |
CN108108700B (en) * | 2017-12-27 | 2020-02-21 | 江苏大学 | Pig feature region identification method based on chord axis transformation |
CN108108700A (en) * | 2017-12-27 | 2018-06-01 | 江苏大学 | A kind of characteristic area recognition methods of the pig based on peg conversion |
CN108846326A (en) * | 2018-05-23 | 2018-11-20 | 盐城工学院 | The recognition methods of pig posture, device and electronic equipment |
CN109410208A (en) * | 2018-11-14 | 2019-03-01 | 成都极致智造科技有限公司 | The machine learning identification of Wear Mechanism of Abrasive Belt and process parameter optimizing method |
CN110008881B (en) * | 2019-03-28 | 2021-03-12 | 北京农业信息技术研究中心 | Method and device for identifying cow behaviors of multiple moving targets |
CN110008881A (en) * | 2019-03-28 | 2019-07-12 | 北京农业信息技术研究中心 | The recognition methods of the milk cow behavior of multiple mobile object and device |
CN110145695A (en) * | 2019-06-03 | 2019-08-20 | 大连理工大学 | A kind of hot duct leakage detection method based on the fusion of the depth confidence network information |
CN110222664A (en) * | 2019-06-13 | 2019-09-10 | 河南牧业经济学院 | A kind of feeding monitoring system of intelligent pigsty based on the analysis of video activity |
CN110222664B (en) * | 2019-06-13 | 2021-07-02 | 河南牧业经济学院 | Intelligent pig housing monitoring system based on video activity analysis |
CN110490161A (en) * | 2019-08-23 | 2019-11-22 | 安徽农业大学 | A kind of captive animal behavior analysis method based on deep learning |
CN111242035A (en) * | 2020-01-14 | 2020-06-05 | 江苏大学 | Local feature-based group-breeding pig identity identification method |
CN111242035B (en) * | 2020-01-14 | 2023-08-22 | 江苏大学 | Group pig raising identity recognition method based on local characteristics |
CN111340061A (en) * | 2020-01-18 | 2020-06-26 | 中国人民解放军国防科技大学 | Multi-mode data fusion and classification method based on SVM model parameter optimization |
CN113033349A (en) * | 2021-03-11 | 2021-06-25 | 北京文安智能技术股份有限公司 | Overlook image selection method for pedestrian re-identification, storage medium and electronic device |
CN113033349B (en) * | 2021-03-11 | 2023-12-26 | 北京文安智能技术股份有限公司 | Overhead image selection method for pedestrian re-recognition, storage medium and electronic equipment |
CN113283405A (en) * | 2021-07-22 | 2021-08-20 | 第六镜科技(北京)有限公司 | Mask detection method and device, computer equipment and storage medium |
CN118089899A (en) * | 2024-04-19 | 2024-05-28 | 中农吉牧(吉林)农业发展有限公司 | Intelligent weighing system of intelligent cattle farm |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105654141A (en) | Isomap and SVM algorithm-based overlooked herded pig individual recognition method | |
Zhou et al. | Evaluation of fish feeding intensity in aquaculture using a convolutional neural network and machine vision | |
CN109034210B (en) | Target detection method based on super-feature fusion and multi-scale pyramid network | |
CN106778902B (en) | Dairy cow individual identification method based on deep convolutional neural network | |
Zhang et al. | Real-time sow behavior detection based on deep learning | |
CN107316307B (en) | Automatic segmentation method of traditional Chinese medicine tongue image based on deep convolutional neural network | |
CN107609601B (en) | Ship target identification method based on multilayer convolutional neural network | |
CN107292298B (en) | Ox face recognition method based on convolutional neural networks and sorter model | |
CN111178197B (en) | Mass R-CNN and Soft-NMS fusion based group-fed adherent pig example segmentation method | |
CN104992223B (en) | Intensive population estimation method based on deep learning | |
CN108846446B (en) | Target detection method based on multi-path dense feature fusion full convolution network | |
CN107194937B (en) | Traditional Chinese medicine tongue picture image segmentation method in open environment | |
CN103514456B (en) | Image classification method and device based on compressed sensing multi-core learning | |
CN111783576B (en) | Pedestrian re-identification method based on improved YOLOv3 network and feature fusion | |
WO2020133636A1 (en) | Method and system for intelligent envelope detection and warning in prostate surgery | |
CN111931811B (en) | Calculation method based on super-pixel image similarity | |
CN109389129A (en) | A kind of image processing method, electronic equipment and storage medium | |
CN112052772A (en) | Face shielding detection algorithm | |
CN109472733A (en) | Image latent writing analysis method based on convolutional neural networks | |
Shafi et al. | Embedded AI for wheat yellow rust infection type classification | |
CN112507904A (en) | Real-time classroom human body posture detection method based on multi-scale features | |
CN109657704A (en) | A kind of coring scene characteristic extracting method based on sparse fusion | |
CN105868711A (en) | Method for identifying human body behaviors based on sparse and low rank | |
Qian et al. | Classification of rice seed variety using point cloud data combined with deep learning | |
Wang et al. | Pig face recognition model based on a cascaded network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160608 |
|
RJ01 | Rejection of invention patent application after publication |