CN112122175A - Material enhanced feature recognition and selection method of color sorter - Google Patents

Material enhanced feature recognition and selection method of color sorter Download PDF

Info

Publication number
CN112122175A
CN112122175A CN202010804337.3A CN202010804337A CN112122175A CN 112122175 A CN112122175 A CN 112122175A CN 202010804337 A CN202010804337 A CN 202010804337A CN 112122175 A CN112122175 A CN 112122175A
Authority
CN
China
Prior art keywords
weighted
feature
materials
characteristic
component
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.)
Granted
Application number
CN202010804337.3A
Other languages
Chinese (zh)
Other versions
CN112122175B (en
Inventor
冯毅雄
岑鸿晋
洪兆溪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University ZJU
Original Assignee
Zhejiang University ZJU
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University ZJU filed Critical Zhejiang University ZJU
Priority to CN202010804337.3A priority Critical patent/CN112122175B/en
Publication of CN112122175A publication Critical patent/CN112122175A/en
Application granted granted Critical
Publication of CN112122175B publication Critical patent/CN112122175B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/34—Sorting according to other particular properties
    • B07C5/342—Sorting according to other particular properties according to optical properties, e.g. colour
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
    • B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
    • B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
    • B07C5/36—Sorting apparatus characterised by the means used for distribution
    • B07C5/361—Processing or control devices therefor, e.g. escort memory
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/90—Determination of colour characteristics
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a material enhanced feature identification and selection method of a color sorter. The method comprises the steps of obtaining a material image by using a color selector, removing a background in the image, and extracting the integral gray characteristic and the local gray characteristic of a single material from the material image one by one; converting each area of an inner circle of the material into a weighted undirected graph node and an edge, constructing a gray level feature weighted undirected graph of a local area of the material, and obtaining a graph feature component and a feature vector of each material; calculating the distribution intervals of the whole HSI component and the graph characteristic component of the material by using the characteristic vector; and when the color sorter works, image data of the materials to be sorted are processed and then compared with the determined distribution intervals of the whole HSI component and the local map characteristic component of the materials, the materials to be sorted which exceed the distribution intervals of the corresponding components are marked as unqualified materials, and the air injection valve is started to realize material separation. The invention combines the integral and local characteristics of a single material, better describes the material properties and improves the color sorting precision.

Description

Material enhanced feature recognition and selection method of color sorter
Technical Field
The invention relates to a material image processing and screening method in the technical field of color sorters, in particular to a material enhanced feature recognition and selection method of a color sorter.
Background
The color sorter is a product combining the fields of machinery, software development, image acquisition and processing and the like. Because the color sorter has the characteristics of high sorting precision, high sorting speed, strong adaptability of sorting algorithm and the like, the color sorter is widely applied to sorting treatment of materials such as grains, tea leaves, ores and the like. In order to improve the quality of products such as grains, tea leaves and the like, the traditional method is to identify the quality of materials to be selected through manual naked eyes, so that not only is great labor consumed, but also the classification precision is poor and the efficiency is low. The color sorter uses a conveying mechanism to convey a large amount of materials to be sorted to an image acquisition and picking device, acquires real-time conveying information of the materials through the image acquisition device such as a CCD (charge coupled device), outputs the positions of unqualified materials to the picking device such as an air jet valve after image processing and sorting rule classification, and opens an air valve to realize material separation.
The RGB color space is commonly used in display systems, which utilize the principle of superposition of three primary colors in physics to produce various colors. In the RGB color space, R, G, B color components are independent of each other. In the HSI color space, H denotes Hue (chroma), S denotes Saturation, and I denotes Intensity (Intensity). Chromaticity is used to distinguish a certain color, such as white, yellow, cyan, green, magenta, red, black, etc.; the saturation refers to the purity of the color, the more vivid the color is, the higher the saturation is, otherwise, the lower the saturation is; the brightness refers to the brightness of the color, and the higher the brightness, the brighter the color. The RGB color space has a disadvantage in that it does not conform to the visual characteristics of the human eye, and the HSI color space is not suitable for the display system, but more conforms to the visual characteristics of the human eye, and thus can be converted into the HSI color space upon color selection.
The current color sorter technology is continuously improved for improving the color sorting precision and reducing the carry-over ratio, and the image processing and sorting algorithm of the materials to be selected play an important role in the color sorting process. The existing color sorting algorithm, such as manual extraction of integral characteristics of materials to be sorted to make division, and classification by combining with a neural network, is limited by the quality of extracted characteristics, but is large in time consumption and cost overhead, and cannot well meet the requirement of improving the material sorting quality.
Disclosure of Invention
In order to solve the problems in the background art, the invention provides a material enhanced feature identification and sorting method of a color sorter.
The technical scheme of the invention is as follows:
the method comprises the following steps: acquiring images of N materials moving in real time by using a color selector to serve as material images, and removing backgrounds in the material images by using motion blur image restoration, median filtering denoising and a threshold method;
step two: extracting the integral gray feature and the local gray feature of a single material one by one from the material image, wherein the integral gray feature is the integral HSI component of the material, and the local gray feature is the gray average value of R, G, B channels in each area of the inner circle of the material;
the second step comprises the following specific steps:
s2.1, removing each ith material A of the backgroundiAs a sample example, i ═ 1, 2., N, i denotes the serial number of the material, and is placed in a blank picture of m × m pixels to obtain a sample example picture;
s2.2, according to the sample, the material A in the pictureiOuter contour point e ofi，1，ei，2，...，ei，nCoordinate calculation of the geometric center Fi：
Figure BDA0002628538240000021
Wherein e isi，1，ei，2，...，ei，nRespectively represent material AiCorresponding 1 st to nth outer contour points;
s2.3, calculating a material AiAll pixel points p in the outer contouri，1，pi，2，...，pi，sThe average value of RGB gray level obtains the whole of the materialVolume gray scale characteristics:
Figure BDA0002628538240000022
wherein p isi，j(R) is Material AiJ (th) pixel point p in outer contouri，jCorresponding grey value, R, of R channel in RGB color spaceiExpressing the gray scale characteristics of an R channel of the material in an RGB color space;
gray scale characteristic G of G channel of material in RGB color spaceiAnd the gray scale characteristic B of the B channeliGrayscale characterization R according to sum R channeliProcessing in the same way to obtain;
s2.4, mixing the material AiThe gray feature of the R, G, B channel is taken as the whole RGB information and converted into HSI information, and then H, S, I components of the HSI information are respectively calculated as the whole HSI component;
s2.5, establishing an inner circle DiInner circle DiCenter of circle MiWith the material AiGeometric center F ofiAre superposed and have an inner circle DiArea T ofiAnd material AiArea S ofiSatisfy Ti：Si＝1∶5；
Wherein, the material AiArea S ofiFrom the outer contour point ei，1，ei，2，...，ei，nThe total number of the pixel points in the surrounded area is obtained, and the inner circle DiArea calculation method and material AiArea S ofiThe calculation mode is the same;
s2.5.1, firstly, the inner circle DiDivided into concentric inner circles Q along the radius quarteringi，1And three rings Qi，2、Qi，3、Qi，4While the inner circle D is drawniEight equal divisions along the circumference to the inner circle Qi，1And three rings Qi，2、Qi，3、Qi，4The device is further divided into eight areas, and the eight areas are divided into 32 areas in total;
s2.5.2, calculating the inner circle D respectivelyiOf R, G, B lanes of 32 zonesAnd obtaining 96 gray average values as local gray features.
Step three: mixing the material AiInner circle D ofiIs converted into a weighted undirected graph Pi(V, E) nodes and edges, constructing a material local area gray level feature weighted undirected graph, and obtaining each material AiAnd feature vector Ki；
The third step comprises the following specific steps:
s3.1, firstly, mixing the material AiInner circle DiArea normalization of (2); respectively calculating the geometric center of each region as the node v of the region in the weighted undirected graphi，1,vi，2,...,vi，32，vi，1Represents Material AiThe corresponding 1 st area is the node in the weighted undirected graph, and all the nodes form the weighted undirected graph Pi(V, E) set of points Vi；
S3.2, combining and matching the 32 areas in pairs, and respectively calculating the color difference coefficient between the two matched areasu，vCoefficient of chromatic aberrationu，vTwo areas smaller than the color difference coefficient threshold phi are associated, and the association relation of the two areas is used as an edge Eu，vFrom all edges Eu，vForm a weighted undirected graph PiSet of edges E of (V, E)i；
The color difference coefficient between the two regionsu，vThe calculation is as follows:
Figure BDA0002628538240000031
wherein r isuRepresenting a grayscale characteristic of an R channel of the region u, Δ R representing a difference between the grayscale characteristic of the R channel of the region u and the grayscale characteristic of the R channel of the region v, Δ G representing a difference between the grayscale characteristic of a G channel of the region u and the grayscale characteristic of the G channel of the region v, and Δ B representing a difference between the grayscale characteristic of a B channel of the region u and the grayscale characteristic of the B channel of the region v;
s3.3, according to the weighted undirected graph PiDistance d between nodes in (V, E)u，vAnd coefficient of chromatic aberrationu，vCalculating the edge Eu，vWeight ω of (d)u，v：
Figure BDA0002628538240000032
S3.4, calculating weighted undirected graph PiGraph feature components of (V, E):
extracting weighted feature path lengths
Figure BDA0002628538240000033
Global efficiency
Figure BDA0002628538240000034
And weighted global clustering coefficients
Figure BDA0002628538240000035
As graph feature components of the material;
weighted feature path length
Figure BDA0002628538240000036
The weighted undirected graph is the average value of the lengths of all possible nodes to the shortest paths in the weighted undirected graph, the node pair is formed by any two nodes, the path length is the sum of the weights of all edges on one path between the two nodes in the node pair, and the length of the shortest path is the minimum value of the lengths of all possible paths between the two nodes in the node pair;
global efficiency
Figure BDA0002628538240000037
Is the average of the inverse of the shortest path length for all possible node pairs in the weighted undirected graph.
The weighted global clustering coefficients in S3.4
Figure BDA0002628538240000038
The formula is adopted for calculation;
Figure BDA0002628538240000039
wherein the content of the first and second substances,
Figure BDA0002628538240000041
is a material AiContains the sum of the geometric mean of the trilateral weights of all triangles of node u,
Figure BDA0002628538240000042
is a material AiF represents the weight sum of all edges where the node u is located, f represents the material AiThe number of nodes.
S3.5, mixing the material AiInner circle D ofiMap feature component of
Figure BDA0002628538240000043
Bulk HSI component Ai(H)、Ai(S)、Ai(S) and Category tag YiSequentially combined to obtain a description material AiFeature vector of attribute features
Figure BDA0002628538240000044
Wherein the class label YiTaking 0 or 1, respectively representing a material AiPass or fail.
Step four: using the feature vector K obtained in step threeiCalculating the distribution intervals of the whole HSI component and the characteristic component of the graph of the material;
the fourth step comprises the following specific steps:
for all materials AiCharacteristic vector K ofiFor the feature vector KiWherein the label of class division YiAll other components obtain the upper and lower limits of the component distribution interval according to the following processing to be used as the weighted characteristic path length of the first component
Figure BDA0002628538240000045
For example, the following are explained:
s4.1, mixing all the materials AiWeighted feature path length of
Figure BDA0002628538240000046
According to the numerical values which are arranged from small to large in sequence, the quartile T1 (L) under the arrangement is calculatedω) Upper quartile T3 (L)ω) And quartile IQR (L)ω)；
S4.2 according to quartile T1 (L)ω) Upper quartile T3 (L)ω) And quartile IQR (L)ω) Calculating to obtain maximum observed value
Figure BDA0002628538240000047
And minimum observed value
Figure BDA0002628538240000048
Figure BDA0002628538240000049
Figure BDA00026285382400000410
Wherein γ represents a distribution correction coefficient;
s4.3, for each material AiWeighted feature path length of
Figure BDA00026285382400000411
The weighted characteristic path length is obtained by performing judgment processing in the following way
Figure BDA00026285382400000412
Upper and lower limits corresponding to the components:
if there is at least one material AiWeighted feature path length of
Figure BDA00026285382400000413
Satisfy the requirement of
Figure BDA00026285382400000414
If the minimum observed value is represented, the lower limit corresponding to the weighted characteristic path length component is taken
Figure BDA00026285382400000415
Otherwise, the lower limit min (L) corresponding to the weighted characteristic path length component is takenω) For all the materials AiWeighted feature path length of
Figure BDA00026285382400000416
Minimum value of (1);
if there is at least one material AiWeighted feature path length of
Figure BDA00026285382400000417
Satisfy the requirement of
Figure BDA00026285382400000418
If the maximum observed value is represented, the upper limit corresponding to the weighted characteristic path length component is taken
Figure BDA00026285382400000419
Otherwise, taking the upper limit max (L) corresponding to the weighted characteristic path length componentω) For all the materials AiWeighted feature path length of
Figure BDA00026285382400000420
Maximum value of (1);
s4.4, respectively calculating the eigenvector K according to the same processing mode of the two stepsiExcept for class label YiOther than the rest
Figure BDA00026285382400000421
...，Ai(I) And obtaining the distribution interval of each component of the feature vector.
Step five: and (4) when the color sorter works, processing image data of the materials to be sorted, comparing the processed image data with the whole HSI component and the local graph characteristic component distribution interval of the materials determined in the step four, marking the materials to be sorted which exceed the distribution interval of the corresponding component as unqualified materials, and starting the air injection valve to realize material separation.
The invention has the beneficial effects that:
the material characteristic extraction method can comprehensively analyze the overall and local characteristics of the materials and better describe the material attributes, so that the color selector can distinguish the subtle differences among the materials and the color selection result is better.
The method for selecting the characteristic gray component of the material is not unique, and a more appropriate characteristic gray component can be selected according to the actual color selection condition, so that the application range of the natural color selection method is expanded.
Drawings
FIG. 1 is a flow chart of an implementation of enhanced feature recognition of the present invention;
fig. 2 is a schematic diagram of material partial feature division.
Detailed Description
The present invention will be described in detail and clearly with reference to the following examples.
As shown in fig. 1, the embodiment of the present invention and its implementation process include the following:
the method comprises the following steps: 116 ores A to be sorted are obtained by using a color sorteri(i-1, 2.., 116) images in real-time motion, and performing motion blur image restoration, median filtering denoising, and thresholding on the images to remove the background.
Step two: extracting the single ore A to be selected one by oneiThe overall gray characteristic is the overall HSI component of the ore, and the local gray characteristic is the gray average value of R, G, B channels in each region of the inner circle of the ore. The method comprises the following specific steps:
a. single ore A without background1As a sample example, the sample example is placed in a blank picture with 256 × 256 pixels to obtain a sample example picture;
b. according to ore A1Outer contour point e of1，1，e1，2，...，e1，829Calculating the geometric center F of the object1：
Figure BDA0002628538240000051
c. Calculating ore A1All pixel points p within the contour1，1，p1，2，...，p1，39321The average RGB gray levels of (a) yield the overall characteristics of a single ore:
Figure BDA0002628538240000052
Figure BDA0002628538240000053
Figure BDA0002628538240000054
wherein p is1,j(R) is ore A1J (th) pixel point p in outer contour1，jCorresponding grey value, R, of R channel in RGB color space1Representing the gray scale feature of an R channel of the ore in an RGB color space;
gray scale feature G of G channel of ore in RGB color space1And the gray scale characteristic B of the B channel1Grayscale characterization R according to sum R channel1Processing in the same way to obtain;
d. conversion of ore A1The overall RGB information of (a) is HSI information, and H, S, I components are calculated respectively:
mixing ore A1Converting the gray characteristic of the R, G, B channel as the whole RGB information into HSI information according to the following formula, and respectively calculating H, S, I components of the HSI information as the whole HSI component;
Figure BDA0002628538240000061
H＝H+2π，ifH＜0
Figure BDA0002628538240000062
Figure BDA0002628538240000063
Max＝maX(R，G，B)
Mln＝min(R，G，B)
wherein, the gray scale characteristics of the R, G, B channels are normalized to the [0, 1] interval before the above formula is applied.
Now with ore A1The overall HSI component is calculated in detail as an example:
Figure BDA0002628538240000064
Figure BDA0002628538240000065
Figure BDA0002628538240000066
e. by inner circle D1Extraction of Ore A1Partial image of (D), inner circle1Center M of1With ore A1Geometric center F1Are superposed and have an inner circle D1Area T of1And ore A1Area S of1Satisfy T1：S11: 5, Ore A1Area S of1From the outer contour point e1，1，e1，2，...，e1，829Obtaining the total number of the pixel points in the surrounded area, and then S139321, inner circle D1Area is calculated similarly and T1＝7864；
f. Firstly, the inner circle D1Divided into concentric inner circles Q along the radius quartering1，1And three rings Q1，2、Q1，3、Q1，4While the inner circle D is drawniEight equal divisions along the circumference to the inner circle Q1，1And three rings Q1，2、Q1，3、Q1，4Further divided into eight regions, totally divided into 32 regionsA domain;
g. respectively calculating the inner circles D1The average value of the gray levels of R, G, B channels in the 32 regions of (1) is obtained as an average value of 96 gray levels, which is used as a local gray level feature.
Step three: mixing ore A1Inner circle D of1Is converted into a weighted undirected graph P1(V, E) nodes and edges; an ore AiWith a weighted undirected graph Pi(V, E), wherein V represents a node set, E represents an edge set, an ore local area gray characteristic weighting undirected graph is constructed, and each ore A is obtainediAnd feature vector KiThe method comprises the following specific steps:
h. firstly, the ore A is mixed1Inner circle D1The numbers of the 32 areas are sequentially ordered from the center to the outside; respectively calculating the geometric center of each region as the node v of the region in the weighted undirected graph1，1,v1，2,...,v1，32，v1，1Represents ore A1The corresponding 1 st area is the node in the weighted undirected graph, and all the nodes form the weighted undirected graph P1(V, E) set of points V1；
i. Combining and matching the 32 areas in pairs, and respectively calculating the color difference coefficient between the two matched areasu，vCoefficient of chromatic aberrationu，vTwo areas smaller than the color difference coefficient threshold phi are associated, and the association relation of the two areas is used as an edge Eu，vFrom all edges Eu，vForm a weighted undirected graph P1Set of edges E of (V, E)1；
Ore A1Inner circle D1Color difference coefficients of area 1 and area 2:
Figure BDA0002628538240000071
then1，2If phi is 300, areas 1 and 2 are connected, and the connection judgment calculation of other matching areas is similar;
j. from weighted undirected graph P1Distance d between nodes in (V, E)u，vAnd coefficient of chromatic aberrationu，vCalculating the edge Eu，vWeight ω of (d)u，v：
Figure BDA0002628538240000072
k. Computing a weighted undirected graph P1Graph feature components of (V, E):
extracting weighted feature path lengths
Figure BDA0002628538240000073
Global efficiency
Figure BDA0002628538240000074
And weighted global clustering coefficients
Figure BDA0002628538240000075
As graph feature components of the ore, as well as morphological features;
weighted feature path length
Figure BDA0002628538240000076
The weighted undirected graph is the average value of the lengths of all possible nodes to the shortest paths in the weighted undirected graph, the node pair is formed by any two nodes, the path length is the sum of the weights of all edges on one path between the two nodes in the node pair, and the length of the shortest path is the minimum value of the lengths of all possible paths between the two nodes in the node pair;
Figure BDA0002628538240000077
Figure BDA0002628538240000078
is the shortest path length between nodes u and v, and
Figure BDA0002628538240000079
where u, p, q.., z, v are nodes on the shortest path between nodes u and v.
Global efficiency
Figure BDA00026285382400000710
Is the average of the reciprocal of the length of the shortest path of all possible node pairs in the weighted undirected graph;
Figure BDA00026285382400000711
Figure BDA0002628538240000081
weighted global clustering coefficients
Figure BDA0002628538240000082
Defining as an average of local clustering coefficients;
Figure BDA0002628538240000083
wherein the content of the first and second substances,
Figure BDA0002628538240000084
is the sum of the geometric means of the trilateral weights of all triangles of ore 1 containing node u, and
Figure BDA0002628538240000085
is the sum of the weights of all edges at which node u of ore 1 is located, and
Figure BDA0002628538240000086
ωu，vrepresenting a triangle edge E containing a node uu，vWeight of (a), ωu，zRepresenting a triangle edge E containing a node uv，zWeight of (a), ωu，zRepresenting a triangle edge E containing a node uu，zThe weight of (c);
mixing ore A1Inner circle D of1Map feature component of
Figure BDA0002628538240000087
Bulk HSI component A1(H)、A1(S)、A1(I) And category label Y1Sequentially combining to obtain the described ore A1Feature vector of attribute features
Figure BDA0002628538240000088
Figure DA00026285382456325
Wherein the class label Y1Take 0, at this time, ore A1Qualified;
step four: using the feature vector K obtained in step threei(i ═ 1, 2.., 116) calculating distribution intervals of the overall HSI components and the map characteristic components of all ores, and specifically comprising the following steps:
for all ores AiA feature vector K of (i ═ 1, 2.., 116)iFor the feature vector KiWherein the label of class division YiAll other components obtain the upper and lower limits of the component distribution interval according to the following processing to be used as the weighted characteristic path length of the first component
Figure BDA00026285382400000810
For example, the following are explained:
m. mixing all ores AiWeighted feature path length of
Figure BDA00026285382400000811
According to the sequential arrangement of the numerical values from small to large, the quartile T1 (L) under the arrangement is calculatedω) Upper quartile T3 (L)ω) And quartile IQR (L)ω)：
IQR(Lω)＝T3(Lω)-T1(Lω)
n. according to quartile T1 (L)ω) Upper quartile T3 (L)ω) And quartile IQR (L)ω) Calculating to obtain maximum observed value
Figure BDA00026285382400000812
And minimum observed value
Figure BDA00026285382400000813
Figure BDA00026285382400000814
Figure BDA00026285382400000815
For each ore AiWeighted feature path length of
Figure BDA00026285382400000816
The weighted characteristic path length is obtained by performing judgment processing in the following way
Figure BDA00026285382400000817
Upper and lower limits corresponding to the components:
if there is at least one ore AiWeighted feature path length of
Figure BDA00026285382400000818
Satisfy the requirement of
Figure BDA00026285382400000819
If the minimum observed value is represented, the lower limit corresponding to the weighted characteristic path length component is taken
Figure BDA00026285382400000820
Otherwise, the lower limit min (L) corresponding to the weighted characteristic path length component is takenω) For all ores AiWeighted feature path length of
Figure BDA00026285382400000821
Minimum value of (1);
if there is at least one ore AiWeighted feature path length of
Figure BDA0002628538240000091
Satisfy the requirement of
Figure BDA0002628538240000092
If the maximum observed value is represented, the upper limit corresponding to the weighted characteristic path length component is taken
Figure BDA0002628538240000093
Otherwise, taking the upper limit max (L) corresponding to the weighted characteristic path length componentω) For all ores AiWeighted feature path length of
Figure BDA0002628538240000094
Maximum value of (1);
p, respectively calculating the characteristic vector K according to the same processing mode of the steps n and oiExcept for class label YiOther than the rest
Figure BDA0002628538240000095
...，Ai(I) And obtaining the distribution interval of each component of the feature vector.
Step five: and (4) when the color sorter works, processing the image data of the ore to be sorted, comparing the processed image data with the whole HSI component and the local graph characteristic component distribution interval of the ore determined in the step four, marking the ore to be sorted which exceeds the distribution interval of the corresponding component as unqualified ore, and starting the air injection valve to realize ore separation.
Therefore, the method can comprehensively analyze the overall and local characteristics of the materials and accurately obtain the material attributes, so that the color selector can distinguish the subtle differences among the materials, the color selection result is good, and the application range of the natural color selection method is expanded.
The above-described embodiments of the present invention are only preferred embodiments, and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims (9)

1. A material enhanced feature recognition and selection method of a color selector is characterized by comprising the following steps:
the method comprises the following steps: using a color selector to obtain images of N materials moving in real time as material images, and removing backgrounds in the images;
step two: extracting the integral gray feature and the local gray feature of a single material one by one from the material image, wherein the integral gray feature is the integral HSI component of the material, and the local gray feature is the gray average value of R, G, B channels in each area of the inner circle of the material;
step three: mixing the material AiInner circle D ofiIs converted into a weighted undirected graph Pi(V, E) nodes and edges, constructing a material local area gray level feature weighted undirected graph, and obtaining each material AiAnd feature vector Ki；
Step four: using the feature vector K obtained in step threeiCalculating the distribution intervals of the whole HSI component and the characteristic component of the graph of the material;
step five: and (4) when the color sorter works, processing image data of the materials to be sorted, comparing the processed image data with the whole HSI component and the local graph characteristic component distribution interval of the materials determined in the step four, marking the materials to be sorted which exceed the distribution interval of the corresponding component as unqualified materials, and starting the air injection valve to realize material separation.
2. The method for identifying and rejecting the materials of the color sorter according to claim 1, wherein: the second step comprises the following specific steps:
s2.1, removing each ith material A of the backgroundiAs a sample example, i ═ 1, 2., N, i denotes the serial number of the material, and is placed in a blank picture of m × m pixels to obtain a sample example picture;
s2.2, according to the sample, the material A in the pictureiOuter contour point e ofi，1，ei，2，...，ei，nCoordinate calculation of the geometric center Fi：
Figure FDA0002628538230000011
Wherein e isi，1，ei，2，...，ei，nRespectively represent material AiCorresponding 1 st to nth outer contour points;
s2.3, calculating a material AiAll pixel points p in the outer contouri，1，pi，2，...，pi，sThe average RGB gray level value obtains the integral gray level characteristics of the material:
Figure FDA0002628538230000012
wherein p isi，j(R) is Material AiJ (th) pixel point p in outer contouri，jCorresponding grey value, R, of R channel in RGB color spaceiExpressing the gray scale characteristics of an R channel of the material in an RGB color space;
gray scale characteristic G of G channel of material in RGB color spaceiAnd the gray scale characteristic B of the B channeliGrayscale characterization R according to sum R channeliProcessing in the same way to obtain;
s2.4, mixing the material AiThe gray feature of the R, G, B channel is taken as the whole RGB information and converted into HSI information, and then H, S, I components of the HSI information are respectively calculated as the whole HSI component;
s2.5, establishing an inner circle DiInner circle DiCenter of circle MiWith the material AiGeometric center F ofiAre superposed and have an inner circle DiArea T ofiAnd material AiArea S ofiSatisfy Ti∶Si＝1∶5；
S2.5.1, firstly, the inner circle DiDivided into concentric inner circles Q along the radius quarteringi，1And three rings Qi，2、Qi，3、Qi，4While the inner circle D is drawniEight equal divisions along the circumference to the inner circle Qi，1And three rings Qi，2、Qi，3、Qi，4The device is further divided into eight areas, and the eight areas are divided into 32 areas in total;
s2.5.2, calculating the inner circle D respectivelyiThe average value of the gray levels of R, G, B channels in the 32 regions of (1) is obtained as an average value of 96 gray levels, which is used as a local gray level feature.
3. The method for identifying and rejecting the materials of the color sorter according to claim 1, wherein: the third step comprises the following specific steps:
s3.1, firstly, mixing the material AiInner circle DiArea normalization of (2); respectively calculating the geometric center of each region as the node v of the region in the weighted undirected graphi，1,vi，2,...,vi，32，vi，1Represents Material AiThe corresponding 1 st area is the node in the weighted undirected graph, and all the nodes form the weighted undirected graph Pi(V, E) set of points Vi；
S3.2, combining and matching the 32 areas in pairs, and respectively calculating the color difference coefficient between the two matched areasu，vCoefficient of chromatic aberrationu，vTwo areas smaller than the color difference coefficient threshold phi are associated, and the association relation of the two areas is used as an edge Eu，vFrom all edges Eu，vForm a weighted undirected graph PiSet of edges E of (V, E)i；
S3.3, according to the weighted undirected graph PiDistance d between nodes in (V, E)u，vAnd coefficient of chromatic aberrationu，vCalculating the edge Eu，vWeight ω of (d)u，v：
Figure FDA0002628538230000021
S3.4, calculating weighted undirected graph PiGraph feature components of (V, E): extracting weighted feature path lengths
Figure FDA0002628538230000022
Global efficiency
Figure FDA0002628538230000023
And weighted global clustering coefficients
Figure FDA0002628538230000024
As graph feature components of the material;
s3.5, mixing the material AiInner circle D ofiMap feature component of
Figure FDA0002628538230000025
Bulk HSI component Ai(H)、Ai(S)、Ai(I) And category label YiSequentially combined to obtain a description material AiFeature vector of attribute features
Figure FDA0002628538230000026
Wherein the class label YiTaking 0 or 1, respectively representing a material AiPass or fail.
4. The method for identifying and rejecting the materials of the color sorter according to claim 1, wherein: the fourth step comprises the following specific steps:
for all materials AiCharacteristic vector K ofiFor the feature vector KiWherein the label of class division YiAll other components obtain the upper and lower limits of the component distribution interval according to the following processing to be used as the weighted characteristic path length of the first component
Figure FDA0002628538230000027
For example, the following are explained:
s4.1, mixing all the materials AiWeighted feature path length of
Figure FDA0002628538230000031
According to the numerical values which are arranged from small to large in sequence, the quartile T1 (L) under the arrangement is calculatedω) Upper quartile T3 (L)ω) And quartile IQR (L)ω)；
S4.2 according to quartile T1 (L)ω) Upper quartile T3 (L)ω) And quartile IQR (L)ω) Calculating to obtain maximum observed value
Figure FDA0002628538230000032
And minimum observed value
Figure FDA0002628538230000033
Figure FDA0002628538230000034
Figure FDA0002628538230000035
Wherein γ represents a distribution correction coefficient;
s4.3, for each material AiWeighted feature path length of
Figure FDA0002628538230000036
The weighted characteristic path length is obtained by performing judgment processing in the following way
Figure FDA0002628538230000037
Upper and lower limits corresponding to the components:
if there is at least one material AiWeighted feature path length of
Figure FDA0002628538230000038
Satisfy the requirement of
Figure FDA0002628538230000039
Figure FDA00026285382300000310
Representing the minimum observed value, then take and addLower bound for weight feature path length component
Figure FDA00026285382300000311
Otherwise, the lower limit min (L) corresponding to the weighted characteristic path length component is takenω) For all the materials AiWeighted feature path length of
Figure FDA00026285382300000312
Minimum value of (1);
if there is at least one material AiWeighted feature path length of
Figure FDA00026285382300000313
Satisfy the requirement of
Figure FDA00026285382300000314
Figure FDA00026285382300000315
If the maximum observed value is represented, the upper limit corresponding to the weighted characteristic path length component is taken
Figure FDA00026285382300000316
Otherwise, taking the upper limit max (L) corresponding to the weighted characteristic path length componentω) For all the materials AiWeighted feature path length of
Figure FDA00026285382300000317
Maximum value of (1);
s4.4, respectively calculating the eigenvector K according to the same processing mode of the two stepsiExcept for class label YiOther than the rest
Figure FDA00026285382300000318
...，Ai(I) And obtaining the distribution interval of each component of the feature vector.
5. The method for identifying and rejecting the materials in the color sorter as claimed in claim 1, wherein: in the first step, the obtained material image uses motion blur image restoration, median filtering denoising and a threshold method to remove the background in the material image.
6. The method for identifying and rejecting the materials in the color sorter as claimed in claim 1, wherein: in the second step, material AiArea S ofiFrom the outer contour point ei，1，ei，2，...，ei，nThe total number of the pixel points in the surrounded area is obtained, and the inner circle DiArea calculation method and material AiArea S ofiThe calculation is the same.
7. The method for identifying and rejecting the materials in the color sorter as claimed in claim 3, wherein: the color difference coefficient between the two regionsu，vThe calculation is as follows:
Figure FDA00026285382300000319
wherein r isuRepresents a grayscale characteristic of an R channel of the region u, Δ R represents a difference between the grayscale characteristic of the R channel of the region u and the grayscale characteristic of the R channel of the region v, Δ G represents a difference between the grayscale characteristic of a G channel of the region u and the grayscale characteristic of the G channel of the region v, and Δ B represents a difference between the grayscale characteristic of a B channel of the region u and the grayscale characteristic of the B channel of the region v.
8. The method for identifying and rejecting the materials in the color sorter as claimed in claim 3, wherein: weighted feature path length in said S3.4
Figure FDA0002628538230000041
Is the average value of the shortest path lengths of all possible node pairs in the weighted undirected graph, the node pair is composed of any two nodes, and the path lengths are two nodes in the node pairThe sum of the weights of all edges on one path among the nodes, wherein the length of the shortest path is the minimum value of the lengths of all possible paths between two nodes in the node pair; global efficiency
Figure FDA0002628538230000042
Is the average of the inverse of the shortest path length for all possible node pairs in the weighted undirected graph.
9. The method for identifying and rejecting the materials in the color sorter as claimed in claim 3, wherein: the weighted global clustering coefficients in S3.4
Figure FDA0002628538230000043
The formula is adopted for calculation;
Figure FDA0002628538230000044
wherein the content of the first and second substances,
Figure FDA0002628538230000045
is a material AiContains the sum of the geometric mean of the trilateral weights of all triangles of node u,
Figure FDA0002628538230000046
is a material AiF represents the weight sum of all edges where the node u is located, f represents the material AiThe number of nodes.
CN202010804337.3A 2020-08-12 2020-08-12 Material enhanced feature recognition and selection method of color sorter Active CN112122175B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010804337.3A CN112122175B (en) 2020-08-12 2020-08-12 Material enhanced feature recognition and selection method of color sorter

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010804337.3A CN112122175B (en) 2020-08-12 2020-08-12 Material enhanced feature recognition and selection method of color sorter

Publications (2)

Publication Number Publication Date
CN112122175A true CN112122175A (en) 2020-12-25
CN112122175B CN112122175B (en) 2021-08-10

Family

ID=73850297

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010804337.3A Active CN112122175B (en) 2020-08-12 2020-08-12 Material enhanced feature recognition and selection method of color sorter

Country Status (1)

Country Link
CN (1) CN112122175B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113414123A (en) * 2021-06-10 2021-09-21 云南永昌硅业股份有限公司 Vein quartz color sorting method and device for improving quality of industrial silicon smelting raw materials
CN113426707A (en) * 2021-06-10 2021-09-24 云南永昌硅业股份有限公司 Silicon slag color separation method and device for effectively improving silicon metal recovery rate
CN115165886A (en) * 2022-09-05 2022-10-11 淄博齐华制衣有限公司 Textile flame-retardant finishing quality detection method based on vision

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005188940A (en) * 2003-12-24 2005-07-14 Matsushita Electric Ind Co Ltd Solder inspection device and solder inspection method
CN101555661A (en) * 2009-04-14 2009-10-14 山东农业大学 Measuring method and system for weight of cotton foreign fiber based on machine vision technique
CN102654466A (en) * 2009-05-13 2012-09-05 株式会社高永科技 Method of measuring measurement target
CN102663391A (en) * 2012-02-27 2012-09-12 安科智慧城市技术（中国）有限公司 Image multifeature extraction and fusion method and system
CN103028555A (en) * 2012-12-03 2013-04-10 天津理工大学 Raisin color sorting detection method based on HSI (Hue-Saturation-Intensity) color features
CN104050674A (en) * 2014-06-27 2014-09-17 中国科学院自动化研究所 Salient region detection method and device
CN109145857A (en) * 2018-09-04 2019-01-04 深圳清沣溪科技有限公司 A method of extracting curve data from curvilinear figure
EP3432272A1 (en) * 2017-07-19 2019-01-23 AcuSolutions Inc. Pseudo h&e image producing method and optical system using same
CN110009618A (en) * 2019-04-02 2019-07-12 浙江大学 A kind of Axle Surface quality determining method and device
CN110276386A (en) * 2019-06-11 2019-09-24 济南大学 A kind of apple grading method and system based on machine vision
CN110434090A (en) * 2019-08-23 2019-11-12 中山德著智能科技有限公司 A kind of carrot surface flaw detecting method
CN110658212A (en) * 2019-09-25 2020-01-07 浙江大学 Shaft part surface full-view image acquisition device for defect identification

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005188940A (en) * 2003-12-24 2005-07-14 Matsushita Electric Ind Co Ltd Solder inspection device and solder inspection method
CN101555661A (en) * 2009-04-14 2009-10-14 山东农业大学 Measuring method and system for weight of cotton foreign fiber based on machine vision technique
CN102654466A (en) * 2009-05-13 2012-09-05 株式会社高永科技 Method of measuring measurement target
CN102663391A (en) * 2012-02-27 2012-09-12 安科智慧城市技术（中国）有限公司 Image multifeature extraction and fusion method and system
CN103028555A (en) * 2012-12-03 2013-04-10 天津理工大学 Raisin color sorting detection method based on HSI (Hue-Saturation-Intensity) color features
CN104050674A (en) * 2014-06-27 2014-09-17 中国科学院自动化研究所 Salient region detection method and device
EP3432272A1 (en) * 2017-07-19 2019-01-23 AcuSolutions Inc. Pseudo h&e image producing method and optical system using same
CN109145857A (en) * 2018-09-04 2019-01-04 深圳清沣溪科技有限公司 A method of extracting curve data from curvilinear figure
CN110009618A (en) * 2019-04-02 2019-07-12 浙江大学 A kind of Axle Surface quality determining method and device
CN110276386A (en) * 2019-06-11 2019-09-24 济南大学 A kind of apple grading method and system based on machine vision
CN110434090A (en) * 2019-08-23 2019-11-12 中山德著智能科技有限公司 A kind of carrot surface flaw detecting method
CN110658212A (en) * 2019-09-25 2020-01-07 浙江大学 Shaft part surface full-view image acquisition device for defect identification

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
冯毅雄等: "基于特征与形貌重构的轴件表面缺陷检测方法", 《浙江大学学报》 *
周友行等: "HSI颜色空间下的直线导轨表面缺陷检测方法", 《中国机械工程》 *
尹建军等: "RGB与HSI颜色空间下番茄图像分割的对比研究", 《农机化研究》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113414123A (en) * 2021-06-10 2021-09-21 云南永昌硅业股份有限公司 Vein quartz color sorting method and device for improving quality of industrial silicon smelting raw materials
CN113426707A (en) * 2021-06-10 2021-09-24 云南永昌硅业股份有限公司 Silicon slag color separation method and device for effectively improving silicon metal recovery rate
CN115165886A (en) * 2022-09-05 2022-10-11 淄博齐华制衣有限公司 Textile flame-retardant finishing quality detection method based on vision
CN115165886B (en) * 2022-09-05 2022-11-18 淄博齐华制衣有限公司 Textile flame-retardant finishing quality detection method based on vision

Also Published As

Publication number Publication date
CN112122175B (en) 2021-08-10

Similar Documents

Publication Publication Date Title
CN112122175B (en) Material enhanced feature recognition and selection method of color sorter
CN103971126B (en) A kind of traffic sign recognition method and device
CN108108761A (en) A kind of rapid transit signal lamp detection method based on depth characteristic study
CN104408449B (en) Intelligent mobile terminal scene literal processing method
CN102324036B (en) Method and device for acquiring human face skin color region from image
Du et al. Color display for hyperspectral imagery
CN109255344A (en) A kind of digital display instrument positioning and Recognition of Reading method based on machine vision
CN107292307B (en) Automatic identification method and system for inverted Chinese character verification code
Prasetyo et al. Mango leaf image segmentation on HSV and YCbCr color spaces using Otsu thresholding
CN106295656A (en) Image outline characteristic extraction method based on image color lump content and device
CN109740572A (en) A kind of human face in-vivo detection method based on partial color textural characteristics
CN105678318B (en) The matching process and device of traffic sign
CN111428556A (en) Traffic sign recognition method based on capsule neural network
CN110866879A (en) Image rain removing method based on multi-density rain print perception
CN108629286A (en) A kind of remote sensing airport target detection method based on the notable model of subjective perception
CN108181316A (en) A kind of bamboo strip defect detection method based on machine vision
CN107169508A (en) A kind of cheongsam Image emotional semantic method for recognizing semantics based on fusion feature
CN103544716B (en) A kind of pixel to image carries out method and the device of color classification
CN112906550A (en) Static gesture recognition method based on watershed transformation
CN114708272A (en) Garment image segmentation model establishing method and garment image segmentation method
Lionnie et al. A comparison of human skin color detection for biometrie identification
Ganesan et al. A comprehensive review of the impact of color space on image segmentation
CN107038401B (en) Lip contour segmentation and feature extraction method
Islami Implementation of HSV-based Thresholding Method for Iris Detection
Moertini et al. Algorithms of clustering and classifying batik images based on color, contrast and motif

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant