CN102524024B - Crop irrigation system based on computer vision - Google Patents

Abstract

The invention relates to the image processing technology and the water-saving irrigation field and discloses a crop irrigation method based on the computer vision. The crop irrigation method adopts the technology of nondestructive crop water lack detection based on the computer vision, in addition, a crop growth data model is combined, finally, the automatic judgment is made, and an electromagnetic valve is controlled for completing the crop irrigation. According to a system, the steps such as crop image obtaining, color space conversion, image dividing and characteristic parameter extraction are firstly carried out, then, the characteristic parameter calculation is carried out through being combined with the crop growth data model, the judgment is completed through a genetic neural network, and finally, a single chip microcomputer is controlled to drive the electromagnetic valve for completing the crop irrigation according to the judgment results. The system provided by the invention can obtain the water demanding information of crops in time and can realize the in-time and accurate irrigation, and the water resource utilization rate is perfectly improved. The system has good practicability and can be applied to crop planting regions such as greenhouses, farmlands and nursery gardens.

Description

Crop irrigation system based on computer vision

Technical field

The present invention relates to computer vision and treatment technology and water-saving irrigation field, particularly relate to a kind of based on Computer Vision Detection crop pattern, change color and the water-saving irrigation method that judges and control.

Background technology

Computer vision is Applied Computer Techniques simulation people's vision mechanism, the things in the obtained visual pattern is carried out the science and technology of discriminator.It is the cross discipline of the collection, image processing and the subjects such as identification, pattern-recognition that integrate image.Along with the computer hardware technique development, computer vision has progressed into practical application.

Development along with science and technology, agricultural production is also promptly changing traditional farming pattern, modern science and industrial technology are penetrated in the agricultural production gradually, agrotechnique is gradually to scientific, informationalized future development, the modern installations agricultural is arisen at the historic moment, and becomes one of most active industry in the world today.

The utilization of agricultural water resources has availability low always, the shortage of water resource and waste and the phenomenon of depositing.Effective utilization of water resource is the subject matter that the China irrigational agriculture development faces, using water wisely, Rational Irrigation, Developing Water-saving Agriculture, be one with strategic problem, and one of effective way that addresses this problem is exactly the degree that wanes by monitoring crop moisture and judges exactly.

All be that the environmental parameters such as environmental temperature of humidity, periphery with soil are as the control parameter of irrigation system to the irrigation system majority of crop at present, can not truly reflect the exsiccosis of crop, therefore control accuracy is low, has caused the waste of the low and water resource of the availability of water resource.

Now existing researcher is by the LVDT(differential transformer type) displacement transducer measures corn, citrus etc., obtained certain water-saving result.But LVDT is used in the mechanical manufacturing field industry measurement, although Measurement Resolution and precision can meet the demands, because contact point has rigidity, can affect the growth of crop, also can't well satisfy the actual requirement that crop is measured simultaneously.Also there is the scholar to utilize sensor or micrometer that fruit or the leaf of crop are measured, but these methods all need to carry out the contact operation with fruit or the blade of crop, the trace of measuring plant organ by sensor or mechanical device changes, need to obtain water information to control irrigation system, these methods are owing to need execute-in-place, waste time and energy, and owing to be that contact is measured, also the growth of crop had certain impact.Current, also there is the scholar to be engaged in abroad and utilizes in-plant infrared remote sensing image to detect the crop water situation, the technique cost is very expensive, and does not reach practical degree.

Computer vision technique is a kind of nondestructive measurement means, is one of a kind of advanced technology means in the plant growth diagnosis research.Crop growth status is one of important content in the crop management decision-making, and the research of this respect is at the early-stage.

The technology used in the present invention is the nondestructive crop water shortage detection based on computer vision, in conjunction with the plant growth data model, judges that also the control magnetic valve is finished the irrigation of crop automatically.

The growth change of the variation of plant growth height, the blade of crop and the water regime of crop have very large correlation, and the tender shoots of crop, the pucker ﹠ bloat of stem of plant etc. also have very strong correlation with the moisture of crop simultaneously.Adopt computer vision diagnosis crop water can realize the target that, continuous monitoring record not disruptive to crop plant, Obtaining Accurate are made moisture information in the object and in time irrigated.

Summary of the invention

The purpose of this invention is to provide a kind of method of irrigating based on the crop intelligent water-saving of computer vision technique, it can realize non-cpntact measurement, does not affect plant growth, and reaches the purpose of water-saving irrigation.

Deficiency for the traditional irrigation mode, the crop intelligent irrigation system based on computer vision that the present invention proposes, at first utilize image acquisition equipment, obtain the crop map picture, and by calculator image is processed, finish the color space conversion of image, cutting apart and feature extraction of image, obtain the stem, leaf of crop, feature and the change color of flower, in conjunction with plant growth data model and genetic neural network, calculate and judge the need water information of crop, remove to control magnetic valve by single-chip microcomputer and irrigate.System has realized the Based Intelligent Control of irrigating is had great importance to the availability that improves water resource.

The crop irrigation system that the present invention is based on computer vision is comprised of image acquisition equipment, calculator, single-chip microcomputer and irrigation system, and system realizes comprising following concrete steps:

(1) obtains the growth original image of crop;

(2) original image is carried out preliminary treatment;

(3) respectively image is carried out binary conversion treatment and from the RGB color space conversion to the HSI color space, and choose H in the HSI color space and S parameter as the color characteristic of crop map picture, in the crop map picture of HSI color space, select the part pixel as seed;

(4) seed region is grown, and will be attached to the neighbor of seed color similar performance on the seed of growth district, and a plurality of sub-blocks of entire image are scanned, and to close on color, zone adjacent on the space merges;

(5) finish the extraction with the plant growth characteristic parameter cut apart of crop, concrete plant growth characteristic parameter has: the color parameter of plant height parameter, leaf area parameter and cauline leaf;

(6) various characteristic parameters and plant growth data model are carried out data comparison and calculating, find out difference value;

(7) calculate and adjudicate based on genetic neural network;

(8) according to the computer decision result, by the Single-chip Controlling magnetic valve, finish the irrigation of crop.

The present invention has set up the plant growth data model, by experiment and the detection to crop, leaf, stem, floral morphology and the change color of crop, the growing height of crop etc. have been studied to the phenomenon of appearance that lack of water shows, find out crop to the rule of water deficit reaction, set up the relation between crop pattern variation and the lack of moisture, thereby formulated corresponding irrigation control index.

System hardware of the present invention mainly is comprised of image acquisition equipment, calculator, single-chip microcomputer and irrigation system, Computer Image Processing of the present invention mainly is made of steps such as image preliminary treatment, image binaryzation, image HSI color notation conversion space, image segmentation, extracted region and calculating.Image is processed the extraction of characteristic parameter of the growth conditions mainly be the color characteristic of the leaf of finishing crop, flower and crop, extracts color and morphological feature parameter and irrigates needed index as judging.In conjunction with the plant growth data model characteristic parameter that obtains is calculated and analyzes again, obtain judging whether lack of water corresponding index of crop, at last by genetic neural network, provide the decision-making output of irrigation.

Crops are in process of growth, and the size of plant forms, plant height, blade and flower shape and the variation of the aspects such as color, fruit shape all are closely related with water demand and the supply of crop.The present invention has set up rational various crop growth characteristics parameter as irrigation index, and major parameter has:

(1) plant height parameter.The growth rate of crop and the moisture content of crop need to have very large relation, and the photosynthesis that how much affects crop of moisture affects absorption and the transhipment of nutriment in the crop, and plant height then is the important parameter of plant growth.

(2) leaf area parameter.It also is a very directly perceived and important feature of plant growth that blade changes, and the shoot and leaf growth of crop is very sensitive to water deficit, during more slight lack of water the growth of leaf just suppressed, blade expansion growth stops ahead of time.

(3) color parameter of cauline leaf.Plant growth is slow when lack of water, and chlorophyll concentration increases relatively, and the leaf look deepens, the cauline leaf color reddens, carbohydrate breakdown is greater than synthetic during the reflection crop drought, and more soluble sugar and the conversion of accumulation forms anthocyanidin in the cell, thereby causes the variation of cauline leaf color.

The present invention has adopted the HSI color space to come the color of crop is differentiated.Usually the image information of obtaining from image acquisition equipment is by the RGB representation in components, and there is no obvious rule, the distribution of the rgb value in the crop map picture can follow, directly utilize these components often can not obtain required effect, be unfavorable for directly as the recognition feature parameter.And the advantage of HSI color space be it with brightness (Intensity) and reflection color intrinsic propesties two parameters---colourity (Hue) and degree of saturation (Saturation) are separated, Color Expression is closer to the observation of human eye, and light is little on the impact of identification.

In the extraction to the crop Color characteristics parameters, choose the H relevant with color in the experiment and the S parameter is used as characteristic parameter, H and S parameter are adopted histogram calculation.

Note Sum (P, x _i) be that a certain eigen value (such as Hue) is x in the image _iPixel count, N is the total pixel number among the regional P, then the histogram of this feature of regional P is

H(P)=(h _x1，h _x2，........h _xi...h _xn)

Wherein

The histogram calculation of S parameter roughly the same;

And it is poor to calculate the H parameter and standard: ${\stackrel{\‾}{S}}_H=\frac{1}{N}\sqrt{\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(H_i-\stackrel{\‾}{H})}^2}$

In the formula,

Be colourity average, H _iBe the chromatic value of certain pixel in the image, the standard deviation compute classes of S parameter together.

Understand the exsiccosis of crop, then must grasp first the normal growth state of crop, in the present invention, set up the plant growth data model, the characteristic parameter of plant growth has been carried out test of many times and data analysis, propose following plant growth data model.

(1) relational model of plant height and time

$Y=\frac{A}{1+B\·e^{\mathrm{CX}}}$

Y is the growing height of crop in the formula, and X is growing degree days (GDD), and A, B, C are the crop modeling parameters, and model parameter is relevant with kind and the vegetative stage of crop.

(2) the leaf relational model of crop

In the process of growth of crop, the growth of the leaf of crop and the relation between the water are also very tight, crop leaf by changing along the wide variation of the leaf of direction of extension, and the wide lw of leaf and the long ll of leaf have certain functional relation, can be represented by the formula:

$\frac{\mathrm{lw}}{\mathrm{LW}}=\mathrm{\α}\·{\left(\frac{\mathrm{ll}}{\mathrm{LL}}\right)}^2+\mathrm{\β}\·\frac{\mathrm{ll}}{\mathrm{LL}}+\mathrm{\γ}$

LL is the length of blade in the formula; LW is the Breadth Maximum of blade; Lw one blade is the width of blade at ll place in length; α, beta, gamma are model parameter.

(3) the area relationship model of crop

Leaf area is long with leaf, wide long-pending being directly proportional of leaf, can be represented by the formula:

LA=j×LW×LL

In the formula, j is correction coefficient, and its numerical value is relevant with leaf morphology, slightly changes with crop varieties and phyllotaxy, and general span is 0.6～0.9.

(4) the leaf color characteristic model of crop

$C_Y=K_1\·W_{\stackrel{\‾}{H}}+K_2$

C wherein _YBe the value of the contained chlorophyll total amount of Unit Weight blade,

Be the average of leaf image H parameter component, K ₁And K ₂It is model parameter.

The present invention adopts the genetic neural network finish hybrid genetic algorithm to train and differentiate.According to reality is differentiated process simulation and the complexity of prediction and the architectural characteristic of neural network model thereof, genetic algorithm (GeneticAlgorithms) is combined with neutral net, utilize genetic algorithm that neutral net is trained, obtained preferably effect.

Among the present invention, design has adopted genetic algorithm to realize study and the design of 3 layers of neutral net.

(1) individual variable is the neuroid weights in the genetic algorithm, adopts the decimal coded mode, and each individual dimension is M (N+O), M wherein, and N, O are respectively hidden layer, input layer and output layer neuron number.

(2) determine network weight and initialization population, if W=is (w ₁, w ₂..., w _n), n is population number, determines that object function E is:

$E=\frac{1}{2R}\underset{R=1}{\overset{R}{\mathrm{\Σ}}}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{[y^{\′}\left(i\right)-y\left(i\right)]}^2,$ And get fitness function be

Wherein R be training sample to sum, m is number of network node, y (i) is the expectation network output valve of i training sample, y ' is the network output valve of i training sample (i), L is model parameter;

(3) carry out population and copy, keep simultaneously the uniformity of population scale, fitness value is sorted from big to small, keep optimum individual and do not carry out the crossover and mutation operation.Carry out the crossover and mutation operation to remaining individuality according to crossover operator Pc and mutation operator Pm, repeat until form population of new generation.

Description of drawings

The structure chart of Fig. 1 crop irrigation system

The structure chart that Fig. 2 crop map picture is processed

Fig. 3 neural network structure figure

Embodiment

System of the present invention is obtained first the image of continuous plant growth situation by industrial camera or digital camera, carried out the steps such as image preliminary treatment, image binaryzation, image segmentation, extracted region and calculating by calculator and obtain the crop growthing state parameters such as cane diameter, tender shoots form and leaf color of crop, in conjunction with the plant growth data model, comprehensively make lack of water and judge, and by the irrigation that discharges water of Single-chip Controlling magnetic valve.

Utilize crop map that image capture device gathers crop as the time, in order to solve the Normalization of image in the measurement, at the rear portion placement rack millimeter paper of measuring crop, and carry out equidistant stain mark at coordinate paper, so that the processing of the characteristic parameter of later image.

System hardware is mainly by calculator and ATMEGA128 or M430F2131 type microprocessor consists of, control circuit is comprised of AT45DB161 storage chip, DS1302 clock chip and MAX232 serial communication chip etc.

For obtaining accurately the growth parameter(s) of crop, duration of test can carry out obtaining of image, the variations such as the height of a crop of automatic monitor for continuously, diameter stem, leaf growth every 10 minutes, the different time intervals such as half an hour, 1 hour.Accurately draw the growth change state of crop.Make the observation of color, morphological index in time, fast, namely when slight metamorphosis occurring, just take measures.Because it is different that the characteristic parameter of Different Crop growth changes, so to the plant growth data model, constantly put into practice and Rule Summary, judge with form and change color that crop is relatively more responsive, be dirty-green such as the heart caudal lobe of peanut and represent lack of water, the jute end pin is straight, and vein obviously represents lack of water.

In the testing process of the crop color of reality, adopt the HSI color space, RGB and HSI are definite by following formula conversion:

$W=\arccos \left[\frac{2R-G-B}{2\sqrt{{(R-G)}^2+(R-B)(G-B)}}\right]$

And H, S, I are respectively: $H=\left\{\begin{array}{cc}2\mathrm{\π}-W& B>G\\ W& B\≤G\end{array}\right.$

$S=1-\frac{3\min (R,G,B)}{R+G+B}$

$I=\frac{R+G+B}{3}$

System of the present invention can in time obtain crop water information, and realizes timely precision irrigation, has well improved the availability of water resource.System has the advantages such as robustness is good, expansion is flexible.Have good practicality, can be applied to the crop-planting zones such as greenhouse, farmland, nursery.

Claims (3)

1. the crop irrigation system based on computer vision is characterized in that system is comprised of image acquisition equipment, calculator, single-chip microcomputer and irrigation system, and system realizes comprising following concrete steps:

(1) obtains the growth original image of crop;

(2) original image is carried out preliminary treatment;

(3) respectively image is carried out binary conversion treatment and from the RGB color space conversion to the HSI color space, and choose H in the HSI color space and S parameter as the color characteristic of crop map picture, in the crop map picture of HSI color space, select the part pixel as seed;

(4) seed region is grown, and will be attached to the neighbor of seed color similar performance on the seed of growth district, and a plurality of sub-blocks of entire image are scanned, and to close on color, zone adjacent on the space merges;

(5) finish the extraction of image segmentation and the plant growth characteristic parameter of crop, concrete plant growth characteristic parameter has: the color parameter of plant height parameter, leaf area parameter and cauline leaf;

(6) various characteristic parameters and plant growth data model are carried out data comparison and calculating, find out difference value, specifically in the foundation of plant growth data model, mainly set up following model:

(a) relational model of plant height and time

$Y=\frac{A}{1+B\·e^{\mathrm{CX}}}$

Y is the growing height of crop in the formula, and X is growing degree days (GDD), and A, B, C are the crop modeling parameters;

(b) the leaf relational model of crop

The wide lw of crop leaf and the long ll of leaf have certain functional relation, represent with following formula:

$\frac{\mathrm{lw}}{\mathrm{LW}}=\mathrm{\α}\·{\left(\frac{\mathrm{ll}}{\mathrm{LL}}\right)}^2+\mathrm{\β}\·\frac{\mathrm{ll}}{\mathrm{LL}}+\mathrm{\γ}$

LL is the length of blade in the formula; LW is the Breadth Maximum of blade; Lw one blade is the width of blade at ll place in length; α, beta, gamma are model parameter;

(c) the area relationship model of crop

Leaf area is long with leaf, wide long-pending being directly proportional of leaf, can be represented by the formula:

LA=j×LW×LL

Wherein, j is correction coefficient, and is relevant with leaf morphology, changes with the difference of crop varieties;

(d) the leaf color characteristic model of crop

$C_Y=K_1\·W_{\stackrel{\‾}{H}}+K_2$

C wherein _YBe the value of the contained chlorophyll total amount of Unit Weight blade,

Be the average of leaf image H parameter component, K ₁And K ₂It is model parameter;

(7) calculate and adjudicate based on genetic neural network;

(8) according to the computer decision result, by the Single-chip Controlling magnetic valve, finish the irrigation of crop.

2. the crop irrigation system based on computer vision according to claim 1 is characterized in that choosing in H and the calculating of S parameter as Color characteristics parameters in step (3), takes the histogram calculation method, and the H parameter histogram of regional P is calculated as:

H(P)=(h _x1，h _x2，........h _xi...h _xn)

Wherein

Sum (P, x _i) be that a certain eigen value is x in the image _iPixel count, N is the total pixel number among the regional P, the histogram calculation of S parameter roughly the same;

And calculate the standard deviation of H parameter:

In the formula,

Be colourity average, H _iBe the chromatic value of certain pixel in the image, the standard deviation compute classes of S parameter together.

3. the crop irrigation system based on computer vision according to claim 1 is characterized in that in the genetic neural network design of step (7) following design feature being arranged specifically:

(1) individual variable is the neuroid weights in the genetic algorithm, adopts the decimal coded mode, and each individual dimension is M (N+O), M wherein, and N, O are respectively hidden layer, input layer and output layer neuron number;

(2) determine network weight and initialization population, if W=is (w ₁, w ₂..., w _n), n is population number, determines that object function E is:

$E=\frac{1}{2R}\underset{R=1}{\overset{R}{\mathrm{\Σ}}}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{[y^{\′}\left(i\right)-y\left(i\right)]}^2,$ And get fitness function be $f=\frac{L}{E}$

Wherein R be training sample to sum, m is number of network node, y (i) is the expectation network output valve of i training sample,

Y ' is the network output valve of i training sample (i), and L is model parameter;

(3) carry out population and copy, keep simultaneously the uniformity of population scale, fitness value is sorted from big to small, keep optimum individual and do not carry out the crossover and mutation operation; Carry out the crossover and mutation operation to remaining individuality according to crossover operator Pc and mutation operator Pm, repeat until form population of new generation.

Images