CN109871900A - The recognition positioning method of apple under a kind of complex background based on image procossing - Google Patents
The recognition positioning method of apple under a kind of complex background based on image procossing Download PDFInfo
- Publication number
- CN109871900A CN109871900A CN201910168484.3A CN201910168484A CN109871900A CN 109871900 A CN109871900 A CN 109871900A CN 201910168484 A CN201910168484 A CN 201910168484A CN 109871900 A CN109871900 A CN 109871900A
- Authority
- CN
- China
- Prior art keywords
- image
- apple
- circle
- value
- density
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of apple identification localization methods, it mainly include Apple image information collection, image preprocessing, image characteristics extraction and apple identification segmentation, Morphological scale-space determines five, each apple center etc. aspect in bianry image, eliminates the complex background in image in addition to apple in conjunction with clustering procedure by statistical picture feature, the center location and fruit size that further determined apple in image, improve Apple image recognition rate and accuracy.The influence that its illumination is removed using Retinex algorithm enhancing original image, provides convenience for further image procossing.
Description
Technical field
The present invention relates to a kind of identifications of apple under field of image processing more particularly to complex background based on image procossing
Localization method.
Background technique
China is maximum apple production state in the world, apple total output in 2010 far super European, America, Africa and big
The sum of foreign continent.With the continuous adjustment of China's structure of agricultural production and the successive optimization of production distribution, the growing environment of apple is able to
Improve, productivity effect is greatly improved, but compared with international most advanced level, there are still obvious gaps.
In apple production operation, it is in production investment link that the picking of fruit, which accounts for the 40%~50% of entire workload,
Period needs are most short, labour's investment most concentration, a link that is most laborious, most cumbersome and having certain danger, and adopt
The quality for plucking operation quality directly influences the storage of apple, processing and sale, to finally influence its market price and economy
Benefit, therefore realize that the picking automatically of fruit is imperative, and its key is how to utilize machine vision technique realization fruit
Accurately identify and be accurately positioned.Nowadays, for the fruit object identification and positioning under blocking, the country yet there are no more perfect
Research achievement generally require to choose a large amount of representative candidate points for the target for reaching high resolution and precise positioning,
But since system processes data amount is big, working efficiency is low, it is not able to satisfy the demand positioned in real time.
Under natural scene, the growth posture of apple is influenced by natural conditions such as soil, season, weather and there are larger
Difference, fruit is more universal by branch eclipse phenomena.Existing conventional algorithm usually can not be good in identification process
The influence for removing illumination, it is not very accurate and inefficient for leading to fruit identification, and because the originals such as branches and leaves block, fruit is overlapped
Because seriously affecting the accurately identifying of fruit, the execution of the accurate positioning of picking point and Picking behavior.
Summary of the invention
The object of the present invention is to provide a kind of apple identification localization methods, mainly include Apple image information collection, figure
As pretreatment, image characteristics extraction and apple identification segmentation, Morphological scale-space determine five, each apple center etc. in bianry image
Aspect eliminates the complex background in image in addition to apple by statistical picture feature in conjunction with clustering procedure, preferable to carry out
The reduction of Apple image and the center location and size that further determined apple in image, improve Apple image recognition rate
And accuracy.The influence that wherein illumination is removed using Retinex algorithm enhancing original image, is mentioned for further image procossing
For convenience.
The present invention realizes with the following method: the identification positioning side of apple under a kind of complex background based on image procossing
Method, comprising:
Step 1: carrying out information collection to original image；
Step 2: being pre-processed to original image information collected；
Step 3: being extracted to primitive image features and carrying out identification segmentation to Apple image；Specially extract RGB figure
As color characteristic and textural characteristics, and clustering method integration various features value is used, identifies and divide Apple image；
Step 4: median filtering denoising is carried out to bianry image and the bianry image after denoising successively carried out opening operation and
Closed operation；
Step 5: determining each apple center in bianry image, Apple image is carried out circle contour fitting operation and judging quasi-
Whether the image after conjunction has overlapping, and intersecting lens of being subject to if having judges area, otherwise Apple image after digital simulation, to obtain
Obtain Apple location information；
In step 2, the pretreatment of image removes the influence of illumination using Retinex algorithm enhancing image, it is assumed that original graph
As S is the product of light image L and albedo image R, formula is as follows:
S (x, y)=R (x, y) L (x, y) (1)
Illumination L is estimated from original image S, to decomposite R, eliminates the influence of uneven illumination, in processes, usually
Image is gone into log-domain, i.e. s=logS, l=logL, r=logR, so that multiplication relationship to be converted to the relationship of sum, formula
It is as follows:
Log (s)=log (RL)
LogS=logR+logL
S=l+r (2)
In step 3, the extraction image texture characteristic specifically: texture analysis is carried out using the gray level image in the channel R；
Describe son to the texture based on regional luminance histogram to calculate, respectively
Mean value
Standard deviation
Smoothness
Consistency
Entropy
Z in formulai--- the stochastic variable of brightness
p(zi) --- the histogram in a region
L --- possible number of greyscale levels
It include that fruit, sky, limb and 4 part of leaf form, after samples normalization, by mean value and side in grayscale image
Difference is used as evaluation parameter, for investigating the diversity of background classes and fruit class；The investigation background classes include limb, leaf, day
It is empty；The difference of mean value represents the distance between two classes；The degree of scatter of variance representative sample；Defined feature evaluation of estimate (EV) description is various
The calculating formula of feature is
AV in formula --- the variance of fruit class
The variance of BV --- background classes
MD --- the difference of fruit class and background classes mean value
According to the EV value of the fruit of estimation and background, mean value m is selected, standard deviation and entropy e totally 3 textural characteristics.
Further, described that identification segmentation is carried out to Apple image specifically: to be based on density using k-medoids is improved
Thought cluster algorithm the feature of extraction is handled, all object of classification are divided into 4 classes: be respectively day empty class, fruit class,
Leaf class and limb class；The gray average mavr for counting every one kind sorts all kinds of gray averages successively in descending order are as follows: sky,
Pixel value in the corresponding window of fruit class is converted to 1 by fruit, leaf and limb, remaining is zero as background transitions, is obtained
Apple bianry image after segmentation；Clustering algorithm based on density thought is specifically expressed as follows:
The k of the standoff distance more than certain threshold range point centered on the points of high-density region is taken, then sample x
Density are as follows:
Density (x)=p ∈ C | dist (x, p)≤r } (9)
Dist is certain away from amount of height, and r is radius, and C is sample set, and formula (9) is expressed as the point centered on x, and r is radius
The sample number that the sphere of composition is included；Wherein r is set as r=u* θ, and θ is that user gives constant, u be two-by-two data object away from
From mean value, can state are as follows:
The averag density of sample are as follows:
Wherein, density initialization process includes:
(1) after the density of each sample calculates, the sample that density is greater than averag density is put in set s, s statement are as follows:
Take dots of maximum density as first cluster centre Z in S1, and deleted from S:
(2) high density o'clock for taking distance Z farthest is as second cluster centre Z2, and deleted from S:
(3)Z3It is chosen according to formula (5), and is deleted from s:
(4) according to formula (6) until finding k central point:
The value of variable i is 1,2 in formula (12)~(16) ..., n.
Further, each apple center further includes first using before identification to threshold region in the determining bianry image
Boundary rectangle label, only calculates the effective coverage of label, after the completion of identification, to the circle contour and threshold value fitted point
The profile cut compares, and retains the circle that pixel registration is greater than 40%, if Euclidean distance between the center of circle is less than Δ d and corresponding
The difference of radius is less than Δ r, then is attributed to same class, assert that it corresponds to the same apple profile, to the center of circle and half in same class
Diameter is averaged；Realize that transformation algorithm specific steps include:
(1) edge detection is carried out to the fruit image of input and calculates the gradient of figure, finally determine circumference；
(2) value of the gradient straight line for drawing out all figures, the sum that adds up in certain coordinate points is bigger, illustrates straight at that point
The number of line intersection is more, is also more likely to be the center of circle；
(3) accumulation threshold T is set, if accumulative frequency is greater than the threshold value, the as center of circle；
(4) distance of some center of circle to all circumferences, as possible radius value are calculated；
(5) maximum radius and least radius are set, the distance value in this section is retained；
(6) statistics with histogram is carried out to the distance remained；
(7) threshold k is set, when the number that distance value occurs is greater than K value, that is, regards as radius value；
(8) (4)~(7) are repeated to remaining center of circle and obtains all radiuses；
(9) to all circle contour pixels and threshold value；
The contour pixel of segmentation compares, and rejects circle of the contour convergence degree less than 40%；The remaining center of circle is sorted out, meter
The mean circle-center and mean radius for calculating every one kind, the apple excircle configuration as fitted；Apple circle is obtained by algorithm above
Heart coordinate and radius further determine that apple position in the picture and size；
For the Apple image for there are branches and leaves to block, combined using clustering procedure and fitting excircle configuration method；For what is identified
Apple is overlapped image, first judges two fitting apple excircle configuration intersection locations, then calculates separately area as cut-off rule,
The biggish outer circle of area is taken to be retained.
Further, the texture parameter of the textural characteristics includes: intensity, density, direction and the roughness of texture.
In conclusion this apple identification location algorithm takes pictures to the apple tree under natural environment by camera to realize number
According to the acquisition of image, operation and analysis are carried out to the collected data image of camera, the influence of natural cause is removed, in complexity
Apple identification is come out in background.This apple identification location algorithm mainly includes Apple image information collection, image preprocessing, figure
As feature extraction and apple identification segmentation, Morphological scale-space determines five, each apple center etc. aspect in bianry image.Wherein,
Apple image information collection is the basis of this algorithm, is mainly used for acquiring RGB Apple image；The pretreatment of image is using algorithm
It removes the influence of illumination in image and image is enhanced；Image characteristics extraction and apple identification segmentation are using clustering algorithm
It identifies acquired image feature (color characteristic and textural characteristics), by calculating acquisition apple bianry image and removing remaining back
Scape image；Morphological scale-space is to be denoised first by median filtering, then passes through opening operation (first corrode and expand afterwards) and closes
Operation (first expanding post-etching) eliminates the cavity in image and carries out edge-smoothing so as to the progress of next step；Determine bianry image
In each apple center be to be handled by circle contour fitting process apple bianry image, it is quasi- by calculating for there is overlapping phenomenon
The area other than crosspoint is calculated after conjunction and takes the greater to be judged, finally obtains the location information of apple in the picture.
It has the beneficial effect that apple identification localization method proposed by the present invention, is improved in detection mode using k-medoids
Based on density thought clustering algorithm detection scheme and combine digital image processing techniques complete apple recognition detection, pass through system
Meter characteristics of image eliminates the complex background (branch, leaf, sky) in image in addition to apple in conjunction with clustering procedure, preferable to carry out
The reduction of Apple image and the center location and size that further determined apple in image, improve Apple image recognition rate
And accuracy.The innovative influence that wherein illumination is removed using Retinex algorithm enhancing original image, is following image
Processing provides convenience.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, the accompanying drawings in the following description is only some embodiments recorded in the present invention, for this field
For those of ordinary skill, without creative efforts, it is also possible to obtain other drawings based on these drawings.
Fig. 1 is the recognition positioning method embodiment process of apple under a kind of complex background of image procossing provided by the invention
Figure；
Fig. 2 is illumination schematic diagram provided by the invention；
Fig. 3 is clustering recognition flow diagram provided by the invention；
Fig. 4 is Morphological scale-space flow chart provided by the invention.
Specific embodiment
The present invention gives a kind of recognition positioning method embodiment of apple under complex background based on image procossing, in order to
So that those skilled in the art is more fully understood the technical solution in the embodiment of the present invention, and makes above-mentioned purpose of the invention, spy
Advantage of seeking peace can be more obvious and easy to understand, is described in further detail with reference to the accompanying drawing to technical solution in the present invention:
Present invention firstly provides a kind of recognition positioning method embodiment of apple under complex background based on image procossing,
As shown in Figure 1, comprising: the present invention realizes with the following method: the knowledge of apple under a kind of complex background based on image procossing
Other localization method, comprising:
Step 1 S101, information collection is carried out to Apple image；
Step 2 S102, image enhancement pretreatment is carried out using Retinex algorithm to Apple image information collected；
Step 3 S103, Apple image feature is extracted and identification segmentation is carried out to Apple image；Specially extract
RGB image color characteristic and textural characteristics, and characteristic value is calculated using clustering method, it identifies and divides Apple image；
Step 4 S104, median filtering denoising is carried out to bianry image and the bianry image after denoising is successively carried out out to fortune
Calculation and closed operation；
Step 5 S105, it determines each apple center in bianry image, Apple image is subjected to contour fitting with circle and judges
Whether the image after fitting has overlapping, judges area if being then subject to intersecting lens, otherwise Apple image after digital simulation, thus
Obtain Apple location information；
In step 2, the pretreatment of image removes the influence of illumination using Retinex algorithm enhancing image, it is assumed that is original
Image S is the product of light image L and albedo image R, and formula is as follows:
S (x, y)=R (x, y) L (x, y) (1)
Illumination L is estimated from original image S, to decomposite R, eliminates the influence of uneven illumination, in processes, usually
Image is gone into log-domain, i.e. s=logS, l=logL, r=logR, so that multiplication relationship to be converted to the relationship of sum, is illustrated
The following formula of Fig. 2 is as follows:
Log (s)=log (RL)
LogS=logR+logL
S=l+r (2)
In step 3, the extraction image texture characteristic specifically: texture analysis is carried out using the gray level image in the channel R；
Describe son to the texture based on regional luminance histogram to calculate, respectively
Mean value
Standard deviation
Smoothness
Consistency
Entropy
Z in formulai--- the stochastic variable of brightness
p(zi) --- the histogram in a region
L --- possible number of greyscale levels
It include that fruit, sky, limb and 4 part of leaf form, after samples normalization, by mean value and side in grayscale image
Difference is used as evaluation parameter, for investigating the diversity of background classes and fruit class；The investigation background classes include limb, leaf, day
It is empty；The difference of mean value represents the distance between two classes；The degree of scatter of variance representative sample；Defined feature evaluation of estimate (EV) description is various
The calculating formula of feature is
AV in formula --- the variance of fruit class
The variance of BV --- background classes
MD --- the difference of fruit class and background classes mean value
According to the EV value of the fruit of estimation and background, mean value m is selected, standard deviation and entropy e totally 3 textural characteristics.
Preferably, described that identification segmentation is carried out to Apple image specifically: to be thought using k-medoids is improved based on density
Think that the algorithm of cluster handles the feature of extraction, all object of classification are divided into 4 classes: being respectively day empty class, fruit class, tree
Leaf class and limb class；The gray average mavr for counting every one kind sorts all kinds of gray averages successively are as follows: sky, fruit in descending order
Real, leaf and limb, are converted to 1 for pixel value in the corresponding window of fruit class, remaining is zero as background transitions, is divided
Apple bianry image after cutting；As shown in figure 3, the clustering algorithm based on density thought is specifically expressed as follows:
The k of the standoff distance more than certain threshold range point centered on the points of high-density region is taken, then sample x
Density are as follows:
Density (x)=p ∈ C | dist (x, p)≤r } (9)
Dist is certain away from amount of height, and r is radius, and C is sample set, and formula (9) is expressed as the point centered on x, and r is radius
The sample number that the sphere of composition is included；Wherein r is set as r=u* θ, and θ is that user gives constant, u be two-by-two data object away from
From mean value, can state are as follows:
The averag density of sample are as follows:
Wherein, density initialization process includes:
(1) after the density of each sample calculates, the sample that density is greater than averag density is put in set s, s statement are as follows:
Take dots of maximum density as first cluster centre Z in S1, and deleted from S:
(2) high density o'clock for taking distance Z farthest is as second cluster centre Z2, and deleted from S:
(3)Z3It is chosen according to formula (5), and is deleted from s:
(4) according to formula (6) until finding k central point:
The value of variable i is 1,2 in formula (12)~(16) ..., n.
Wherein, image denoising uses median filtering method, and median filtering is the non-linear processing methods for inhibiting noise.Use intermediate value
Filter method handles the local image of 3*3 pixel, after 9 gray values are sorted by sequence from small to large, with the 5th (i.e. central)
Gray value of the gray value of serial number as object pixel.
The principle of median filtering method is to be ranked up the variate-value in several collected periods, then takes and sequences sequence
Worth centre value, this method can effectively prevent by sudden impulse disturbances data entrance.It is actually using
When, the quantity in the period of sequence will select suitably: if the quantity of selection is too small, may not have the effect of removal interference；Such as
The quantity of fruit selection is excessive, and the time delay that will cause sampled data is excessive, causes poor system performance.Certainly in actual use,
It is not possible that a kind of method is only used only, but the various digital filtering techniques of integrated use, for example be added and put down in median filtering method
Mean filter, so as to improving the performance of filtering.
For given n numerical value al, a2, one, an), by their ordered arrangements by size.When n is odd number, it is located at
That numerical value in middle position is known as the intermediate value of this n numerical value；When n is even number, centrally located two values are put down
Mean value is known as the intermediate value of this n numerical value, is denoted as med (al, a2, an).Median filtering is exactly such a transformation, in image
The output of certain pixel is equal to the intermediate value of each pixel grey scale in pixel neighbour city after filtering.The size of neighborhood is determined in how many a numerical value
In seek intermediate value, the shape decision of window takes element to calculate intermediate value in which type of geometric space.To two dimensional image, the shape of window
Shape can be rectangle, circle and cross etc., its center is normally on point processed.Window size and shape are sometimes to filter
Wave influential effect is very big.
Bianry image after denoising is successively carried out to opening operation and first corrodes the operation mode expanded afterwards and closed operation i.e. first
Expand the operation mode of post-etching.Wisp can be eliminated using opening operation, the separating objects at very thin point, smooth larger object
Boundary；Closed operation can fill tiny cavity, connect approaching object, smoothly its boundary, process are as shown in Figure 4:
Preferably, each apple center further includes first using to threshold region outer before identification in the determining bianry image
Rectangle marked is connect, only the effective coverage of label is calculated, after the completion of identification, to the circle contour and Threshold segmentation fitted
Profile compare, retain the circle that pixel registration is greater than 40%, if Euclidean distance between the center of circle is less than Δ d and corresponding half
The difference of diameter be less than Δ r, then be attributed to same class, assert its correspond to the same apple profile, in same class the center of circle and radius
It averages；Realize that transformation algorithm specific steps include:
(1) edge detection is carried out to the fruit image of input and calculates the gradient of figure, finally determine circumference；
(2) value of the gradient straight line for drawing out all figures, the sum that adds up in certain coordinate points is bigger, illustrates straight at that point
The number of line intersection is more, is also more likely to be the center of circle；
(3) accumulation threshold T is set, if accumulative frequency is greater than the threshold value, the as center of circle；
(4) distance of some center of circle to all circumferences, as possible radius value are calculated；
(5) maximum radius and least radius are set, the distance value in this section is retained；
(6) statistics with histogram is carried out to the distance remained；
(7) threshold k is set, when the number that distance value occurs is greater than K value, that is, regards as radius value；
(8) (4)~(7) are repeated to remaining center of circle and obtains all radiuses；
(9) to all circle contour pixels and threshold value；
The contour pixel of segmentation compares, and rejects circle of the contour convergence degree less than 40%；The remaining center of circle is sorted out, meter
The mean circle-center and mean radius for calculating every one kind, the apple excircle configuration as fitted；Apple circle is obtained by algorithm above
Heart coordinate and radius further determine that apple position in the picture and size；
For the Apple image for there are branches and leaves to block, combined using clustering procedure and fitting excircle configuration method；For what is identified
Apple is overlapped image, first judges two fitting apple excircle configuration intersection locations, then calculates separately area as cut-off rule,
The biggish outer circle of area is taken to be retained.
Preferably, the texture parameter of the textural characteristics includes: intensity, density, direction and the roughness of texture.
To sum up, set forth herein a kind of apple identification location algorithm by camera to the apple tree under natural environment take pictures come
The acquisition for realizing data image carries out operation and analysis to the collected data image of camera, removes the influence of natural cause,
Apple identification is come out in complex background.This apple identification location algorithm mainly includes Apple image information collection, and image is pre-
Processing, image characteristics extraction and apple identification segmentation, Morphological scale-space determine five, each apple center etc. side in bianry image
Face.Wherein, Apple image information collection is the basis of this algorithm, is mainly used for acquiring RGB Apple image；The pretreatment of image is
The influence of illumination in image is removed using algorithm and image is enhanced；Image characteristics extraction and apple identification segmentation be using
Clustering algorithm identifies acquired image feature (color characteristic and textural characteristics), by calculating acquisition apple bianry image and going
Fall remaining background image；Morphological scale-space is to be denoised first by median filtering, and it is (swollen after first corroding then to pass through opening operation
It is swollen) cavity in image is eliminated with closed operation (first expanding post-etching) and carries out edge-smoothing so as to the progress of next step；It determines
Each apple center is to be handled by circle contour fitting process apple bianry image in bianry image, logical for there is overlapping phenomenon
It crosses the area calculated other than crosspoint after digital simulation and takes the greater to be judged, finally obtain the position of apple in the picture
Information.
Above embodiments are to illustrative and not limiting technical solution of the present invention.Appointing for spirit and scope of the invention is not departed from
What modification or part replacement, are intended to be within the scope of the claims of the invention.
Claims (4)
1. the recognition positioning method of apple under a kind of complex background based on image procossing characterized by comprising
Step 1: carrying out information collection to original image；
Step 2: being pre-processed to original image information collected；
Step 3: being extracted to primitive image features and carrying out identification segmentation to Apple image；Specially extract RGB image face
Color characteristic and textural characteristics, and clustering method integration various features value is used, it identifies and divides Apple image；
Step 4: carrying out median filtering denoising to bianry image and the bianry image after denoising successively being carried out to opening operation and closes fortune
It calculates；
Step 5: each apple center in bianry image is determined, after carrying out circle contour fitting operation by Apple image and judge fitting
Image whether have overlapping, intersecting lens of being subject to if having judges area, otherwise Apple image after digital simulation, to obtain apple
Fruits location information；
In step 2, the pretreatment of image removes the influence of illumination using Retinex algorithm enhancing image, it is assumed that is original image
S is the product of light image L and albedo image R, and formula is as follows:
S (x, y)=R (x, y) L (x, y) (1)
Illumination L is estimated from original image S, to decomposite R, eliminates the influence of uneven illumination, it in processes, usually will figure
As going to log-domain, i.e. s=logS, l=logL, r=logR, so that multiplication relationship to be converted to the relationship of sum, formula is as follows:
Log (s)=log (RL)
LogS=logR+logL
S=l+r (2)
In step 3, the extraction image texture characteristic specifically: texture analysis is carried out using the gray level image in the channel R；To base
Describe son in the texture of regional luminance histogram to be calculated, respectively
Mean value
Standard deviation
Smoothness
Consistency
Entropy
Z in formulai--- the stochastic variable of brightness
p(zi) --- the histogram in a region
L --- possible number of greyscale levels
In grayscale image, includes that fruit, sky, limb and 4 part of leaf form, after samples normalization, mean value and variance are made
For evaluation parameter, for investigating the diversity of background classes and fruit class；The investigation background classes include limb, leaf, sky；?
The difference of value represents the distance between two classes；The degree of scatter of variance representative sample；Defined feature evaluation of estimate (EV) describes various features
Calculating formula be
AV in formula --- the variance of fruit class
The variance of BV --- background classes
MD --- the difference of fruit class and background classes mean value
According to the EV value of the fruit of estimation and background, mean value m is selected, standard deviation and entropy e totally 3 textural characteristics.
2. the recognition positioning method of apple, feature under a kind of complex background based on image procossing as described in claim 1
It is, it is described that identification segmentation is carried out to Apple image specifically: to use the improved calculation based on density thought cluster of k-medoids
Method handles the feature of extraction, and all object of classification are divided into 4 classes: being respectively day empty class, fruit class, leaf class and limb
Class；The gray average mavr for counting every one kind sorts all kinds of gray averages successively are as follows: sky, fruit, leaf and branch in descending order
It is dry, pixel value in the corresponding window of fruit class is converted to 1, remaining is zero as background transitions, the apple after being divided
Bianry image；Clustering algorithm based on density thought is specifically expressed as follows:
The k of the standoff distance more than certain threshold range point centered on the points of high-density region is taken, then sample x's is close
Degree are as follows:
Density (x)=p ∈ C | dist (x, p)≤r } (9)
Dist is certain away from amount of height, and r is radius, and C is sample set, and formula (9) is expressed as the point centered on x, and r is radius composition
The sphere sample number that is included；Wherein r is set as r=u* θ, and θ is that user gives constant, and u is data object distance two-by-two
Mean value can be stated are as follows:
The averag density of sample are as follows:
Wherein, density initialization process includes:
(1) after the density of each sample calculates, the sample that density is greater than averag density is put in set s, s statement are as follows:
Take dots of maximum density as first cluster centre Z in S1, and deleted from S:
(2) high density o'clock for taking distance Z farthest is as second cluster centre Z2, and deleted from S:
(3)Z3It is chosen according to formula (5), and is deleted from s:
(4) according to formula (6) until finding k central point:
The value of variable i is 1,2 in formula (12)~(16) ..., n.
3. the recognition positioning method of apple, special under a kind of complex background based on image procossing as claimed in claim 1 or 2
Sign is that each apple center further includes that boundary rectangle mark is first used to threshold region before identification in the determining bianry image
Note, only calculates the effective coverage of label, after the completion of identification, to the profile of the circle contour and Threshold segmentation that fit into
Row comparison retains the circle that pixel registration is greater than 40%, if the Euclidean distance between the center of circle is less than Δ d and the difference of corresponding radius is small
In Δ r, be then attributed to same class, assert its correspond to the same apple profile, in same class the center of circle and radius average；
Realize that transformation algorithm specific steps include:
(1) edge detection is carried out to the fruit image of input and calculates the gradient of figure, finally determine circumference；
(2) value of the gradient straight line for drawing out all figures, the sum that adds up in certain coordinate points is bigger, illustrates straight line phase at that point
The number of friendship is more, is also more likely to be the center of circle；
(3) accumulation threshold T is set, if accumulative frequency is greater than the threshold value, the as center of circle；
(4) distance of some center of circle to all circumferences, as possible radius value are calculated；
(5) maximum radius and least radius are set, the distance value in this section is retained；
(6) statistics with histogram is carried out to the distance remained；
(7) threshold k is set, when the number that distance value occurs is greater than K value, that is, regards as radius value；
(8) (4)~(7) are repeated to remaining center of circle and obtains all radiuses；
(9) to all circle contour pixels and threshold value；
The contour pixel of segmentation compares, and rejects circle of the contour convergence degree less than 40%；The remaining center of circle is sorted out, is calculated every
A kind of mean circle-center and mean radius, the apple excircle configuration as fitted；The apple center of circle is obtained by algorithm above to sit
Mark and radius further determine that apple position in the picture and size；
For the Apple image for there are branches and leaves to block, combined using clustering procedure and fitting excircle configuration method；For the apple identified
It is overlapped image, two fitting apple excircle configuration intersection locations is first judged, then calculates separately area as cut-off rule, take face
The biggish outer circle of product is retained.
4. the recognition positioning method of apple, special under a kind of complex background based on image procossing as claimed in claim 1 or 2
Sign is that the texture parameter of the textural characteristics includes: intensity, density, direction and the roughness of texture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910168484.3A CN109871900A (en) | 2019-03-06 | 2019-03-06 | The recognition positioning method of apple under a kind of complex background based on image procossing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910168484.3A CN109871900A (en) | 2019-03-06 | 2019-03-06 | The recognition positioning method of apple under a kind of complex background based on image procossing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109871900A true CN109871900A (en) | 2019-06-11 |
Family
ID=66919944
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910168484.3A Pending CN109871900A (en) | 2019-03-06 | 2019-03-06 | The recognition positioning method of apple under a kind of complex background based on image procossing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109871900A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110751687A (en) * | 2019-09-17 | 2020-02-04 | 山东科技大学 | Apple size grading method based on computer vision minimum and maximum circles |
| CN111160180A (en) * | 2019-12-16 | 2020-05-15 | 浙江工业大学 | Night green apple identification method of apple picking robot |
| CN111598001A (en) * | 2020-05-18 | 2020-08-28 | 哈尔滨理工大学 | Apple tree pest and disease identification method based on image processing |
| CN111738271A (en) * | 2020-03-04 | 2020-10-02 | 沈阳工业大学 | Method for identifying shielded fruits in natural environment |
| CN111833435A (en) * | 2020-06-28 | 2020-10-27 | 江苏大学 | Unmanned aerial vehicle near-field remote sensing mature crop density high-flux measurement method |
| CN112418043A (en) * | 2020-11-16 | 2021-02-26 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951905A (en) * | 2017-02-16 | 2017-07-14 | 江苏大学 | Apple identification and localization method on a kind of tree based on TOF camera |
-
2019
- 2019-03-06 CN CN201910168484.3A patent/CN109871900A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951905A (en) * | 2017-02-16 | 2017-07-14 | 江苏大学 | Apple identification and localization method on a kind of tree based on TOF camera |
Non-Patent Citations (3)
| Title |
|---|
| 司永胜等: "基于K-均值聚类的绿色苹果识别技术", 《农业机械学报》 * |
| 姚丽娟等: "一种新的k-medoids聚类算法", 《计算机工程与应用》 * |
| 许鹏等: "基于ARM的hough圆优化的苹果识别算法", 《湖北工业大学学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110751687A (en) * | 2019-09-17 | 2020-02-04 | 山东科技大学 | Apple size grading method based on computer vision minimum and maximum circles |
| CN110751687B (en) * | 2019-09-17 | 2023-05-26 | 山东科技大学 | Apple size grading method based on computer vision minimum and maximum circle |
| CN111160180A (en) * | 2019-12-16 | 2020-05-15 | 浙江工业大学 | Night green apple identification method of apple picking robot |
| CN111738271A (en) * | 2020-03-04 | 2020-10-02 | 沈阳工业大学 | Method for identifying shielded fruits in natural environment |
| CN111738271B (en) * | 2020-03-04 | 2023-05-02 | 沈阳工业大学 | Method for identifying blocked fruits in natural environment |
| CN111598001A (en) * | 2020-05-18 | 2020-08-28 | 哈尔滨理工大学 | Apple tree pest and disease identification method based on image processing |
| CN111833435A (en) * | 2020-06-28 | 2020-10-27 | 江苏大学 | Unmanned aerial vehicle near-field remote sensing mature crop density high-flux measurement method |
| CN112418043A (en) * | 2020-11-16 | 2021-02-26 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
| CN112418043B (en) * | 2020-11-16 | 2022-10-28 | 安徽农业大学 | Corn weed occlusion determination method and device, robot, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109871900A (en) | The recognition positioning method of apple under a kind of complex background based on image procossing | |
| CN110120042B (en) | Crop image pest and disease damage area extraction method based on SLIC super-pixel and automatic threshold segmentation | |
| CN105718945B (en) | Apple picking robot night image recognition method based on watershed and neural network | |
| WO2022099598A1 (en) | Video dynamic target detection method based on relative statistical features of image pixels | |
| CN107862667B (en) | Urban shadow detection and removal method based on high-resolution remote sensing image | |
| CN109636784B (en) | Image saliency target detection method based on maximum neighborhood and super-pixel segmentation | |
| CN106340016B (en) | A kind of DNA quantitative analysis method based on microcytoscope image | |
| CN110610506B (en) | Image processing technology-based agaricus blazei murill fruiting body growth parameter detection method | |
| CN106127735B (en) | A kind of facilities vegetable edge clear class blade face scab dividing method and device | |
| CN103295013A (en) | Pared area based single-image shadow detection method | |
| Das Choudhury et al. | Automated stem angle determination for temporal plant phenotyping analysis | |
| CN111784764A (en) | Tea tender shoot identification and positioning algorithm | |
| CN113298777A (en) | Cotton leaf blight detection method and system based on color features and super-pixel clustering | |
| CN110310291A (en) | A kind of rice blast hierarchy system and its method | |
| Ji et al. | In-field automatic detection of maize tassels using computer vision | |
| CN115115627B (en) | Soil saline-alkali soil monitoring method based on data processing | |
| CN101526994A (en) | Fingerprint image segmentation method irrelevant to collecting device | |
| CN109584253A (en) | Oil liquid abrasive grain image partition method | |
| CN111754538A (en) | Threshold segmentation method for USB surface defect detection | |
| CN111199245A (en) | Rape pest identification method | |
| CN115731257A (en) | Leaf form information extraction method based on image | |
| CN116721121A (en) | Plant phenotype color image feature extraction method | |
| CN111753903B (en) | Soybean variety identification method based on vein topological characteristics | |
| CN115841600B (en) | Deep learning-based sweet potato appearance quality classification method | |
| Cai et al. | Novel image segmentation based on machine learning and its application to plant analysis |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190611 |
|
| WD01 | Invention patent application deemed withdrawn after publication |