CN102789579A - Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology - Google Patents
Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology Download PDFInfo
- Publication number
- CN102789579A CN102789579A CN2012102602590A CN201210260259A CN102789579A CN 102789579 A CN102789579 A CN 102789579A CN 2012102602590 A CN2012102602590 A CN 2012102602590A CN 201210260259 A CN201210260259 A CN 201210260259A CN 102789579 A CN102789579 A CN 102789579A
- Authority
- CN
- China
- Prior art keywords
- color
- algorithm
- image
- characteristic
- crop
- 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
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to an identification method for a stressed state of a water fertilizer of a greenhouse crop on basis of a computer vision technology. According to the identification method provided by the invention, a crop under a greenhouse environment is taken as a research object; a computer vision monitoring platform is constructed; a plant image cutting method which adapts to the change in natural illumination and a complex scene is researched; an obtained plant blade image is extracted at the aspects of morphology, color, grain and the like, and sufficient characteristic sets are constructed; a heuristic search algorithm, such as a genetic algorithm, a simulated annealing algorithm, an ant colony algorithm, a particle swarm optimization, or the like, is combined with a neural network technique for searching for the optimal characteristic subset; and a BP (Back Propagation) neural network is utilized to identify a stressed characteristic of the crop. A camera is moved by using a horizontal positioning system, so that the plant image is all-dimensionally obtained; the algorithm operation is realized by using a CUDA (Compute Unified Device Architecture) hardware platform, so as to meet the real-time demand on monitoring; and the invention provides a technical method for measuring destructiveness under the stressed state of the water fertilizer of the greenhouse crop and the application prospect is wide.
Description
Technical field
The invention belongs to hothouse plants monitoring field, be specifically related to a kind of chamber crop liquid manure and coerce state identification method based on computer vision.
Background technology
Industrialized agriculture has become the important channel that solves population, grain, soil contradiction as novel agricultural production mode.Current; The ubiquity scientific and technological content was low during China industrialized agriculture was produced, labour intensity big, production level and benefit are low; Press for technological improvement, particularly utilize modern information technologies, realize the facility greenhouse is controlled and intelligent management automatically; To improve China's industrialized agriculture production technology level, further increase economic efficiency and resource utilization.Realize the accurate control of greenhouse, obtain the plant physiology state and comprise that information such as moisture, nutrition are vital.Traditional greenhouse moisture, nutrient solution control are to carry out according to artificial observation or parameter preset, rather than to the plant specific needs of particular moment.The contact measurement is usually used in confirming the physiological characteristic of plant, but has inconvenient operation, wastes time and energy, has destructiveness, is subject to the subjective factor influence, is not suitable for characteristics such as monitoring in real time.When crop water deficient, nutritional deficiency etc. occur and coerces characteristic, can have a strong impact on its growth.For this reason, identification is as early as possible screened, so that further take measures to control, avoids suffering a loss.The plant stress state reflects through the blade face, like crinkle, withered, sagging, jaundice.This structural change of blade meeting changes reflection of light, and the slight change of color that causes or texture aspect can be used for the physiological change [1] of monitoring plant.Utilize computer vision system to plant growth carry out non-cpntact measurement have quick nondestructive, promptly and accurately, characteristics [2] such as round-the-clock; And can obtain growth parameter(s), integrality information and its specific demand of identification of crop,, effectively improve the utilization of resources for irrigation and environment control rationally provides reliable basis; Energy savings consumes; Promote product quality, reduce operating cost, reach the purpose that improves output, saving cost, protection environment.
(1) moisture state monitoring aspect.Plant utilization moisture is kept blade health, in the time of can't satisfying transpiration when the moisture undersupply, and the blade stomatal closure, the rising minimizing, photosynthesis weakens, blade wither [3].No matter the crop water shortage that physiologic factor or non-physiologic factor cause all can influence plant growth and metabolism, serious meeting influences crop yield.The early diagnosis of crop water shortage for increase output, reduce the loss particularly important.Leaf water content, stemflow method etc. are owing to the direct information that the plant moisture state is provided is widely used in plant moisture state analysis [4]; But there are directly contact, destructive sampling, gather the limited shortcoming of sample, be difficult to obtain large-scale business promotion and use.Human eye is bigger to the perception individual difference of light, and different people is just variant to the color and style estimation of lack of water symptom, and computer vision can remedy this defective.(1992) [5] such as Seginer I show the vertical movement of blade tip and lack of water, carbon dioxide absorption rate height correlation through the observation experiment to full ripe tomato leaf.(1996) [6] such as Kurata K utilize graphical analysis to obtain the inclination angle of tomato rachis, and carry out related with plant moisture it.(1998) [7] such as Revollon P utilize the variation of angle between blade tip axis and the horizontal line of ornamental plant that plant hydropenia is launched research.(2002) [8] such as Kacira M then utilize integral shroud top projected area, and (Top Projected Canopy Area is TPCA) as recognition feature and set up the detection index of plant hydropenia.(2004) [9] such as Foucher P utilize neural network to cut apart the crop map picture; TPCA is calculated three shape facilities such as single order invariant moments, fractal dimension, average stem stalk skeleton length; Contrast experiment through to two groups of potted plants of lack of water and normal irrigation shows; Through choosing appropriate threshold, three characteristics can be diagnosed out the lack of water symptom of crop.6 textural characteristics have been extracted in (2008) [1] such as Ondimu S N from the gray scale of mossery picture and colored covariance matrix, and utilize the multilayer perceptron neural network that exsiccosis is discerned.
Chinese patent " based on the crop water-requesting information determination of computer vision " (application number: 200510041045); Utilize reference substance to measure the new method of crop cane size and fruit size; Being used to detect crop needs water information, and detection system is made up of the suitable reference substance of size, image capture device, image pick-up card and computing machine; Utilize computer vision non-cpntact measurement crop cane and fruit,, obtain crop cane size and fruit size, need water information thereby obtain crop through corresponding image processing algorithm through being placed on object of reference near the crop cane and fruit to be measured." based on the irrigation water truck system of computer vision " (application number: 2006100287346), comprise closing comprising database hub and in order to the monitoring and the remote supervisory and controlling equipment of real-time display message; The irrigation control box that embedded board machine, programmable controller and digital display equipment constitute; A ccd video camera that carries out real time image collection; Variable-frequence governor and the relay that is connected with variable-frequence governor, solenoid valve and the topworks of control waterwheel speed of travel and start and stop; And the photoimpact scrambler and the pulsed flowmeter that obtains current irrigation volume that are used to obtain the current waterwheel speed of travel and travel distance; Carry out the wireless ethernet device of real-time radio communication; Be used to control the solenoid valve of little spray nozzle switch and the compositions such as walking track of irrigating waterwheel.Relatively confirm irrigation volume according to stored historical information in plant size of obtaining and the database.
(2) nutritional deficiency monitoring aspect.Element such as nitrogen, calcium is extremely important to the growth of plant, often occurs but soil is lacked the excessive fertilising that effective information causes, and causes very big threat to environment.In addition, in greenhouse production, plant has deficiency symptom usually, is particularly blooming and period as a result, has a strong impact on output and the quality [10] of crop.Early diagnosis and timely potential problems such as the tip burn on leaf handled to the crop nutritional deficiency can promote resource utilization, improve the quality and the output of agricultural product.The early symptom of nutritional deficiency is also not obvious, even if veteran expert also is difficult to definitely be diagnosed.Whether nitrogen stress will depend on the identification that leaf color changes to plant identification usually, and crop alimentary is bad mainly to be reflected through blade, and mainly embodies [11] through color and texture.(1996) [12] such as Ahmad I S utilize color characteristic to represent plant hydropenia or nutritional deficiency level, utilize low, the middle nitrogen level of the mean value differentiation crop of R (redness), G (green), H (tone) component.(2004) [13] such as Borhan M S are based on the grey level histogram of R, G component and the multispectral figure image of coloured image; Four characteristics such as average, deviation, energy, entropy have been extracted; Utilize multiple linear regression analysis; Plant chlorophyll and nitrogen level are predicted that recognition result is superior to manual observation.(2011) [11] such as Xu G L utilize the histogram feature of b* component in the CIE color space; In conjunction with Fourier transform and wavelet analysis the plant leaf blade nutritional deficiency is discerned; For lucifuge line and other Effect of Environmental, Image Acquisition is in the sampling box of a sealing, to carry out.(2010) [14] such as David Story are in the controllable environment of radiation source at LED; Utilize 1 morphological feature of plant leaf blade, 2 colors and 4 textural characteristics that the lettuce calcium deficiency is discerned, the result shows that the comparable artificial vision of computer vision found the nutritional deficiency state in early 1-2 days.Domestic, Li Changying etc. (2003) [15] utilize the variation of integral shroud projected area to reflect the fertilizer deficiency situation of plant, disturb (like wind speed) because the calculating of integral shroud projected area is subject to external condition, have limited the application of this method.12 characteristics that the rare equality (2003) [16] of hair is extracted leaf color and texture are discerned the tomato nutritional deficiency.
Computer vision technique is coerced crop and has been obtained remarkable progress aspect the characteristic monitoring; But; Also have some problems, because the agricultural crops ambient lighting is uneven, natural lighting changes greatly, image-forming condition is undesirable, background is complicated, cut apart difficulty, most of research is carried out under the controlled laboratory condition of illumination (1); So that the gray difference of target crop and background is bigger, the target segmentation problem that adapts to natural lighting variation, complex background image fails effectively to be solved; (2) image processing algorithm in the vision system is to be the special object customization, lacks robustness when condition changes, and is difficult to expand and reuse during in the face of different problem; (3) in the past research is to monitor to the single characteristic (like lack of water or nitrogen stress) of coercing of plant, and the successful visual monitor system requirement of a cover can be discerned a plurality of characteristics of coercing simultaneously; (4) extracting one or more characteristics after the image segmentation discerns; The characteristic of coercing of Different Crop different conditions is different; How to construct identification and coerce characteristic characteristic set required, that quantity is enough big, and self-adaptation is optimized the shortage validity method of selecting to Different Crop, from aspects such as form, color, textures; (5) image processing algorithm is consuming time, how to select economy, rationally, realize that quick computing also is a problem with the requirement of satisfying monitoring in real time on the efficient hardware platform.
The present invention is directed to above problem; Through making up the computer vision monitoring platform; With the crop in the greenhouse is research object; Exploitation adapts to plant image dividing method under natural lighting variation, the complex scene; And the enough big characteristic set of structure quantity, and adaptively selected optimal feature subset carry out the Computer Vision Recognition method that crop liquid manure is coerced characteristic. with preceding two kinds of technology through the size of the different growth phases of plant confirm water requirement different be that the present invention selects optimal characteristics to gather from various features such as the texture of crop leaf, color, form to judge water, fertile information, the identification of realization water, fertile two states; Satisfy chamber crop moisture, fertilizer coerce state in real time, non-destructive monitoring requirement, for environment control provides foundation.
List of references
[1] Ondimu S N, Murase H. Comparison of plant water stress detection ability of color and gray level texture in sunagoke moss. Transactions of the ASABE,2008, Vol. 51(3): 1111~11120。
[2] woods smiles, Xu Lihong, Wu Junhui. the progress of computer vision technique in the plant growth monitoring. and " EI ", 2004,20 (2): 279 ~ 283.
[3] Nilsson H. Remote sensing and image analysis in plant pathology. Annual Review of Phytopathology, 15(1995):489~527。
[4] Wang D, Gartung J. Infrared canopy temperature of early-ripening peach trees under postharvest deficit irrigation. Agricultural Water Management,97(2010):1787~1794。
[5] Seginer I, Elster R T, Goodrum, J W, Rieger M W. Plant wilt detection bycomputer-vision tracking of leaf tips. Transactions of the ASABE, 35 (1992):1563~1567。
[6] Kurata K, Yan J. Water stress estimation of tomato canopy based on machine vision. Acta Horticulturae,440(1996):389~394。
[7]Revollon P, Chasseriaux G, Riviere L M et al.The use of image processing for tracking the morphological modification of Forsythia following an interruption of watering. In Proc. International Conference on Agricultural Engineering, 872–873. AgEng OSLO98. 1998,August 24~27. Oslo, Norway。
[8] Kacira M, Ling P P, Short T H. Machine vision extracted plant movement for early detection of plant water stress. Transactions of the ASAB , 2002,45 (4):1147~1153。
[9] Foucher P, Revollon1 P, Vigouroux B,Chasseriaux G. Morphological Image Analysis for the Detection of Water Stress in Potted Forsythia. Biosystems Engineering (2004),89 (2):131~138。
[10] Mercure, S A, Daoust B, Samson G. Causal relationship between growth inhibition, accumulation of phenolic metabolites, and changes of UV-induced fluorescences in nitrogen-deficient barley plants. Can. J. Bot. 82(2004): 815~821。
[11] Xu G L,Zhang F L, Shah S G et al. Use of leaf color images to identify nitrogen and potassium deficient tomatoes. Pattern Recognition Letters ,32 (2011) :1584~1590。
[12] Ahmad I S, Reid J F. Evaluation of colour representations for maize images. Journal of Agricultural and Engineering Research, 63(1996):185~196。
[13] Borhan M S, Panigrahi S, Lorenzen J H, Gu H. Multispectral and color imaging techniques for nitrate and chlorophyll determination of potato leaves in a controlled environment. Transactions of the American Society of Agricultural Engineers,2004, Vol. 47(2): 599~608。
[14] David Story, Murat Kacira, Chieri Kubota et al. Lettuce calcium deficiency detection with machine vision computed plant features in controlled environments. Computers and Electronics in Agriculture, 74 (2010): 238~243。
[15] Li Changying, Teng Guanghui, Zhao Chunjiang etc. utilize computer vision technique to realize non-destructive monitoring to the hothouse plants growth. EI, 2003,19 (3): 140 ~ 143.
[16] Mao Hanping, Xu Guili, Li Pingping. the identification that wanes based on the tomato nutrient element of computer vision. agricultural mechanical journal, 2003,34 (2): 73 ~ 75.
Summary of the invention
The object of the present invention is to provide a kind of chamber crop to coerce state identification method based on computer vision; The present invention makes up chamber crop computer vision monitoring technology platform; To crop in the greenhouse carry out continuously, non-destructive monitoring, realize that crop liquid manure coerces the early diagnosis of characteristic.
The chamber crop based on computer vision that the present invention proposes is coerced state identification method, and concrete steps are following:
(1) makes up the computer vision monitoring platform; Said monitoring platform is made up of CCD camera 1, support 9, computing machine 3, controllor for step-by-step motor 4, directions X motor 5, Y direction motor 6, directions X guide rail 7 and Y traversing guide 8; Wherein: guide rail 8 is positioned at both sides before and after the top, greenhouse, and guide rail 7 two ends are positioned on the guide rail 8; Y direction motor 6 control directions X guide rails 7 are in guide rail 8 upper edge Y direction motions; CCD camera 1 is installed on the directions X guide rail 7 through support 9, and directions X motor 5 control supports 9 move around at guide rail 7 upper edge directions Xs, and CCD camera 1 is positioned at the chamber crop top; CCD camera 1 is through the image pick-up card 2 even input end of computing machine 3, and the output terminal of computing machine 3 connects the input end of controllor for step-by-step motor 4, and the output terminal of controllor for step-by-step motor 4 connects directions X motor 5 and Y direction motor 6 respectively.Computing machine is realized the horizontal location of CCD camera above the greenhouse through sending instruction to controllor for step-by-step motor 4, gathers the crop map picture successively; Computing machine has the CUDA hardware platform, is used to realize the complicated image processing operations.Motor 5, motor 6 are stepper motor.
(2) image to obtaining selects the normalization color space to be used for cluster segmentation, changes the influence to segmentation result to eliminate natural lighting; On the basis of color quantizing, utilize fuzzy C-means clustering algorithm (FCM) to carry out image segmentation, with background separation such as plant leaf blade image and soil; To cutting apart the back image, utilize the mathematical morphology computing to handle, remove noise; To image after the denoising, utilization element marking algorithm tag image also carries out BLOB and analyzes, and removes weeds according to the BLOB analysis result, fills the blade hole, and all leaf images that extract in the scene are used for subsequent treatment.
(3) to the plant leaf blade image, from aspects such as form, color and texture structures quantity enough big coerce characteristic set; To each plant leaf blade that obtains, extract the girth, the centre of form, area of blade, highly, 13 morphological features such as width, inner ellipse major axis, inner ellipse minor axis, area girth ratio, degree of compacting, length growth rate, length breadth ratio logarithm, girth width compare, perimeter length compares.Color provides abundant information to identification of targets, and color is described as a three-dimensional vector usually, and promptly each color is a coordinate in the color space.Obtaining view data is rgb format, and HSI meets human eye perception psychology, and IE (L*a*b*) is that even color space can be measured little aberration with Euclidean distance.The average and the gray average that obtain R, G, B, H, S, L, a*, each color component of b* are as color characteristic.Texture is estimating of reflection target surface brightness variation, like smoothness, roughness, rule degree etc.The two dimensional gray co-occurrence matrix is usually used in texture analysis, because the space distribution relation of gray-scale value in its ability statistical picture.Such as entropy is used for describing the randomness of intensity profile, and when crop subalimentation, surperficial complexity reduces, and entropy also can reduce; Energy is index of reflection gray scale intensities, when lack of water, and yellow leaf, brightness increases, and energy also increases, and contrast improves.Compare with gray level image, coloured image provides more color characteristic in visible spectrum, and in view of the above, this project is intended and utilized colored co-occurrence matrix to carry out the color texture analysis.Each component for R, G, B, H, S, I, L*, a*, b*; According to distance is that 1 angle is 0 spatial relationship calculating co-occurrence matrix; On this basis; Calculating energy (Energy), entropy (Entropy), contrast (Contrast), homogeneity (Homogeneity), reciprocal difference square (Inverse Difference Moment, IDM), simple crosscorrelation (Correlation), average and (Sum Mean), variance (Variance), type trend (Cluster Tendency), maximum probability (Maximum Probability); Each component calculates 10 Haralick textural characteristics, totally 90 characteristics.
(4), utilize heuristic search algorithm to combine neural network to select water, fertile state optimization character subset to the characteristic set of structure.
(5) according to water, the fertile state optimization character subset selected, utilize the BP neural network that crop is coerced characteristic and discern.
Among the present invention, the described Image Acquisition of step (1) is specially, and computing machine sends signal and gives controllor for step-by-step motor 4, the motion of control directions X stepper motor 5 and Y direction stepper motor 6, and then mobile CCD camera is to assigned address, images acquired above the crop.The structural advantage of the present invention be through the horizontal positioning system mobile camera to assigned address, can comprehensively obtain the crop map picture, reduce the installation quantity of video camera.Another innovative point is to image processing algorithm problem consuming time on the hardware, utilizes CUDA (Compute Unified Device Architecture, unified calculation equipment framework) to carry out the Flame Image Process computing, to improve arithmetic speed.CUDA is the software and hardware solution that is improved the video card image-capable and be suitable for parallel computation by being intended to of releasing of NVIDIA (tall and handsome reaching), through utilizing the processing power of GPU (graphics processing unit), can significantly promote calculated performance.Reached millions ofly based on GPU (video card) sales volume of the tall and handsome CUDA of reaching, software developer, scientist and researchist be the tall and handsome CUDA that reaches of utilization in every field just, comprises image and Video processing, calculation biology and chemistry, fluid mechanics simulation etc.Therefore utilize CUDA to improve the arithmetic speed of FCM algorithm, easy popularization with low cost.
Among the present invention; It is that the image of n is divided into
individual sub- color space in rgb space with pixel quantity that the said normalization color space of step (2) adopts based on the split plot design of minimum color variance; The pixel number of each subspace is
; Satisfy following relation
Q takes off formula as quantized values.
The FCM algorithm is with image data set
is divided into
type; The degree of membership of arbitrary sample
class to
is
in
, and classification results can be used a fuzzy membership matrix
expression.In the standard FC M algorithm; Each pixel
all will be participated in the interative computation of FCM algorithm; Operand is big; And after carrying out color quantizing, the representative color and the number of colors of each subspace are known, therefore only need to carry out computing with representative color; Do not need repetitive operation, can significantly improve arithmetic speed.Graphical representation after the quantification is (Q; H); Color sub-spaces
, the quantity that each subspace representative color is corresponding is
.Degree of membership matrix notation based on the FCM algorithm classification result of color quantizing is
; Fuzzy C-Means Clustering is to realize by the object function
that minimizes about degree of membership matrix U and cluster centre V
Behind the color quantizing, identical color does not need double counting, and objective function is calculated as follows (4)
In
formula; The computing formula of degree of membership matrix
adopts (5) formula; Dimension after the quantification is that c * n becomes c * q; Calculated amount greatly reduces,
(5)
Wherein
(6)
Behind the color quantizing, center calculation adopts following formula,
(7)
Because q is much smaller than the n event, the central points amount also significantly reduces.
In the FCM algorithm of the present invention,, more even in the distribution of rgb space because view data has been passed through pre-service; No longer assemble,, need not transform to the even color space of CIE so can directly utilize the RGB numerical value of image slices vegetarian refreshments to carry out cluster along near the diagonal line; Can reduce calculated amount; And do not influence the cluster effect, behind the color quantizing, significantly improve arithmetic speed not influencing under the prerequisite of cutting apart quality.
Among the present invention, step (3) Haralick textural characteristics is defined as,
Energy,
(8)
Entropy,
(9)
Contrast,
(10)
Homogeneity,
(11)
The reciprocal difference square,
(12)
Simple crosscorrelation,
(13)
Average with,
(14)
Variance,
(15)
Class trend,
(16)
Maximum probability,
(17)
P [i; J] be (i of co-occurrence matrix; J) individual element,
is respectively the average and the standard deviation of co-occurrence matrix.
Among the present invention, step (4) is chosen optimal feature subset from characteristic set method is to adopt heuristic search algorithm such as genetic algorithm, simulated annealing, ant group algorithm or particle cluster algorithm to seek the optimal characteristics combination; The objective function of algorithm is the minimum identification error of neural network; After the algorithm convergence, add up the frequency of utilization of each characteristic, select the most frequently used characteristic as the strongest characteristic of recognition capability; And the structural attitude subclass, the characteristic quantity of subclass is controlled at 20-30.When algorithm was selected characteristic, for untapped characteristic, the input of neural network was made as 0.After optimizing, the quantity of characteristic set descends, and so, can reduce the neural network computation complexity, improve generalization ability.
Among the present invention, step (5) adopts the characteristic set of BP neural network utilization optimization that the liquid manure state of crop is discerned.Neural network can go out regular knowledge from the extracting data of input, output when training, remember in the weights of network, and have generalization ability, and therefore selecting neural network to be used for plant stress characteristic feature identification is a kind of natural selection.Be used for the neural network of feature selecting and the neural network that is used for feature identification of reaching the standard grade at last all adopts three layers of BP (error back propagation) neural network, both difference are that network input layer and hidden neuron quantity are inconsistent.The latter's input feature value is the character subset after optimizing, and scale of neural network is little many than the former.Input layer is the characteristic quantity that is used to discern, and output layer is lack of water, nutritional deficiency, characteristic quantity such as normal, and the neuronal quantity in middle layer rule of thumb is worth definite.
Beneficial effect of the present invention is: the present invention is a research object with the crop in the greenhouse; Structure computer vision monitoring platform; Research adapts to plant image dividing method under natural lighting variation, the complex scene; To the plant leaf blade image that obtains; Extract and the enough characteristic sets of structure quantity from aspects such as form, color, textures, adopt heuristic search algorithm such as genetic algorithm, simulated annealing, ant group algorithm, particle cluster algorithm to combine nerual network technique to seek optimal feature subset, utilize the BP neural network that crop is coerced characteristic at last and discern.Adopt the horizontal positioning system mobile camera, can comprehensively obtain plant image.For computing algorithm consuming time, adopt CUDA (Compute Unified Device Architecture) hardware platform to realize, to satisfy the real-time requirement of monitoring.The present invention coerces destructive measurement of state expense for the liquid manure of hothouse plants a kind of technological means is provided, and has broad application prospects.
Description of drawings
Fig. 1 is a monitoring platform structural diagrams of the present invention.
Fig. 2 is a computer organization of the present invention.
Fig. 3 is the inventive method process flow diagram.
Fig. 4 is an image processing process.Wherein: (a) former figure, (b) image segmentation, (c) mathematical morphology computing, (d) Blob filters, and (e) hole is filled, (f) abstract image.
Label among the figure: 1 is the CCD camera, and 2 is image pick-up card, and 3 for having the computing machine of CUDA platform, and 4 is the stepper controller, and 5 is the directions X stepper motor, and 6 is Y direction stepper motor, and 7 is the directions X guide rail, and 8 is the Y traversing guide, and 9 is support, and 10 is chamber crop.
Embodiment
In order better to understand the present invention, the present invention is elaborated below in conjunction with accompanying drawing.
Embodiment 1:
As shown in Figure 1, computing machine adopts at interval according to preset image, sends instruction; Motor through controllor for step-by-step motor control X, Y both direction; Video camera (CCD camera) is moved to assigned address, and images acquired then is stored in and is used for graphical analysis in the computing machine.Camera is installed in the crop top among Fig. 1, helps obtaining the plant leaf blade image.Fig. 2 is the computer structure composition, and computer has the CUDA hardware platform, to improve the Flame Image Process arithmetic speed.
Fig. 3 is an image analysis algorithm flow process process flow diagram, and Fig. 4 is the image segmentation process, and each step of algorithm process is following:
(1) like Fig. 3; After obtaining image, carry out pre-service earlier, i.e. degree of comparing adjustment and decorrelation stretching conversion (Decorrelation Stretch Transformation; DST); So that view data being more evenly distributed in rgb space, and then the dynamic range of raising image, handle back FCM cluster and can directly in rgb space, carry out.
(2) after the image pre-service, carry out color quantizing, utilize fuzzy C-means clustering algorithm (FCM) to carry out image segmentation then, soil is separated with plant.Based on the FCM algorithm of color quantizing, as previously mentioned.Fig. 4 (a) is a plant image, and Fig. 4 (b) is the result after cutting apart based on the FCM of color quantizing.
(3) to cutting apart the back image, utilize the mathematical morphology computing to handle, remove noise, Fig. 4 (b) is carried out the morphology denoising after the result shown in Fig. 4 (c).
(4) to image after the denoising; Utilization element marking algorithm tag image also carries out BLOB and analyzes; Remove weeds, big block (like Fig. 4 (d)) according to the BLOB analysis result; Fill blade hole (like Fig. 4 (e)), all leaf images (shown in Fig. 4 (f)) that extract in the scene are used for subsequent treatment.
(5) each plant leaf blade to obtaining, the average and the gray average that obtain R, G, B, H, S, L, a*, each color component of b* are as color characteristic.
(6) extract the girth, the centre of form, area of blade, highly, 13 morphological features such as width, inner ellipse major axis, inner ellipse minor axis, area girth ratio, degree of compacting, length growth rate, length breadth ratio logarithm, girth width compare, perimeter length compares.
(7) confirm suitable spatial relationship (angle is 0, and distance is 1),, calculate co-occurrence matrix, extract aforementioned 10 Haralick textural characteristics on this basis, amount to 90 characteristics for each component of R, G, B, H, S, I, L*, a*, b*.
(8) constructing neural network is coerced feature identification, utilizes heuristic search algorithm such as adopting genetic algorithm, simulated annealing, ant group algorithm, particle cluster algorithm to seek optimal feature subset.
(9) utilize optimal feature subset, structure BP neural network is utilized the training of characteristic feature picture, and thing was coerced the ONLINE RECOGNITION of eigenstate under the realization complex scene was done.
(10) for computing algorithm consuming time, the flat algorithm framework (structure is as shown in Figure 2) of research and designs C UDA hardware improves arithmetic speed, requirement of real time.Reached millions ofly based on GPU (video card) sales volume of the tall and handsome CUDA of reaching, software developer, scientist and researchist be the tall and handsome CUDA that reaches of utilization in every field just, comprises image and Video processing, calculation biology and chemistry, fluid mechanics simulation etc.Therefore utilize CUDA to improve the arithmetic speed of algorithm, not only method is feasible and with low cost.
Claims (5)
1. the chamber crop based on computer vision is coerced state identification method, it is characterized in that concrete steps are following:
(1) makes up the computer vision monitoring platform; Said monitoring platform is formed the camera horizontal positioning system by CCD camera (1), support (9), computing machine (3), controllor for step-by-step motor (4), directions X motor (5), Y direction motor (6), directions X guide rail (7) and Y traversing guide (8); Wherein: guide rail (8) is positioned at both sides before and after the top, greenhouse, and guide rail (7) two ends are positioned on the guide rail (8); Y direction motor (6) control directions X guide rail (7) is the motion of Y direction in guide rail (8) upper edge; CCD camera (1) is installed on the directions X guide rail (7) through support (9), and directions X moves around directions X motor (5) control support (9) in guide rail (7) upper edge, and CCD camera (1) is positioned at the chamber crop top; CCD camera (1) is through image pick-up card (2) the even input end of computing machine (3), and the output terminal of computing machine (3) connects the input end of controllor for step-by-step motor (4), and the output terminal of controllor for step-by-step motor (4) connects directions X motor (5) and Y direction motor (6) respectively; Computing machine is realized the horizontal location of CCD camera above the greenhouse through sending instruction to controllor for step-by-step motor (4), gathers the crop map picture successively; Computing machine has the CUDA hardware platform, is used to realize the complicated image processing operations; Motor (5), motor (6) are stepper motor;
(2) image to obtaining selects the normalization color space to be used for cluster segmentation, changes the influence to segmentation result to eliminate natural lighting; On the basis of color quantizing, utilize the fuzzy C-means clustering algorithm to carry out image segmentation, with background separation such as plant leaf blade image and soil; To cutting apart the back image, utilize the mathematical morphology computing to handle, remove noise; To image after the denoising, utilization element marking algorithm tag image also carries out BLOB and analyzes, and removes weeds according to the BLOB analysis result, fills the blade hole, and all leaf images that extract in the scene are used for subsequent treatment;
(3) to the plant leaf blade image, from aspects such as form, color and texture structures quantity enough big coerce characteristic set; To each plant leaf blade that obtains; Extract the girth, the centre of form, area of blade, highly, width, inner ellipse major axis, inner ellipse minor axis, area girth ratio, degree of compacting, length growth rate, length breadth ratio logarithm, girth width compare, perimeter length is than 13 morphological features; Color is described as a three-dimensional vector usually, and promptly each color is a coordinate in the color space; Obtaining view data is rgb format, and HSI meets human eye perception psychology, and CIE (L*a*b*) is that even color space can be measured little aberration with Euclidean distance; The average and the gray average that obtain R, G, B, H, S, L, a*, each color component of b* are as color characteristic; Texture is estimating of reflection target surface brightness variation; The two dimensional gray co-occurrence matrix is used for texture analysis, and entropy is used for describing the randomness of intensity profile, and when crop subalimentation, surperficial complexity reduces, and entropy also can reduce; Energy is index of reflection gray scale intensities, when lack of water, and yellow leaf, brightness increases, and energy also increases, and contrast improves; Compare with gray level image, coloured image provides more color characteristic in visible spectrum, and in view of the above, this project is intended and utilized colored co-occurrence matrix to carry out the color texture analysis; Each component for R, G, B, H, S, I, L*, a*, b*; According to distance is that 1 angle is 0 spatial relationship calculating co-occurrence matrix; Calculating energy, entropy, contrast, homogeneity, reciprocal difference square, simple crosscorrelation, average and, variance, type trend, maximum probability and each component calculate 10 Haralick textural characteristics, totally 90 characteristics;
(4), utilize heuristic search algorithm to combine neural network to select water, fertile state optimization character subset to the characteristic set of structure;
(5) according to water, the fertile state optimization character subset selected, utilize the BP neural network that crop is coerced characteristic and discern.
2. the chamber crop based on computer vision according to claim 1 is coerced state identification method; It is characterized in that it is that the image of n is divided into
individual sub- color space in rgb space with pixel quantity that the said normalization color space of step (2) adopts based on the split plot design of minimum color variance; The pixel number of each subspace is
; Satisfy following relation
Q takes off formula as quantized values,
FCM algorithm to image data sets
into
class,
any sample
on
class membership degree
, classification results using a fuzzy membership matrix
representation; standard FCM algorithm, each pixel
to be involved in FCM algorithm iteration, computing capacity, and for color quantization, each sub- indicates the color space and color number is known, only that represent the color operation, a substantial increase in operation speed; quantized image is expressed as (Q, H), the color sub-space
, on behalf of each sub-space corresponding to the color The number is
; FCM algorithm based on color classification results quantify the membership degree matrix is represented as
, Fuzzy C-Means clustering is by minimizing About membership matrix U and cluster center V The objective function
achieved,
Behind the color quantizing; Identical color does not need double counting, and objective function is calculated as follows
(4)
In the formula; The computing formula of degree of membership matrix
adopts (5) formula; Dimension after the quantification is that c * n becomes c * q; Calculated amount greatly reduces,
(5)
Wherein
Behind the color quantizing, center calculation adopts following formula,
(7)
Because q is much smaller than n, the central points amount also significantly reduces;
3. the chamber crop based on computer vision according to claim 1 is coerced state identification method, it is characterized in that step (3) Haralick textural characteristics is defined as,
Energy,
(8)
Entropy,
(9)
Contrast,
(10)
Homogeneity,
(11)
The reciprocal difference square,
(12)
Simple crosscorrelation,
(13)
Average with,
(14)
Variance,
(15)
Class trend,
(16)
Maximum probability,
(17)
P [i; J] be (i of co-occurrence matrix; J) individual element,
is respectively the average and the standard deviation of co-occurrence matrix.
4. the chamber crop liquid manure based on computer vision according to claim 1 is coerced state identification method; It is characterized in that step (4) chooses the method for optimal feature subset and be from characteristic set: adopt heuristic search algorithm such as genetic algorithm, simulated annealing, ant group algorithm or particle cluster algorithm to seek the optimal characteristics combination; The objective function of algorithm is the minimum identification error of neural network; After the algorithm convergence, add up the frequency of utilization of each characteristic, select the most frequently used characteristic as the strongest characteristic of recognition capability; And the structural attitude subclass, the characteristic quantity of subclass is controlled at 20-30; When algorithm was selected characteristic, for untapped characteristic, the input of neural network was made as 0.
5. the chamber crop based on computer vision according to claim 1 is coerced state identification method, it is characterized in that step (5) adopts the characteristic set of BP neural network utilization optimization that the liquid manure state of crop is discerned; Neural network can go out regular knowledge from the extracting data of input, output when training, remember in the weights of network, and have generalization ability, and therefore selecting neural network to be used for plant stress characteristic feature identification is a kind of natural selection; Be used for the neural network of feature selecting and the neural network that is used for feature identification of reaching the standard grade at last all adopts three layers of BP neural network, both difference are that network input layer and hidden neuron quantity are inconsistent; The latter's input feature value is the character subset after optimizing, and scale of neural network is little many than the former; Input layer is the characteristic quantity that is used to discern, and output layer is lack of water, nutritional deficiency or normal characteristic quantity, and the neuronal quantity in middle layer rule of thumb is worth definite.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210260259.0A CN102789579B (en) | 2012-07-26 | 2012-07-26 | Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210260259.0A CN102789579B (en) | 2012-07-26 | 2012-07-26 | Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102789579A true CN102789579A (en) | 2012-11-21 |
| CN102789579B CN102789579B (en) | 2015-06-03 |
Family
ID=47154980
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210260259.0A Expired - Fee Related CN102789579B (en) | 2012-07-26 | 2012-07-26 | Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102789579B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103048266A (en) * | 2012-12-11 | 2013-04-17 | 江苏大学 | Automatic recognizing method and device for nitrogen phosphorus and potassium stress of protected tomatoes |
| CN103105246A (en) * | 2012-12-31 | 2013-05-15 | 北京京鹏环球科技股份有限公司 | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm |
| CN103487374A (en) * | 2013-10-14 | 2014-01-01 | 无锡艾科瑞思产品设计与研究有限公司 | Machine-vision-based device and method for qualitatively and rapidly detecting clenbuterol |
| CN103493794A (en) * | 2013-09-26 | 2014-01-08 | 北京农业信息技术研究中心 | Seedbed monitoring management system and method |
| CN104298987A (en) * | 2014-10-09 | 2015-01-21 | 西安电子科技大学 | Handwritten numeral recognition method based on point density weighting online FCM clustering |
| CN105588930A (en) * | 2015-12-17 | 2016-05-18 | 镇江市高等专科学校 | Method for measuring parameters of soil in greenhouse |
| CN106097372A (en) * | 2016-06-23 | 2016-11-09 | 北京农业信息技术研究中心 | Crop plant water stress Phenotypic examination method based on image procossing |
| CN106718363A (en) * | 2017-01-06 | 2017-05-31 | 安徽农业大学 | A kind of irrigation tests method and its test platform towards precision agriculture |
| CN107622236A (en) * | 2017-09-15 | 2018-01-23 | 安徽农业大学 | Based on bee colony and gradient lifting decision Tree algorithms crops disease diagnosing method for early warning |
| CN107730504A (en) * | 2017-10-17 | 2018-02-23 | 太原理工大学 | Image partition method based on improved BP |
| CN107766938A (en) * | 2017-09-25 | 2018-03-06 | 南京律智诚专利技术开发有限公司 | A kind of plant cover cultivation methods based on BP neural network |
| CN108596216A (en) * | 2018-04-04 | 2018-09-28 | 格薪源生物质燃料有限公司 | Biomass fuel quality determining method and system |
| CN109699271A (en) * | 2018-12-24 | 2019-05-03 | 柳州铁道职业技术学院 | System and its control method are applied in water-fertilizer precision filling in tea place |
| CN110021177A (en) * | 2019-05-06 | 2019-07-16 | 中国科学院自动化研究所 | Heuristic random searching traffic lights timing designing method, system |
| CN112857440A (en) * | 2021-01-08 | 2021-05-28 | 成都农业科技职业学院 | Intelligent agricultural greenhouse control system and control method |
| CN113966714A (en) * | 2021-10-27 | 2022-01-25 | 山东润浩水利科技有限公司 | Fertilizing device and fertilizing method for automatic field irrigation |
| CN116519688A (en) * | 2023-04-26 | 2023-08-01 | 中国科学院植物研究所 | High-throughput acquisition and automatic analysis method and system for berry phenotype characteristics |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1813675A1 (en) * | 2006-01-31 | 2007-08-01 | Pioneer Hi-Bred International, Inc. | Method for high throughput transgene function analysis for agronomic traits in maize |
| CN102334422A (en) * | 2010-07-27 | 2012-02-01 | 中国农业科学院蔬菜花卉研究所 | Machine vision based real-time diagnosis method and system of vegetable leaf diseases |
| CN102506938A (en) * | 2011-11-17 | 2012-06-20 | 江苏大学 | Detecting method for greenhouse crop growth information and environment information based on multi-sensor information |
-
2012
- 2012-07-26 CN CN201210260259.0A patent/CN102789579B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1813675A1 (en) * | 2006-01-31 | 2007-08-01 | Pioneer Hi-Bred International, Inc. | Method for high throughput transgene function analysis for agronomic traits in maize |
| CN102334422A (en) * | 2010-07-27 | 2012-02-01 | 中国农业科学院蔬菜花卉研究所 | Machine vision based real-time diagnosis method and system of vegetable leaf diseases |
| CN102506938A (en) * | 2011-11-17 | 2012-06-20 | 江苏大学 | Detecting method for greenhouse crop growth information and environment information based on multi-sensor information |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103048266B (en) * | 2012-12-11 | 2015-06-10 | 江苏大学 | Automatic recognizing method and device for nitrogen phosphorus and potassium stress of protected tomatoes |
| CN103048266A (en) * | 2012-12-11 | 2013-04-17 | 江苏大学 | Automatic recognizing method and device for nitrogen phosphorus and potassium stress of protected tomatoes |
| CN103105246A (en) * | 2012-12-31 | 2013-05-15 | 北京京鹏环球科技股份有限公司 | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm |
| CN103493794A (en) * | 2013-09-26 | 2014-01-08 | 北京农业信息技术研究中心 | Seedbed monitoring management system and method |
| CN103493794B (en) * | 2013-09-26 | 2015-04-29 | 北京农业信息技术研究中心 | Seedbed monitoring management system and method |
| CN103487374B (en) * | 2013-10-14 | 2016-03-30 | 无锡艾科瑞思产品设计与研究有限公司 | The qualitative device for fast detecting of clenbuterol hydrochloride based on machine vision and method |
| CN103487374A (en) * | 2013-10-14 | 2014-01-01 | 无锡艾科瑞思产品设计与研究有限公司 | Machine-vision-based device and method for qualitatively and rapidly detecting clenbuterol |
| CN104298987A (en) * | 2014-10-09 | 2015-01-21 | 西安电子科技大学 | Handwritten numeral recognition method based on point density weighting online FCM clustering |
| CN104298987B (en) * | 2014-10-09 | 2017-07-25 | 西安电子科技大学 | The Handwritten Digit Recognition method of online FCM clusters is weighted based on dot density |
| CN105588930A (en) * | 2015-12-17 | 2016-05-18 | 镇江市高等专科学校 | Method for measuring parameters of soil in greenhouse |
| CN106097372A (en) * | 2016-06-23 | 2016-11-09 | 北京农业信息技术研究中心 | Crop plant water stress Phenotypic examination method based on image procossing |
| CN106718363B (en) * | 2017-01-06 | 2022-06-28 | 安徽农业大学 | Irrigation test method and test platform for fine agriculture |
| CN106718363A (en) * | 2017-01-06 | 2017-05-31 | 安徽农业大学 | A kind of irrigation tests method and its test platform towards precision agriculture |
| CN107622236A (en) * | 2017-09-15 | 2018-01-23 | 安徽农业大学 | Based on bee colony and gradient lifting decision Tree algorithms crops disease diagnosing method for early warning |
| CN107622236B (en) * | 2017-09-15 | 2020-12-04 | 安徽农业大学 | Crop disease diagnosis and early warning method based on swarm and gradient lifting decision tree algorithm |
| CN107766938A (en) * | 2017-09-25 | 2018-03-06 | 南京律智诚专利技术开发有限公司 | A kind of plant cover cultivation methods based on BP neural network |
| CN107730504A (en) * | 2017-10-17 | 2018-02-23 | 太原理工大学 | Image partition method based on improved BP |
| CN108596216A (en) * | 2018-04-04 | 2018-09-28 | 格薪源生物质燃料有限公司 | Biomass fuel quality determining method and system |
| CN109699271A (en) * | 2018-12-24 | 2019-05-03 | 柳州铁道职业技术学院 | System and its control method are applied in water-fertilizer precision filling in tea place |
| CN110021177A (en) * | 2019-05-06 | 2019-07-16 | 中国科学院自动化研究所 | Heuristic random searching traffic lights timing designing method, system |
| CN110021177B (en) * | 2019-05-06 | 2020-08-11 | 中国科学院自动化研究所 | Heuristic random search traffic signal lamp timing optimization method and system |
| CN112857440A (en) * | 2021-01-08 | 2021-05-28 | 成都农业科技职业学院 | Intelligent agricultural greenhouse control system and control method |
| CN113966714A (en) * | 2021-10-27 | 2022-01-25 | 山东润浩水利科技有限公司 | Fertilizing device and fertilizing method for automatic field irrigation |
| CN113966714B (en) * | 2021-10-27 | 2022-12-06 | 山东润浩水利科技有限公司 | Fertilizing device and fertilizing method for automatic field irrigation |
| CN116519688A (en) * | 2023-04-26 | 2023-08-01 | 中国科学院植物研究所 | High-throughput acquisition and automatic analysis method and system for berry phenotype characteristics |
| CN116519688B (en) * | 2023-04-26 | 2024-05-14 | 中国科学院植物研究所 | High-throughput acquisition and automatic analysis method and system for berry phenotype characteristics |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102789579B (en) | 2015-06-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102789579B (en) | Identification method for stressed state of water fertilizer of greenhouse crop on basis of computer vision technology | |
| CN102564593B (en) | Plant growth condition monitoring system based on compute vision and internet of things | |
| Fan et al. | The future of Internet of Things in agriculture: Plant high-throughput phenotypic platform | |
| WO2022253057A1 (en) | Intelligent moisture precise irrigation control system and method for fruit and vegetable cultivation in solar greenhouse | |
| Zhou et al. | Diagnosis of winter-wheat water stress based on UAV-borne multispectral image texture and vegetation indices | |
| CN102524024B (en) | Crop irrigation system based on computer vision | |
| Tian et al. | Application status and challenges of machine vision in plant factory—A review | |
| Kaneda et al. | Multi-modal sliding window-based support vector regression for predicting plant water stress | |
| CN110163138B (en) | Method for measuring and calculating wheat tillering density based on multispectral remote sensing image of unmanned aerial vehicle | |
| Lin et al. | Intelligent greenhouse system based on remote sensing images and machine learning promotes the efficiency of agricultural economic growth | |
| Lin et al. | A review on computer vision technologies applied in greenhouse plant stress detection | |
| Liang et al. | Segmentation and weight prediction of grape ear based on SFNet-ResNet18 | |
| CN113158750A (en) | Self-feedback learning evaluation method of plant growth model based on convolutional neural network | |
| Nadafzadeh et al. | Design and fabrication of an intelligent control system for determination of watering time for turfgrass plant using computer vision system and artificial neural network | |
| Pandey et al. | Smart agriculture: Technological advancements on agriculture—A systematical review | |
| Miao et al. | Crop weed identification system based on convolutional neural network | |
| Lin et al. | Data-driven modeling for crop growth in plant factories | |
| Ambildhuke et al. | IoT based Portable Weather Station for Irrigation Management using Real-Time Parameters | |
| Islam et al. | IoT-Smart Agriculture: Comparative Study on Farming Applications and Disease Prediction of Apple Crop using Machine Learning | |
| Sahu et al. | A Study on Weather based Crop Prediction System using Big Data Analytics and Machine Learning | |
| Raval et al. | Computer vision and machine learning in agriculture | |
| Li et al. | A longan yield estimation approach based on uav images and deep learning | |
| Mangla et al. | Statistical growth prediction analysis of rice crop with pixel-based mapping technique | |
| Muruganandam et al. | IoT Based Agriculture Monitoring and Prediction of Paddy Growth using Enhanced Conquer Based Transitive Clustering | |
| Liyanage et al. | Sustainable Growth Through Automation: Machine Learning and Computer Vision Advancements in Sri Lankan Floriculture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150603 Termination date: 20180726 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |