CN109359531A - Fruit harvesting area automatic positioning method facing natural scene - Google Patents

Fruit harvesting area automatic positioning method facing natural scene Download PDF

Info

Publication number
CN109359531A
CN109359531A CN201811062603.9A CN201811062603A CN109359531A CN 109359531 A CN109359531 A CN 109359531A CN 201811062603 A CN201811062603 A CN 201811062603A CN 109359531 A CN109359531 A CN 109359531A
Authority
CN
China
Prior art keywords
fruit
neighborhood
pixel
extract
area
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
CN201811062603.9A
Other languages
Chinese (zh)
Other versions
CN109359531B (en
Inventor
庄家俊
唐宇
骆少明
侯超钧
郭琪伟
苗爱敏
陈亚勇
张恒涛
刘泽锋
孙胜
朱耀宗
高升杰
程至尚
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.)
Zhongkai University of Agriculture and Engineering
Original Assignee
Zhongkai University of Agriculture and Engineering
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 Zhongkai University of Agriculture and Engineering filed Critical Zhongkai University of Agriculture and Engineering
Priority to CN201811062603.9A priority Critical patent/CN109359531B/en
Publication of CN109359531A publication Critical patent/CN109359531A/en
Application granted granted Critical
Publication of CN109359531B publication Critical patent/CN109359531B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/77—Retouching; Inpainting; Scratch removal
    • 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
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

本发明涉及一种面向自然场景的果实采收区域自动定位方法，包括如下步骤：提取训练数据中的荔枝果实区域，统计其加权三原色亮度分布特征作为客观亮度基准，采用迭代式Retinex算法增强加权三原色亮度分量；结合修正色差图、阈值化处理和数学形态学方法提取亮度增强后的果实区域；通过局部邻域像素的位置关系和色调信息重构HSI色彩空间中的色调分量，根据色调分布特征提取枝条区域；采用细化算法提取枝条骨架，通过角点检测和角点邻域像素的模式分布规则提取骨架上的关键角点，结合果实与枝条的相对位置约束和关键角点的空间分布特性，自动定位荔枝果实采收区域。本发明可以提高自然场景中荔枝果实采收区域自动定位的自适应性和准确性。

The invention relates to an automatic positioning method for a fruit harvesting area oriented to natural scenes, comprising the following steps: extracting a litchi fruit area in training data, calculating the brightness distribution characteristics of its weighted three primary colors as an objective brightness reference, and using an iterative Retinex algorithm to enhance the weighted three primary colors Luminance component; combined with corrected color difference map, thresholding and mathematical morphology methods to extract the fruit area after brightness enhancement; reconstruct the hue component in HSI color space through the positional relationship and hue information of local neighborhood pixels, and extract according to hue distribution characteristics Branch area; the branch skeleton is extracted by the thinning algorithm, and the key corner points on the skeleton are extracted through the corner point detection and the pattern distribution rules of the pixels in the neighborhood of the corner points. Automatically locate the lychee fruit harvesting area. The invention can improve the adaptability and accuracy of the automatic positioning of the litchi fruit picking area in the natural scene.

Description

A kind of fruit harvesting region automatic positioning method towards natural scene
Technical field
The present invention relates to IT application to agriculture, precision agriculture, machine vision and technical field of image processing, and in particular to a kind of Fruit harvesting region automatic positioning method towards natural scene.
Background technique
Modern orchard gradually appears the status that management cost is constantly incremented by, manual labor constantly successively decreases, close with labour Picking fruit mode based on collection type gradually shows trend (the Gongal A., Amatya of sustainable developability reduction S.,Karkee M.,et al,2015.Sensors and systems for fruit detection and localiza tion.Comput.Electron.Agric.116,8-19).The intelligent harvesting equipment of fruit is a kind of traditional artificial side of change The automated job of formula is equipped, and the critical components such as mobile platform, control system, the detection of woods fruit and end effector are generally included, For step by step input into the automatic recovery operations of a variety of mountain area woods fruits, the main problem that it is faced has been nature into slice at present The reliable detection with harvesting point/region accurate positionin in fruit region under part.
Fruit detection generallys use machine vision and image processing techniques with harvesting point/zone location.For natural scene Influence of the middle forest fruit regional imaging quality vulnerable to illumination condition, (Xu L.M., the Lv J.D., 2018.Recognition such as Xu method for apple fruit based on SUSAN and PCNN.Multimed.Tools Appl.77(6), 7205-7219) using the fruit image acquired in the homographic filtering method processing weaker situation of illumination.Wang etc. (Wang C.L., Lee W.S,Zou X.J.,et al,2018.Detection and counting of immature green citrus fruit based on the Local Binary Patterns(LBP)feature using illumination- Normalized images.Precision Agric.17 (6), 678-697) use the Retinex algorithm based on bilateral filtering Improve the overall brightness in fruit region in image.But the above method does not consider the INTELLIGENT IDENTIFICATION of input picture illumination property, Processing brightness is higher or the more uniform image of Luminance Distribution is easy to appear " overexposure " phenomenon or changes the original in fruit region There is hue information, subsequent fruit region detection and harvesting point/zone location accuracy may be influenced, so being not suitable for certainly Woods fruit image under dynamic processing natural scene.Zhuang etc. (Zhuang J.J., Luo S.M., Hou C.J., et al, 2018.Detection of orchard citrus fruits using a monocular machine vision- based method for automatic fruit picking applications.Comput.Electron.Agric. 152,64-73) the even problem of uneven illumination of fruit area image has been automatically processed using local block homographic filtering method, but should Method includes more adjustment parameter, may need to readjust when environmental factor significantly changes, treatment process is relatively complicated. (Wang C.L., Zou X.J., Tang Y.C., et al, the 2016.Localisation of litchi in an such as Wang unstructured environment using binocular stereo vision.Biosyst.Eng.145,39-51) In conjunction with K- means Method and the method for registering based on tag template extracts the red litchi fruits in binocular vision image sequence There is preferable detection effect in region to the lesser woods fruit image of backcolor difference.Point is harvested for the lichee under pneumatic environment Orientation problem, (Xiong J.T., He Z.L., Lin R., et al, the 2018.Visual positioning such as Xiong technology of picking robots for dynamic litchi clusters with disturbance.Co Mput.Electron.Agric.151,226-237) the tone distributed knowledge based on litchi fruits region, using clustering algorithm from Mature litchi fruits region and its mass center are extracted in the chrominance component of HSI image, through pendulum criterion from binocular vision image Harvesting point is estimated, but automatic positioning error is larger, is suitable for the wider end effector of beta pruning job area.To avoid environment Influence of the factor to lichee harvesting spot placement accuracy, Xiong etc. (Xiong J.T., Lin R., Liu Z., et al, 2018.The recognition of litchi clusters and the calculation of picking point In a nocturnal natural environment.Biosyst.Eng.166,44-57) using based on LED orientation light filling Nighttime imaging system, non-woods fruit in night scenes image and interference limb region, knot are filtered out by fuzzy C-means clustering It closes Otsu Threshold segmentation and Harris corner detection approach orients the harvesting point of litchi fruits, this method is mainly for the treatment of preceding back The more significant night woods fruit image of scape difference.
Although currently based on the litchi fruits of machine vision and image processing techniques harvesting point/region automatic positioning method Achieved certain effect, but in the continually changing natural scene of environmental factor, more reliable, more accurate harvesting point or Region automatic positioning method is harvested to explore there is still a need for further.
Summary of the invention
The present invention provides one kind towards natural field to overcome at least one defect (deficiency) described in the above-mentioned prior art The fruit harvesting region automatic positioning method of scape, it is intended to improve fruit detection and harvesting zone location based on machine vision technique The adaptivity and accuracy of system.
To achieve the purpose of the present invention, it is achieved using following technical scheme:
A kind of fruit harvesting region automatic positioning method towards natural scene, comprising the following steps:
(1) the weighting three primary colors luminance component V for extracting current RGB image I, utilizes modified R X chromaticity difference diagram (Modified Red and Green/Blue (X) Chromatic Mapping, MRXCM) method generates modified R X (Red&Green/ from I Blue) chromaticity difference diagram Cm, from CmMiddle extraction part fruit region RoIs (Regions of Interest) calculates RoIs Luminance Distribution Statistical nature, Retinex algorithm processing V is iteratively used according to statistical nature and reconstructs I, when iterative process terminates The enhanced RGB image I ' of brightness is exported by color space conversion model；
(2) modified R X chromaticity difference diagram C is therefrom extracted by MRXCM methodm', to Cm' execution imagethresholding obtains Cm＂, and C is handled using mathematical Morphology Algorithmm＂ therefrom extracts potential mature or immature fruit region Of；
(3) the chrominance component H ' for extracting I ', measures the positional relationship of local neighborhood pixel and the centre of neighbourhood, and it is adjacent to measure part The hue information difference of domain pixel and the centre of neighbourhood establishes the weight distribution model of local neighborhood centering heart pixel tonal value, leads to It crosses and the centre of neighbourhood pixel by noise pollution is isolated, reconstructs chrominance component H ' in a manner of weighted meanc, to H 'cExecute image threshold Value is divided to obtain Hc＂ therefrom extracts potential sepia branch region Os；
(4) O is extractedsSkeleton Qs', detect Qs' on angle point and extract it is relevant to branch bifurcated attribute key angle point, Utilize O under gravityfAnd Qs' relative position constraint and crucial angle point spatial characteristics, positioning fruit harvesting point distribution Section.
Further, step (1) the weighting three primary colors luminance component V for extracting current RGB image I, specifically, with becoming Amount R, G and B respectively indicate the grey scale pixel value of the red, green and blue Color Channel of I, calculate weighting three primary colors brightness point according to the following formula Measure V:
V=α R+ β G+ γ B
Further, step (1) is described utilizes modified R X chromaticity difference diagram (Modified Red and Green/Blue (X) Chromatic Mapping, MRXCM) method generates modified R X (Red&Green/Blue) chromaticity difference diagram C from Im, specifically, root The red color tone or green hue shown according to fruit region calculates the monochromatic ratio of R and G or R and B as scale factor, and weights Into conventional red green difference figure or reddish blue difference figure, modified chromaticity difference diagram C is generatedm。
Further, step (1) is described from CmMiddle extraction part fruit region RoIs (Regions of Interest), The statistical nature for calculating RoIs Luminance Distribution, specifically, to CmIt executes Threshold segmentation and erosion operation obtains bianry image, extract Local fruit region RoIs calculates the average brightness value M of mask images institute overlay area in V as mask imagesRoIs。
Further, step (1) is described iteratively uses Retinex algorithm processing V according to statistical nature and reconstructs I exports the enhanced RGB image I ' of brightness by color space conversion model when iterative process is terminated, specifically, extracting more The three primary colors weighted luminance component of a training image counts the average brightness value M in fruit regiontrain, and execute following steps:
S1. judge whether MRoIs≥Mtrain, if so then execute step S2, S3 is thened follow the steps if not；
S2. I as the enhanced RGB image I ' of brightness and is exported into I '；
S3. enhanced luminance component V ' is obtained using Retinex algorithm, passes through RGB/HSI color space conversion model The tone H and color saturation S of current RGB image are extracted, V ' is merged to generate the enhanced RGB image I ' of brightness, extracts again The average M of I 'RoIs' judges whether MRoIs' >=MtrainIf otherwise continuing to execute step S3, if continuing to execute step S4；
S4. I ' is exported.
Preferably, step (2) is described handles C using mathematical Morphology Algorithmm＂ therefrom extracts potential mature or prematurity Fruit region Of, specifically, successively using the morphological erosion operation based on coarse scale structures member, based on small scale structures member Dilation operation and holes filling handle Cm＂, using 8- connectivity criteria from treated CmPotential fruit region O is extracted in ＂f。
Further, step (3) positional relationship for measuring local neighborhood pixel and the centre of neighbourhood, specifically, leading Domain center is (xc,yc) m × n neighborhood in, coordinates computed be (xi,yi) ith pixel and centre of neighbourhood distance dis ((xi,yi),(xc,yc)), the positional relationship w of ith pixel and the centre of neighbourhood is obtained according to the following formulapi:
wpi=exp {-λPmax(dis((xi,yi),(xc,yc)))}
In formula, λpFor position amplification coefficient, i=1,2 ..., m × n.
Further, step (3) the hue information difference for measuring local neighborhood pixel and the centre of neighbourhood, specifically, In field, centre coordinate is (xc,yc) m × n neighborhood in, calculate neighborhood in coordinate be (xi,yi) ith pixel tone value hiWith neighborhood tone intermediate value hmDifference diff (hi,hm), and the neighborhood territory pixel tone value standard deviation sigma in addition to central pointc, according to The different information w of following formula calculating ith pixel and the centre of neighbourhoodHi:
In formula, λHFor tone difference amplification coefficient, i=1,2 ..., m × n.
Further, step (3) the weight distribution model for establishing local neighborhood centering heart pixel tonal value, by every From by noise pollution centre of neighbourhood pixel, chrominance component H ' is reconstructed in a manner of weighted meanc, specifically, passing through wpiAnd wHi Point multiplication operation distribution ith pixel weight wciIf centre of neighbourhood tone value hcWith hmDeviation is larger, then field center is arranged Weight be 0, be otherwise provided as 1, calculate the weighted average of each pixel tonal value in neighborhood, the weight as centre of neighbourhood pixel Structure tone value h 'c, the chrominance component H ' reconstructed after all pixels of H ' is traversed based on above-mentioned processc。
Preferably, step (3) is described to H 'cIt executes imagethresholding and obtains Hc＂ therefrom extracts potential sepia Branch region Os, specifically, according to branch hue information distribution setting segmentation threshold in multiple training images, to H 'cExecute image Threshold division obtains Hc＂, using 8- connectivity criteria from HcPotential branch region O is extracted in ＂s。
Further, step (4) the extraction OsSkeleton Qs', detect Qs' on angle point and extract with branch bifurcated category Property relevant crucial angle point, specifically, thinning algorithm is utilized to extract OsSkeleton Qs', it is extracted using Harris Corner Detection Algorithm Qs' on whole angle points utilize crotch angle point adjacent according to the bifurcated feature between fruit branches different in the case of more fruit adhesions The mode distribution rule and skeleton connectivity of domain pixel judge whether crotch angle point belongs to crucial angle point and extract crucial angle point.
Further, step (4) is described utilizes O under gravityfAnd Qs' relative position constraint and crucial angle point sky Between distribution character, position fruit harvesting point distributed area, specifically, according to fruit position of centre of gravity under gravity lower than its connect Strong connectedness feature between branch, fruit carpopodium and branch calculates carpopodium and branch join domain ArMiddle all pixels space is sat Target statistical nature, and combine the statistical nature of crucial angular coordinate extract only with OfBranch skeleton with connectivity extracts Harvesting central point of the crucial angle point as fruit on the skeleton far from carpopodium, with ArMiddle pixel and angled key space of points column coordinate Dispersion degree as deviate harvesting point regional standard it is poor, according to harvesting central point and regional standard difference position fruit harvesting point Distributed area.
Compared with prior art, the beneficial effect of technical solution of the present invention is: (1) iterative Retinex algorithm can be automatic The overall brightness distribution that input RGB image is adjusted according to the light characteristic in litchi fruits region, is improving image overall brightness The color information for remaining to maintain each region of original image when distribution, has preferable scene adaptive, is suitable for natural environment Under image preprocessing；(2) simultaneously using in the positional relationship of local neighborhood pixel and hue information reconstruct HSI color space Chrominance component can not only eliminate noise jamming, moreover it is possible to restore the practical hue information of centre of neighbourhood pixel to greatest extent, favorably In the reliability for improving branch extracted region；(3) it is distributed based on Harris Corner Detection Algorithm and the mode of angle point neighborhood territory pixel Rule Extraction key angle point, is avoided that the interference of noise angle point to a certain extent, be conducive to improve litchi fruits harvesting point or Harvest the positional accuracy in region.
Detailed description of the invention
Fig. 1 is the implementation example figure of the fruit harvesting region automatic positioning method process towards natural scene in the present embodiment.
Fig. 2 a is the image embodiment figure of a secondary illumination abundance.
Fig. 2 b is the adaption brightness reinforcing effect implementation example figure of image shown in Fig. 2 a.
Fig. 2 c is the weaker image embodiment figure of a secondary illumination.
Fig. 2 d is the adaption brightness reinforcing effect implementation example figure of image shown in Fig. 2 c.
Fig. 3 a is the litchi fruits extraction effect implementation example figure of image shown in Fig. 2 a.
Fig. 3 b is the litchi fruits extraction effect implementation example figure of image shown in Fig. 2 c.
Fig. 4 a be the centre of neighbourhood not by noise pollution when tone value reconstruction result implementation example figure.
Fig. 4 b be the centre of neighbourhood by noise pollution when tone value reconstruction result implementation example figure.
Fig. 5 a is the lichee branch extraction effect implementation example figure of image shown in Fig. 2 a.
Fig. 5 b is the lichee branch extraction effect implementation example figure of image shown in Fig. 2 c.
Fig. 6 a is that the litchi fruits of image shown in Fig. 2 a harvest point location effect example figure.
Fig. 6 b is that the litchi fruits of image shown in Fig. 2 c harvest point location effect example figure.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent actual product Size；
To those skilled in the art, it is to be understood that certain known features and its explanation, which may be omitted, in attached drawing 's.
In the description of the present invention, unless otherwise indicated, the meaning of " plurality " is two or more.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment
As shown in Figure 1, a kind of fruit harvesting region automatic positioning method towards natural scene, comprising the following steps:
(1) the weighting three primary colors luminance component V for extracting current RGB image I, utilizes modified R X chromaticity difference diagram (Modified Red and Green/Blue (X) Chromatic Mapping, MRXCM) method generates modified R X (Red&Green/ from I Blue) chromaticity difference diagram Cm, from CmMiddle extraction part fruit region RoIs (Regions of Interest) calculates RoIs Luminance Distribution Statistical nature, according to statistical nature export the enhanced RGB image I ' of brightness；
(2) modified R X chromaticity difference diagram C is therefrom extracted by MRXCM methodm', to Cm' execution imagethresholding obtains Cm＂, and C is handled using mathematical Morphology Algorithmm＂ therefrom extracts potential mature or immature fruit region Of；
(3) the chrominance component H ' for extracting I ', measures the positional relationship of local neighborhood pixel and the centre of neighbourhood, and it is adjacent to measure part The hue information difference of domain pixel and the centre of neighbourhood establishes the weight distribution model of local neighborhood centering heart pixel tonal value, leads to It crosses and the centre of neighbourhood pixel by noise pollution is isolated, reconstructs chrominance component H ' in a manner of weighted meanc, to H 'cExecute image threshold Value segmentation, therefrom extracts potential sepia branch region Os；
(4) O is extractedsSkeleton Qs', detect Qs' on angle point and extract it is relevant to branch bifurcated attribute key angle point, Utilize O under gravityfAnd Qs' relative position constraint and crucial angle point spatial characteristics, positioning fruit harvesting point distribution Section.
The imaging effect in fruit region is easy to be influenced by illumination condition in visual pattern, it is therefore necessary to examine in fruit The illumination compensation surveyed and first carry out image before harvesting point location, and the hue information in fruit region cannot occur significantly to change after compensating Become, and and brightness value higher input picture uniform for luminance information, is not then considered as illumination compensation, otherwise will appear " mistake Degree exposure " phenomenon and cause significantly changing for the original hue information in fruit region.For this purpose, the present invention passes through iterative Retinex Algorithm, it is automatic to enhance according to the Luminance Distribution feature in fruit region in input image lightness information and illumination abundance/homogeneous image The brightness of input picture.
By taking litchi fruits as an example.
When algorithm starts, if current RGB image I is input RGB image Ic.When compensating the luminance information of I, in order to the greatest extent may be used The original hue information of image can not be changed, the weighting three primary colors luminance component V of I is extracted according to formula (1).
V=α R+ β G+ γ B
Wherein, R, G and B respectively indicate the grey scale pixel value of the red, green and blue Color Channel of I, and s.t. indicates constraint condition. Using red color tone litchi fruits region detection as target in the present embodiment, and three parameters being arranged in formula (1) be respectively α= 0.6, β=0.3, γ=0.1, to protrude the grayscale information of red component in image.
In order to isolate the background area different from fruit hue information, setting red component (R) and green (G) from I Or the ratio of blue component (B), as impact factor, the factor value is higher, illustrates the difference between fruit and other background areas It is different bigger, and be weighted in conventional color difference drawing method, obtain the calculating of the modified R X chromaticity difference diagram as shown in formula (2)~(4) Model extracts chromaticity difference diagram C from Im。
Cm=λ × R-X (2)
λ=R/X (3)
In order to extract red color tone litchi fruits region, β > γ is set in the present embodiment.
C is divided using Otsu algorithmmTo obtain corresponding bianry imageIn order to guarantee that the region of subsequent extracted is Litchi fruits part is corroded by the circular configuration member that radius is more than 8 pixels in the present embodimentIt is secondary corresponding to obtain one The mask images in fruit region, the luminance picture in litchi fruits region is extracted by the point multiplication operation of mask images and V, and is calculated The average value M of the luminance pictureRoIs.On the other hand, it is manually chosen from the training image data comprising red color tone litchi fruits Fruit region, and calculate the weighting three primary colors luminance component and average brightness value M in these regionstrain。
Compare MRoIsAnd Mtrain:
If MRoIs≥Mtrain, then not executing the brightness based on Retinex algorithm to V enhances, directly output I '=Ic；
If MRoIs< Mtrain, then executing the brightness based on Retinex algorithm to V enhances, and obtains new luminance component V ', passes through RGB/HSI color space conversion model extraction IcChrominance component H and color saturation component S, and merge V ' generation brightness enhancing RGB image I ' afterwards, enables I againc=I ' continues according to above-mentioned workflow management IcMRoIs, enhance I in an iterative mannercIt is bright Degree, until meeting condition MRoIs≥MtrainWhen, export the enhanced RGB image I ' of brightness.
Fig. 2 a~Fig. 2 d gives with above-mentioned iterative manner using the effect contrast figure of Retinex algorithm before and after the processing.Figure 2a is the image of a secondary illumination abundance, and brightness reinforcing effect is as shown in Figure 2 b, and Fig. 2 c is the weaker image of a secondary illumination, bright It is as shown in Figure 2 d to spend reinforcing effect.
By formula (2)~(4) from the middle extraction modified R X chromaticity difference diagram C of I 'm', using Otsu algorithm to Cm' execute thresholding Segmentation obtains bianry image Cm＂ successively uses the erosion operation based on coarse scale structures member (radius can be used for 4 pixels Circular configuration member), dilation operation based on small scale structures member (radius can be used first for the circular configuration of 3 pixels) and hole Filling processing Cm＂, by 8- connectivity criteria from treated Cm" in extract potential fruit region, and be labeled as Of.Above-mentioned " big ruler " big " and " small " in very little structural elements ", " small scale structures member " is in processing CmThe erosion operation and dilation operation carried out when ＂ What the structural elements of use were compared.Fig. 3 gives through treated litchi fruits extracted region effect picture, and Fig. 3 a is The litchi fruits extraction effect implementation example figure of image shown in Fig. 2 a, Fig. 3 b are that the litchi fruits extraction effect of image shown in Fig. 2 c is real Apply example diagram.
In order to guarantee the accurate extraction in sepia branch region, the present invention is given birth to again using local hue information restructing algorithm At filtered chrominance component, to eliminate the noise jamming in input picture.By RGB/HSI color space conversion model from I ' Middle extraction chrominance component H '.In any m × n neighborhood of H ', if the coordinate and tone value of the centre of neighbourhood are respectively (xc,yc) and hc, the coordinate and tone value respectively (x of i-th of (i=1,2 ..., m × n) pixeli,yi) and hi, i-th is determined according to formula (5) A pixel (xi,yi) with the positional relationship w of the centre of neighbourhoodpi。
wpi=exp {-λpmax(dis((xi,yi),(xc,yc)))},(xi,yi)≠(xc,yc) (5)
Wherein, λpFor position amplification coefficient, λ is taken in the present embodimentp=2；dis((xi,yi),(xc,yc)) it is metric point (xi,yi) and (xc,yc) the distance between function namely ith pixel taken in the present embodiment at a distance from the centre of neighbourhood chessboard away from From i.e. dis ((xi,yi),(xc,yc))=max (| xi-xc|,|yi-yc|)。
In addition, another key factor for influencing the estimation of m × n centre of neighbourhood hue information is the tone value of neighborhood territory pixel, it is Reduce influence of the centre of neighbourhood polluted by noise to its hue information reconstruction result, the present invention, which measures, removes center in neighborhood Pixel tone standard deviation sigma outside pointcAnd calculating field intermediate value hm, local neighborhood pixel (x is measured by formula (6) and (7)i,yi) With hmDifference on hue information, result queue wHi。
Wherein, λHFor tone difference amplification coefficient, λ is taken in the present embodimentH=1；diff(hi,hc) be ith pixel color Tone pitch hiWith the difference of neighborhood tone intermediate value hm, diff (h is taken in the present embodimenti,hc)=| | hi-hc||2, NcFor except point (xc,yc) Outer neighborhood territory pixel set.
Since the tone value of centre of neighbourhood pixel can be influenced by the position of m × n neighborhood territory pixel and tone value simultaneously, this It invents while considering wpiAnd wHiCarry out reconstruction point (xc,yc) hue information, allow it closer to practical tone value, adopted in the present embodiment Neighborhood territory pixel (x is formed with the mode of dot producti,yi) weight, i.e. wci=wpi×wHi.Particularly, point (xc,yc) before reconstitution Tone may be interfered by noise, so for (xi,yi)=(xc,yc) situation, if hcWith hmDeviation is larger, then w is arrangedci =0, otherwise w is setci=1.Finally the weighted mean of m × n neighborhood hue information is calculated as point (x according to formula (8)c,yc) Reconstruction result.
After whole pixels to H ' complete above-mentioned calculating, the reconstructed image H ' of H ' can be obtainedc.Fig. 4 gives m × n The centre of neighbourhood pixel tonal value reconstruct numerical operation result figure, Fig. 4 a be centre of neighbourhood pixel not by noise pollution when tone Be worth reconstruction result, Fig. 4 b be centre of neighbourhood pixel by noise pollution when tone value reconstruction result, take m=5 in the present embodiment, N=5.Each grid in Fig. 4 a and Fig. 4 b represents a pixel, and the first row numerical value in grid indicates the point before reconstitution Tone value, the numerical value in the second row bracket indicates contribution weight of the hue information of the point in restructuring procedure, neighborhood midpoint In the third line numerical value indicate the tone value of point after reconstitution.
Litchi branch strip area is manually extracted from training image data, and counts these areas after being transformed into HSI color space The average tone value h in domainavg, as to H 'cCarry out the segmentation threshold of binary conversion treatment, H 'cMiddle tone value is less than or equal to havgPicture Element is set to prospect, is otherwise set to background, and potential branch region O is extracted in the foreground image obtained by as 8- connectivity criterias。 Fig. 5 gives the effect picture of litchi branch strip area extraction, and Fig. 5 a is the lichee branch extraction effect embodiment of image shown in Fig. 2 a Figure, Fig. 5 b are the lichee branch extraction effect implementation example figure of image shown in Fig. 2 c.
O is handled using thinning algorithms, extract the skeleton O of litchi branch strip areas', O is detected by Harris algorithms' in All angle points (set its quantity as k).In order to judge whether certain angle point is located at branch bifurcation, all angle points are traversed, for jth (j=1,2 ..., k) a angle point, if its neighborhood territory pixel meets mode distribution rule shown in formula (9) simultaneously, by angle point j Coordinate be added to angled key point set AcIn, otherwise delete angle point j.
Wherein, N8(Pi) indicate that 8 neighborhoods of angle point j (do not include the pixel quantity that value is 1 in angle point j), P2、P4、P6With P8For the 4 neighborhood territory pixel values of angle point j, and they are located at top, right, lower section and the left of angle point j.
To OfMorphological dilations operation (radius can be used for the circular configuration of 3 pixels member) is executed, O is obtainedf', by Os′ With Of' region superposition is carried out, therefrom search for the join domain A of litchi fruits and branchr, and calculate ArMiddle all pixels point row is sat Target mean value Pr1With the mean value P of column coordinater2.Then according to the opposite position of litchi fruits under gravity and litchi branch strip area It sets and adhesion relation, calculates AcThe mean value P of middle key angle point row coordinatec1With the mean value P of column coordinatec2If AcFor empty set or Pr1 < Pc1, from Os' middle rejecting and Of' do not have connectivity branch region, extract only with OfBranch skeleton with connectivity.Most Afterwards, from AcThe middle the smallest angle point of search row coordinate value, and centered on the angle point, with Pr2With Pc2Between absolute difference be standard Difference determines the harvesting point distributed area of litchi fruits.Fig. 6 a, Fig. 6 b give the effect picture of litchi fruits harvesting point location, figure 6a is that the litchi fruits of image shown in Fig. 2 a harvest point location effect example figure, and Fig. 6 b is the litchi fruits of image shown in Fig. 2 c Harvest point location effect example figure.
The same or similar label correspond to the same or similar components；
Positional relationship is described in attached drawing only for illustration, should not be understood as the limitation to this patent；
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention Protection scope within.

Claims (10)

1.一种面向自然场景的果实采收区域自动定位方法，其特征在于，包括以下步骤：1. an automatic positioning method for the fruit harvesting area facing natural scenes, is characterized in that, comprises the following steps: (1)提取当前RGB图像I的加权三原色亮度分量V，利用修正RX色差图(Modified Redand Green/Blue(X)Chromatic Mapping,MRXCM)方法从I中生成修正RX(Red&Green/Blue)色差图Cm，从Cm中提取局部果实区域RoIs(Regions of Interest)，计算RoIs亮度分布的统计特征，根据统计特征以迭代方式采用Retinex算法处理V并重构I，当迭代过程终止时通过颜色空间转换模型输出亮度增强后的RGB图像I′；(1) Extract the weighted three primary color luminance components V of the current RGB image I, and use the Modified Red and Green/Blue(X) Chromatic Mapping (MRXCM) method to generate a modified RX (Red & Green/Blue) color difference map C m from I , extract the local fruit region RoIs (Regions of Interest) from C m , calculate the statistical characteristics of the brightness distribution of RoIs, use the Retinex algorithm to process V and reconstruct I in an iterative manner according to the statistical characteristics, and convert the model through the color space when the iterative process is terminated. Output the RGB image I' after brightness enhancement; (2)通过MRXCM方法从中提取修正RX色差图Cm′，对Cm′执行图像阈值化分割得到Cm＂，并利用数学形态学算法处理Cm＂，从Cm＂中提取潜在成熟或未成熟果实区域Of；(2) Extract the corrected RX chromatic aberration map C m ' from the MRXCM method, perform image threshold segmentation on C m ' to obtain C m ", and use mathematical morphology algorithm to process C m ", extract potential mature or Immature fruit area Of; (3)提取I′的色调分量H′，衡量局部邻域像素与邻域中心的位置关系，衡量局部邻域像素与邻域中心的色调信息差异，建立局部邻域对中心像素色调值的权重分配模型，通过隔离受噪声污染的邻域中心像素、以加权均值的方式重构色调分量Hc′，对Hc′执行图像阈值化分割得到Hc＂，从中提取潜在的棕褐色枝条区域Os；(3) Extract the hue component H' of I', measure the positional relationship between the local neighborhood pixels and the neighborhood center, measure the difference in hue information between the local neighborhood pixels and the neighborhood center, and establish the weight of the local neighborhood to the central pixel's hue value The allocation model, by isolating the center pixel of the neighborhood polluted by noise, reconstructing the hue component H c ' in the way of weighted mean, performing image thresholding segmentation on H c ' to obtain H c ", and extracting the potential tan branch region O s ; (4)提取Os的骨架Qs′，检测Qs′上的角点并提取与枝条分叉属性相关的关键角点，利用重力作用下Of和Qs′的相对位置约束和关键角点的空间分布特性，定位果实采收点分布区间。(4) Extract the skeleton Q s ′ of O s , detect the corner points on Q s ′ and extract the key corner points related to branch bifurcation properties, and use the relative position constraints and key angles of O f and Q s ′ under the action of gravity The spatial distribution characteristics of points, and the distribution interval of fruit harvesting points is located. 2.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(1)所述提取当前RGB图像I的加权三原色亮度分量V，具体为，用变量R、G和B分别表示I的红、绿和蓝颜色通道的像素灰度值，按照下式计算加权三原色亮度分量V：2. the automatic positioning method of the fruit harvesting area facing natural scene according to claim 1, is characterized in that, described in step (1) extracts the weighted three primary color luminance component V of current RGB image I, is specifically, with variable R, G and B represent the pixel gray values of the red, green and blue color channels of I, respectively, and the weighted three primary color luminance components V are calculated according to the following formula: V＝αR+βG+γBV=αR+βG+γB 3.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(1)所述利用修正RX色差图(Modified Red and Green/Blue(X)Chromatic Mapping,MRXCM)方法从I中生成修正RX(Red&Green/Blue)色差图Cm，具体为，根据果实区域表现出的红色调或绿色调，计算R与G或R与B的单色比值作为比例因子，并加权到常规的红绿色差图或红蓝色差图中，生成修正的色差图Cm。3. the automatic positioning method of the fruit harvesting area oriented to natural scene according to claim 1, is characterized in that, described in step (1) utilizes modified RX chromatic aberration map (Modified Red and Green/Blue (X) Chromatic Mapping, MRXCM ) method to generate a corrected RX (Red&Green/Blue) color difference map C m from I, specifically, according to the red or green tone exhibited by the fruit area, calculate the monochromatic ratio of R to G or R to B as a scale factor, and Weighted to a conventional red-green difference map or a red-blue difference map to generate a corrected color difference map C m . 4.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(1)所述从Cm中提取局部果实区域RoIs(Regions of Interest)，计算RoIs亮度分布的统计特征，具体为，对Cm执行阈值分割和腐蚀运算得到二值图像，提取局部果实区域RoIs作为掩模图像，并计算该掩模图像在V中所覆盖区域的平均亮度值MRoIs。4. The method for automatically locating a fruit harvesting area for a natural scene according to claim 1, wherein the step (1) extracts the local fruit region RoIs (Regions of Interest) from C m , and calculates the RoIs brightness distribution Specifically, perform threshold segmentation and erosion operation on C m to obtain a binary image, extract the local fruit area RoIs as a mask image, and calculate the average brightness value M RoIs of the area covered by the mask image in V. 5.根据权利要求4所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(1)所述根据统计特征以迭代方式采用Retinex算法处理V并重构I，当迭代过程终止时通过颜色空间转换模型输出亮度增强后的RGB图像I′，具体为，提取多个训练图像的三原色加权亮度分量，统计果实区域的平均亮度值Mtrain，并执行以下步骤：5. the automatic positioning method of the fruit harvesting area oriented to natural scene according to claim 4, is characterized in that, described in step (1), adopts Retinex algorithm to process V and reconstruct I according to statistical feature in iterative manner, when iterative process. At the termination, the RGB image I′ with enhanced brightness is output through the color space conversion model. Specifically, the three-primary-color weighted brightness components of multiple training images are extracted, the average brightness value M train of the fruit area is counted, and the following steps are performed: S1.判断是否MRoIs≥Mtrain，若是则执行步骤S2，若否则执行步骤S3；S1. Determine whether M RoIs ≥ M train , if so, execute step S2, if otherwise, execute step S3; S2.将I作为亮度增强后的RGB图像I′并输出I′；S2. Take I as the RGB image I' after the brightness enhancement and output I'; S3.利用Retinex算法获得增强后的亮度分量V′，通过RGB/HSI颜色空间转换模型提取当前RGB图像的色调H和色饱和度S，融合V′以生成亮度增强后的RGB图像I′，重新提取I′的平均MRoIs＇，判断是否MRoIs＇≥Mtrain，若否则继续执行步骤S3，若是则继续执行步骤S4；S3. Use the Retinex algorithm to obtain the enhanced brightness component V', extract the hue H and color saturation S of the current RGB image through the RGB/HSI color space conversion model, and fuse V' to generate the enhanced brightness RGB image I', re- Extract the average M RoIs ' of I', determine whether M RoIs '≥M train , if otherwise, continue to perform step S3, if so, continue to perform step S4; S4.输出I′。S4. Output I'. 6.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(3)所述衡量局部邻域像素与邻域中心的位置关系，具体为，在领域中心为(xc,yc)的m×n邻域中，计算坐标为(xi,yi)的第i个像素与邻域中心的距离dis((xi,yi),(xc,yc))，按照下式得到第i个像素与邻域中心的位置关系wpi：6. The method for automatically locating a fruit harvesting area for a natural scene according to claim 1, wherein the step (3) measures the positional relationship between the local neighborhood pixel and the neighborhood center, specifically, in the field center In the m×n neighborhood of (x c , y c ), calculate the distance between the ith pixel whose coordinates are (x i , y i ) and the center of the neighborhood dis((x i ,y i ),(x c ,y c )), the positional relationship w pi between the i-th pixel and the neighborhood center is obtained according to the following formula: wpi＝exp{-λPmax(dis((xi,yi),(xc,yc)))}w pi =exp{-λ P max(dis((x i ,y i ),(x c ,y c ))))} 式中，λp为位置增幅系数，i＝1,2,…,m×n。In the formula, λ p is the position amplification coefficient, i=1,2,...,m×n. 7.根据权利要求1或6所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(3)所述衡量局部邻域像素与邻域中心的色调信息差异，具体为，在领域中心坐标为(xc,yc)的m×n邻域中，计算邻域中坐标为(xi,yi)的第i个像素的色调值hi与邻域色调中值hm的差异diff(hi,hm)，以及除中心点外的邻域像素色调值标准差σc，按照下式计算第i个像素与邻域中心的差异信息wHi：7. the method for automatically locating the fruit harvesting area of the natural scene according to claim 1 or 6, is characterized in that, described in step (3) to measure the difference in hue information of local neighborhood pixel and neighborhood center, specifically, In the m×n neighborhood where the coordinates of the center of the domain are (x c , y c ), calculate the hue value h i of the ith pixel whose coordinates are (x i , y i ) in the neighborhood and the median hue value h of the neighborhood The difference diff(h i ,h m ) of m , and the standard deviation σ c of the hue value of the neighboring pixels except the center point, calculate the difference information w Hi between the ith pixel and the neighborhood center according to the following formula: (xi,yi)≠(xc,yc) (x i ,y i )≠(x c ,y c ) 式中，λH为色调差异增幅系数，i＝1,2,…,m×n。In the formula, λ H is the hue difference amplification coefficient, i=1, 2, . . . , m×n. 8.根据权利要求7所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(3)所述建立局部邻域对中心像素色调值的权重分配模型，通过隔离受噪声污染的邻域中心像素、以加权均值的方式重构色调分量H′c，具体为，通过wpi和wHi的点乘运算分配第i个像素的权重wci，若邻域中心色调值hc与hm偏差较大，则设置领域中心的权重为0，否则设置为1，计算邻域中各像素色调值的加权平均值，作为邻域中心像素的重构色调值h′c，基于上述流程遍历H′的所有像素后得到重构的色调分量H′c。8. The method for automatically locating a fruit harvesting area for a natural scene according to claim 7, wherein the step (3) establishes a weight distribution model for the central pixel tone value of the local neighborhood, and is polluted by noise by isolating The neighborhood center pixel of , reconstructs the hue component H' c in the form of weighted mean value, specifically, assigns the weight w ci of the ith pixel through the dot product operation of w pi and w Hi , if the neighborhood center hue value h c If the deviation from h m is large, set the weight of the center of the field to 0, otherwise set it to 1, and calculate the weighted average of the hue values of each pixel in the neighborhood as the reconstructed hue value h' c of the pixel in the center of the neighborhood, based on the above The process traverses all pixels of H' to obtain the reconstructed hue component H' c . 9.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(4)所述提取Os的骨架Qs′，检测Qs′上的角点并提取与枝条分叉属性相关的关键角点，具体为，利用细化算法提取Os的骨架Qs′，采用Harris角点检测算法提取Qs′上的全部角点，根据多果粘连情况下不同果实枝条之间的分叉特点，利用分叉处角点邻域像素的模式分布规则和骨架连通性判断分叉处角点是否属于关键角点并提取关键角点。9 . The method for automatically locating a fruit harvesting area for a natural scene according to claim 1 , wherein in step (4), the skeleton Q s ′ of O s is extracted, and the corner points on Q s ′ are detected and extracted. 10 . The key corner points related to branch bifurcation attributes are, specifically, using the refinement algorithm to extract the skeleton Q s ' of O s , and using the Harris corner detection algorithm to extract all the corner points on Q s '. For the bifurcation characteristics between fruit branches, the pattern distribution rules and skeleton connectivity of the neighborhood pixels of the bifurcations are used to determine whether the bifurcation corners belong to the key corners and extract the key corners. 10.根据权利要求1所述的面向自然场景的果实采收区域自动定位方法，其特征在于，步骤(4)所述利用重力作用下Of和Qs′的相对位置约束和关键角点的空间分布特性，定位果实采收点分布区间，具体为，根据重力作用下果实重心位置低于其连接枝条、果实果柄与枝条间的强连通性特点，计算果柄与枝条连接区域Ar中所有像素空间坐标的统计特征，并结合关键角点坐标的统计特征提取仅与Of具有连通性的枝条骨架，提取该骨架上远离果柄的关键角点作为果实的采收中心点，以Ar中像素和关键角点空间列坐标的离散程度作为偏离采收点的区域标准差，根据采收中心点和区域标准差定位果实采收点分布区间。10. The method for automatically locating a fruit harvesting area for a natural scene according to claim 1, wherein the step (4) utilizes the relative position constraints of O f and Q s ' under the action of gravity and the relative position constraints of the key corners. Spatial distribution characteristics, locate the distribution interval of fruit harvesting points, specifically, according to the strong connectivity characteristics between the fruit stalk and the branch, the fruit stalk and the branch connection area Ar Statistical features of all pixel space coordinates, combined with the statistical features of key corner coordinates to extract the branch skeleton that only has connectivity with O f , and extract the key corner points on the skeleton far from the fruit stalk as the harvesting center point of the fruit, with A The degree of dispersion of the spatial column coordinates of pixels and key corner points in r is taken as the regional standard deviation from the harvesting point, and the distribution interval of fruit harvesting points is located according to the harvesting center point and the regional standard deviation.
CN201811062603.9A 2018-09-12 2018-09-12 An automatic positioning method for fruit harvesting area oriented to natural scenes Active CN109359531B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811062603.9A CN109359531B (en) 2018-09-12 2018-09-12 An automatic positioning method for fruit harvesting area oriented to natural scenes

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811062603.9A CN109359531B (en) 2018-09-12 2018-09-12 An automatic positioning method for fruit harvesting area oriented to natural scenes

Publications (2)

Publication Number Publication Date
CN109359531A true CN109359531A (en) 2019-02-19
CN109359531B CN109359531B (en) 2021-12-14

Family

ID=65350966

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811062603.9A Active CN109359531B (en) 2018-09-12 2018-09-12 An automatic positioning method for fruit harvesting area oriented to natural scenes

Country Status (1)

Country Link
CN (1) CN109359531B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112541860A (en) * 2019-09-23 2021-03-23 深圳开阳电子股份有限公司 Skin color beautifying correction method and device
CN118058074A (en) * 2024-02-20 2024-05-24 广东若铂智能机器人有限公司 Method for judging burst interference in string-type fruit picking process

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101726251A (en) * 2009-11-13 2010-06-09 江苏大学 Automatic fruit identification method of apple picking robot on basis of support vector machine
CN103727877A (en) * 2013-12-20 2014-04-16 北京农业信息技术研究中心 Fruit identifying and locating method, device and system
CN104700404A (en) * 2015-03-02 2015-06-10 中国农业大学 Fruit location identification method
CN105373794A (en) * 2015-12-14 2016-03-02 河北工业大学 Vehicle license plate recognition method
CN105701829A (en) * 2016-01-16 2016-06-22 常州大学 Bagged green fruit image segmentation method
US9462749B1 (en) * 2015-04-24 2016-10-11 Harvest Moon Automation Inc. Selectively harvesting fruits
CN106326894A (en) * 2016-08-31 2017-01-11 西南交通大学 Adverse state detection method of transverse pins of rotation double lugs of high-speed rail overhead contact line equipment
CN106327443A (en) * 2016-08-24 2017-01-11 电子科技大学 Night image enhancement method based on improved genetic algorithm
CN108319973A (en) * 2018-01-18 2018-07-24 仲恺农业工程学院 Detection method for citrus fruits on tree

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101726251A (en) * 2009-11-13 2010-06-09 江苏大学 Automatic fruit identification method of apple picking robot on basis of support vector machine
CN103727877A (en) * 2013-12-20 2014-04-16 北京农业信息技术研究中心 Fruit identifying and locating method, device and system
CN104700404A (en) * 2015-03-02 2015-06-10 中国农业大学 Fruit location identification method
US9462749B1 (en) * 2015-04-24 2016-10-11 Harvest Moon Automation Inc. Selectively harvesting fruits
CN105373794A (en) * 2015-12-14 2016-03-02 河北工业大学 Vehicle license plate recognition method
CN105701829A (en) * 2016-01-16 2016-06-22 常州大学 Bagged green fruit image segmentation method
CN106327443A (en) * 2016-08-24 2017-01-11 电子科技大学 Night image enhancement method based on improved genetic algorithm
CN106326894A (en) * 2016-08-31 2017-01-11 西南交通大学 Adverse state detection method of transverse pins of rotation double lugs of high-speed rail overhead contact line equipment
CN108319973A (en) * 2018-01-18 2018-07-24 仲恺农业工程学院 Detection method for citrus fruits on tree

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
CHENGLIN WANG等: "A robust fruit image segmentation algorithm against varying illumination for vision system of fruit harvesting robot", 《OPTIK》 *
刘阳: "自然环境下目标物的高速图像检测算法研究", 《中国博士学位论文全文数据库 信息科技辑》 *
李毅 等: "尺度变化的Retinex红外图像增强", 《液晶与显示》 *
沈甜: "苹果采摘机器人重叠果实快速动态识别及定位研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
熊俊涛 等: "基于Retinex图像增强的不同光照条件下的成熟荔枝识别", 《农业工程学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112541860A (en) * 2019-09-23 2021-03-23 深圳开阳电子股份有限公司 Skin color beautifying correction method and device
CN118058074A (en) * 2024-02-20 2024-05-24 广东若铂智能机器人有限公司 Method for judging burst interference in string-type fruit picking process

Also Published As

Publication number Publication date
CN109359531B (en) 2021-12-14

Similar Documents

Publication Publication Date Title
CN105718945B (en) Apple picking robot night image recognition method based on watershed and neural network
CN106951836B (en) Crop Coverage Extraction Method Based on Prior Threshold Optimizing Convolutional Neural Network
CN108573276B (en) A change detection method based on high-resolution remote sensing images
CN108319973B (en) Detection method for citrus fruits on tree
Huang et al. Using fuzzy mask R-CNN model to automatically identify tomato ripeness
Liu et al. A method of segmenting apples at night based on color and position information
CN108537239B (en) Method for detecting image saliency target
CN107578418A (en) A Contour Detection Method for Indoor Scenes by Fusion of Color and Depth Information
CN109636784A (en) Saliency object detection method based on maximum neighborhood and super-pixel segmentation
Narvekar et al. Grape leaf diseases detection & analysis using SGDM matrix method
CN113255434B (en) An apple recognition method that combines fruit features and deep convolutional neural network
CN105574514A (en) Greenhouse immature tomato automatic identification method
CN108629289A (en) The recognition methods in farmland and system, applied to the unmanned plane of agricultural
Changhui et al. Overlapped fruit recognition for citrus harvesting robot in natural scenes
Patil Pomegranate fruit diseases detection using image processing techniques: a review
CN109359531A (en) Fruit harvesting area automatic positioning method facing natural scene
CN109035223A (en) A kind of intelligent evaluation method for satellite remote sensing images availability
Salazar-Gonzalez et al. Retinal blood vessel segmentation via graph cut
Tran et al. Automatic dragon fruit counting using adaptive thresholds for image segmentation and shape analysis
CN111832480B (en) A remote sensing identification method of rape planting area based on spectral features
CN117274981B (en) Pepper pathology prediction method and device based on image recognition
Li et al. Image processing for crop/weed discrimination in fields with high weed pressure
Areni Automatic counting of chili ripeness on computer vision for industri 4.0
CN116563721B (en) Tobacco field extraction method based on layered classification thought
Yang et al. Cherry recognition based on color channel transform

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