CN110084145A - A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow - Google Patents
A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow Download PDFInfo
- Publication number
- CN110084145A CN110084145A CN201910277867.4A CN201910277867A CN110084145A CN 110084145 A CN110084145 A CN 110084145A CN 201910277867 A CN201910277867 A CN 201910277867A CN 110084145 A CN110084145 A CN 110084145A
- Authority
- CN
- China
- Prior art keywords
- image
- value
- layer
- measuring point
- follows
- 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
- 241000607479 Yersinia pestis Species 0.000 title claims abstract description 69
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 67
- 201000010099 disease Diseases 0.000 title claims abstract description 66
- 230000006378 damage Effects 0.000 title claims abstract description 37
- 238000011017 operating method Methods 0.000 title claims abstract description 5
- 238000000034 method Methods 0.000 claims abstract description 62
- 238000012545 processing Methods 0.000 claims abstract description 25
- 239000002689 soil Substances 0.000 claims abstract description 18
- 238000005516 engineering process Methods 0.000 claims abstract description 10
- 230000005540 biological transmission Effects 0.000 claims abstract description 8
- 238000007619 statistical method Methods 0.000 claims abstract description 4
- QVFWZNCVPCJQOP-UHFFFAOYSA-N chloralodol Chemical compound CC(O)(C)CC(C)OC(O)C(Cl)(Cl)Cl QVFWZNCVPCJQOP-UHFFFAOYSA-N 0.000 claims abstract description 3
- 210000004218 nerve net Anatomy 0.000 claims description 52
- 230000006870 function Effects 0.000 claims description 24
- 230000009418 agronomic effect Effects 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 20
- 210000002364 input neuron Anatomy 0.000 claims description 15
- 238000003709 image segmentation Methods 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 10
- 238000004458 analytical method Methods 0.000 claims description 7
- 238000012544 monitoring process Methods 0.000 claims description 7
- 230000003321 amplification Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 210000005036 nerve Anatomy 0.000 claims description 5
- 230000009514 concussion Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 230000003466 anti-cipated effect Effects 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 238000013139 quantization Methods 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 6
- 241000238631 Hexapoda Species 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 206010059866 Drug resistance Diseases 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000009313 farming Methods 0.000 description 1
- 239000003337 fertilizer Substances 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 239000000575 pesticide Substances 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Catching Or Destruction (AREA)
Abstract
The invention particularly discloses a kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow, including OV7725 camera child node acquisition unit and LoRa transmission child node transmission unit, soil entropy sense signals node acquisition unit, the hardware platforms such as stm32 endpoint processing unit based on technology of Internet of things, the image of acquisition and agricultural land information are transmitted to monitoring center；The data statistical analysis method of base TensorFlow is carried out in monitoring center, crop map picture is analyzed and determined with agricultural land information, obtains accurate pest and disease damage identification judgement.
Description
Technical field
The invention belongs to agricultural pests to identify field, and in particular to one kind (is compiled based on TensorFlow based on data flow
The symbolic mathematical system of journey) the multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method.
Background technique
Farmland diseases and pests of agronomic crop monitoring is important one for realizing IT application to agriculture, and then realizing agriculture wisdom
Step, whole work process design numerous step and method.In these years diseases and pests of agronomic crop seems increasingly severe, various diseases
Disease is intersected continuous.It on the one hand is that Hu of the pesticide in type and dosage uses excessively, crop pests gradually develop drug resistance；It is another
Aspect agricultural product price drops all the way, and peasant is eager to increase income, and the usage amount of chemical fertilizer increases every year, leads to soil constituent
Variation, is suitble to the growth of autochthonal crop pests.Diseases and pests of agronomic crop is increasingly difficult to administer, and is considered making for agriculture chemical merely
Dosage cause be it is inaccurate, need to be integrated by the leaf image and various soil, air data of diseases and pests of agronomic crop
Analysis, wherein pest and disease damage automatic identification is one of major issue, needs to find suitable farmland diseases and pests of agronomic crop identification
Calculation method.
Farmland diseases and pests of agronomic crop, which monitors traditional method generally, plant protection personnel field investigation, field sampling etc., but this
A little methods have destructiveness, and time and effort consuming to crops itself, and there are the representative poor, subjectivities of Points replacing surfaces by force, timeliness
The drawbacks such as difference, it is difficult to meet a wide range of pest and disease damage real-time monitoring requirement, application can not be popularized in an all-round way.In recent years, farmland farming
Object pest and disease damage identification technology has carried out the method with modern science and technology technology, for example Bayes Discriminatory Method, core K-mean value are gathered
The application scenario of class method, coloration moments method.Since various method and technologies have its own application conditions and limitation, these methods are all
Adjustment picture size and feature that cannot be adaptive be judged, if sometimes in practice only using a kind of parameter as explain according to
According to, it may be difficult to it achieves the desired result, therefore, in practice generally requires and carry out automatic adjusument with different situations using a kind of
Method.
With the development of science and technology, diseases and pests of agronomic crop identification technology in farmland all achieves significant progress and development,
The accuracy of pest and disease damage identification and speed are continuously improved.But various recognition methods have the application premise and limitation of its own,
In practical applications, most methods are all only by environmental ideals at present, and method itself is there is no innovating, so in farmland agriculture
Traditional method is sometimes difficult to obtain preferable effect in crop disease and insect monitoring.Therefore, in farmland wisdom crop diseases and pest
In the application of evil monitoring, scientific and technical personnel should select reasonable data processing side according to field medium physical property and characteristics of image
Method.
Currently, there is no special image and data integrated treatment based on TensorFlow in the identification of farmland diseases and pests of agronomic crop
Design scheme.In the construction of farmland monitoring works, the crops that are generally mounted with based on camera and data transmission module
Monitoring device, can be to collected agricultural land information, the crops picture taken and to aerial temperature and humidity, crops germ spore
Son, pest and disease damage quantity and disease and insect information displacement are monitored.However, being only mostly after existing image and data acquisition
Manually differentiated, is identified by the pest and disease damage that subjective factor is influenced greatly be neatly realized different croplands, be also unfavorable for agriculture
Industry is intelligentized universal.
Summary of the invention
It includes that manual identified, MATLAB identification, Machine Vision Recognition etc. are a variety of that farmland diseases and pests of agronomic crop, which knows method for distinguishing,
Transmission method.The scene and effective distance of different types of pest and disease damage recognition methods application are often different, the sense of external manifestation
It is also not identical to know technology, therefore first has to research diseases and pests of agronomic crop image information and agricultural land information.In order to more accurately and timely
Ground carries out pest and disease damage identification to the farmland crops in complex environment, collaborative perception ambient enviroment or oneself state is needed, to it
Adaptive, time-frequency domain, multiple dimensioned processing method are carried out, obtains effective information, and root from agricultural land soil and crop map picture
It is actively responded according to respective rule, carries out preliminary treatment and diseases and pests of agronomic crop identification.Therefore, the crop disease of most complex scenarios
Identifying pest technical research is to ensure one of the important link of agriculture wisdom.
In conclusion analyzing leads to agricultural pest present invention incorporates the actual conditions of agricultural pest identification
Agricultural land information and crop information, the characteristics of identification for agricultural pest, when proposing the pest and disease damage based on TensorFlow
The multiple dimensioned identifying system of frequency domain and method: 1, crops occur autochthonal pest and disease damage will lead to pest and disease damage point and surrounding medium it
Between there are obvious temperature and humidity differences to compare and analyze to need to acquire soil entropy；2, pest and disease damage occurs for crops
When leaf characteristic and leaf characteristic when pest and disease damage does not occur be different, to need to acquire in farmland by camera
Leaf image, compare and analyze；3, the crop map picture of shooting, which does not ensure that, to be apparent and accurately, to need
Multi-scale transform is carried out, treating capacity is reduced at the position that pest and disease damage occurs in lock image, increases processing accuracy；4, disease pest occurs
Fine distinction is had between harmful crops and the crops that pest and disease damage does not occur, human eye is difficult to discover under normal circumstances, from
And it needs to carry out crop map picture and agricultural land information the judgement on time-frequency domain；5, it will treated agricultural land information and image information
Integrated treatment is carried out, judgment threshold is changed using the method based on TensorFlow automatically, adaptive adjusting parameter, it is accurate to obtain
Pest and disease damage identify judgement.
A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain based on TensorFlow, including based on technology of Internet of things
OV7725 camera child node acquisition unit and LoRa transmission child node transmission unit, soil entropy sense signals node acquisition list
The hardware platforms such as member, stm32 endpoint processing unit, are transmitted to monitoring center for the image of acquisition and agricultural land information；In monitoring
The heart carries out the data statistical analysis method based on TensorFlow, and crop map picture is analyzed and determined with agricultural land information, is obtained
Accurate pest and disease damage identification judgement out.
A kind of operating method of the multiple dimensioned identifying system of pest and disease damage time-frequency domain based on TensorFlow, including walk as follows
It is rapid:
The first step finds out the temperature field in crops farmland and the gradient distribution of moisture field；
Second step, farmland crop map is as multi-scale transform；
Third step, farmland crop map are converted as time domain；
4th step, the transformation of farmland crops image frequency domain；
5th step, data fusion, the agricultural land soil temperature and humidity that the first step, third step and the 4th step are obtained, crop map
Sentence as time-frequency domain information obtains accurate pest and disease damage identification using the method progress data statistic analysis of the double-deck Propagation Neural Network
It is disconnected.
Further, the TensorFlow method is the mathematical statistics analysis and processing method based on adaptive change.
Further, the gradient distribution in the temperature field for finding out crops farmland and moisture field, concrete operations are as follows:
Diseases and pests of agronomic crop identifies that agricultural land soil temperature acquisition treatment process is as follows:
Obtain the temperature value W (t) of the current measuring point n of current time tnWith the temperature value W (t) of upper measuring point n-1n-1It carries out
Gradient algorithm, calculation formula are as follows:
Tt=W (t)n-W(t)n-1
W(t)nFor the temperature value of the current measuring point n of current time t；W(t)n-1For the upper measuring point n-1 of current time t
Temperature value；TtFor temperature gradient value.By TtAfter temperature gradient value calculates, the gradient value for carrying out each measuring point is described, and makes correspondence
The temperature gradient field distribution of point；
Diseases and pests of agronomic crop identifies that agricultural land soil humidity collection treatment process is as follows:
Obtain the humidity value H (t) of the current measuring point n of current time tnWith the humidity value H (t) of upper measuring point n-1n-1It carries out
Gradient algorithm, calculation formula are as follows:
Dt=H (t)n-H(t)n-1
H(t)nFor the humidity value of the current measuring point n of current time t；H(t)n-1For the upper measuring point n-1 of current time t
Humidity value；DtFor moist gradient value.By DtAfter moist gradient value calculates, the gradient value for carrying out each measuring point is described, and makes correspondence
The moist gradient field distribution of point.
Further, for the farmland crop map as multi-scale transform, concrete operations are as follows:
Diseases and pests of agronomic crop identifies that image dividing processing process is as follows:
The image R for occupying entire area of space is obtained, R is segmented the image into1、R2、R3、R4Four regions, process can be with
It indicates are as follows:
R1+R2+R3+R4=R
R is the entire area of space that image occupies；R1For the upper left image block of image segmentation, R2For the upper right of image segmentation
Image block, R3For the lower-left image block of image segmentation, R4For the bottom right image block of image segmentation.This time image is divided twice
It cuts, the figure layer after segmentation amplifies processing, that is, reaches multi-scale transform purpose；
The diseases and pests of agronomic crop identification image amplification transformation treatment process is as follows:
The one-dimensional abscissa function B (x) for obtaining original image is shortened by the one-dimensional abscissa function that processing obtains original image
The function B (2x-1) that one times of obtained function B (x), B (2x) are obtained to one unit of right translation, calculation formula are as follows:
A (x)=2 [B (2x)+B (2x-1)]
A (x) is twice of magnification function on the one-dimensional abscissa of image, and B (x) is the one-dimensional abscissa function of original image, B
(2x) is that the one-dimensional abscissa function of original image is shortened to one times of obtained function, and B (2x-1) is by B (2x) to right translation one
The function that unit obtains.One-dimensional ordinate function can finally be enlarged into image original four times also according to the processing of this method.
Multi-scale transform of the present invention has image segmentation twice and image enhanced processing, preferably, using first image
Segmentation, then image amplify, again image segmentation, again the process of image amplification, and the image finally obtained is original image length and width
On be exaggerated four times, 16 times are exaggerated on area.Image Multiscale transformation of the invention is just completed since then.
Further, for the farmland crop map as time domain transformation, specific operation process is as follows:
It obtains and represents the coordinate (r, g, b) of diseases and pests of agronomic crop image red, green, blue, r, the maximum max in g and b
And a smallest min, calculation formula are as follows in r, g, b:
S=((max-min)/max) * 100/255
V=max*100/255
H is the tone size on the red, green, blue coordinate (r, g, b) of image, and S is the red, green, blue coordinate (r, g, b) of image
On saturation dimension, V be image red, green, blue coordinate (r, g, b) on brightness size.H is obtained, after S, V, to quantization
The row pixel and column pixel of image afterwards carry out complementation, are matched with 72 color regions, and the color for finally extracting image is special
It levies and is matched with color libraries, obtained no disease and pests harm.
Further, the farmland crops image frequency domain transformation, specific operation process are as follows:
Obtain Gabor filter direction and scale μ and ν, image slices vegetarian refreshments coordinate (x, y) indicated with z, frequency domain Gauss
Envelope σ controls the width, concussion part wavelength and the function k in direction of Gaussian windowμ,ν.Calculation formula is as follows:
Ψμ,vIt (z) is Gabor core amplitude characteristic；μ and ν respectively indicates direction and the scale of Gabor filter；Z represents figure
As pixel coordinate (x, y)；| | | | indicate norm；σ is Gaussian envelope；kμ,νControl width, the concussion part wavelength of Gaussian window
And direction, it is defined as
kν=kmax/fυFor filter sample frequency, kmaxFor maximum frequency, fυBetween limitation frequency domain Kernel Function distance
Every the factor；
Obtain Gabor core amplitude characteristic Ψμ,v(z) after, convolution is carried out with original image, apparent characteristics of image can be obtained,
To carry out pest and disease damage judgement.
Further, the step is fifth is that respectively obtain farmland temperature field, moisture field, Time Domain Processing by above step
It is worth, after frequency domain processing costs, data statistic analysis is carried out using the double-deck Propagation Neural Network, to input vector XiAt the normalization done
Reason, calculation formula are as follows:
XiFor the input vector of each neuron, the i.e. soil moisture fields of the invention, temperature field, time-domain calculation value, frequency domain
Calculated value.Xi *For to input vector XiThe normalized done.
Data can be passed to first layer nerve net by not all input neuron node, and only optimal value just may be used
With calculation formula is as follows:
Xi *For to current measuring point n from input neuron to first layer nerve net input vector XiThe normalized done.It is current measuring point n from input neuron to first layer nerve net input vector XiThe normalization connection weighted value done, normalizing
Change method and XiMethod for normalizing is identical.SjIt (n) is Xi *WithThe maximum value namely first layer optimal value of sum of products.
First layer optimal value in order to obtain needs repeatedly to adjust weighted value, and current measuring point n is from input neuron to first layer
The connection weighted value modified computing formulae of nerve net is as follows:
Normalization for current measuring point n from input neuron to 4 input values of first layer nerve net connects weighting
Value, initial assignment is respectively 0.1,0.2,0.3,0.4.α is learning rate, is between one 0-1 as training constantly reduces
Value, initial assignment 0.3.Xi *For to current measuring point n from input neuron to first layer nerve net input vector XiThe normalization done
Processing.Connection weighted value for revised current measuring point n from input neuron to first layer nerve net.It repairs repeatedly
Positive weight, until weight is constant, i.e.,It is equal toAt this timeEqual to zero.
Equally, data can be passed to second layer nerve net by not all first layer nerve net neuron node,
Only optimal value just can be with calculation formula is as follows:
VjkConnection weighted value for current measuring point n from first layer nerve net neuron node to second layer nerve net.Sj(n)
For Xi *WithThe maximum value namely first layer nerve net optimal value of sum of products.LjIt (n) is VjkWith Sj(n) maximum of sum of products
Value namely second layer nerve net optimal value；
Second layer optimal value in order to obtain, needs repeatedly to adjust weighted value, first layer nerve net to second layer nerve net it
Between weighted value modified computing formulae it is as follows:
Vjk(n+1)=Vjk(n)+βbj[(yk-Dk]
VjkIt (n) is connection weighted value of the current measuring point n from first layer nerve net to second layer nerve net, initial assignment is all
It is 0.5,0.5.β is learning rate, is the value between one 0-1, initial assignment 0.4 as training constantly reduces.bjFor competition
The two-value output vector of layer, works as Lj(n) when getting, bj=1；No person bj=0.DkFor the anticipated output value of second layer nerve net.Vjk
It (n+1) is connection weighted value of the revised current measuring point n from first layer nerve net to second layer nerve net.ykFor second layer mind
Real output value through netting, calculation formula are as follows:
VkjFor the connection weighted value of first layer nerve net to second layer nerve net, Vkj *For first layer nerve net to the second layer
The connection weighted value V of nerve netkjWith Sj(n) product, works as Lj(n) when getting, i.e. first layer nerve net optimal value and the second layer
When nerve net connects weighted value sum of products maximum, b at this timej=1, Vkj *For Lj(n)；No person bj=0, yk=0.
The utility model has the advantages that
The present invention overcomes the problems, such as that the farmland diseases and pests of agronomic crop identification of existing most complex scenarios is difficult, can accomplish section's knowledge
Other farmland diseases and pests of agronomic crop.
Detailed description of the invention
Fig. 1 is the multiple dimensioned identifying system of pest and disease damage time-frequency domain and method flow diagram of the invention based on TensorFlow；
Fig. 2 is that agricultural land Image Multiscale of the invention divides enlarged diagram；
Fig. 3 is agricultural land image time domain HSV model schematic of the invention；
Fig. 4 is the double-deck Propagation Neural Network schematic diagram.
Specific embodiment
Below in conjunction with specific embodiment, the present invention is described in further detail:
Fig. 1 is the pest and disease damage time-frequency of the invention based on TensorFlow (symbolic mathematical system based on data flow programming)
The multiple dimensioned identifying system in domain and method flow diagram.System is initialized first, measures soil temperature and humidity and the agriculture of current measuring point n
Crop image information；Then gradient solution procedure, the humidity field gradient of agricultural land soil in the temperature field of agricultural land soil are successively carried out
Solution procedure, crop map picture first time amplification process, is divided for second of crop map picture crop map picture first time cutting procedure
Process, second of amplification process of crop map picture, crop map are cut as time domain conversion process, crops image frequency domain are transformed
Journey；Temperature and humidity gradient fields and crop map most are respectively obtained as time-frequency domain information through above step afterwards, propagate nerve using bilayer
Network method judges whether current measuring point n occurs pest and disease damage to data processing for statistical analysis.
Fig. 2 is that agricultural land Image Multiscale of the invention divides enlarged diagram.1 is the first figure layer, is as shot
The original image arrived or the image by simple physical processing, are divided into four parts for the first figure layer, one piece of figure layer of upper right are taken to be put
Greatly 2, the second figure layer is formed, is exaggerated four times on the first figure layer upper right portion area at this time.Continue for the second figure layer to be divided into
Four parts, taking one piece of figure layer of upper right to amplify is 3, forms third figure layer, is amplified on the second figure layer upper right portion area at this time
Four times, 16 times are amplified on the first figure layer upper right portion area.According to this method, first figure layer each section can be put
Greatly, that is, multi-scale transform is realized.
Fig. 3 is agricultural land image time domain HSV model schematic of the invention.1 is red tone place angle；2 are
Angle where yellow tone；3 be green tone place angle；4 be cyan color tone place angle；5 be blue color place angle；
6 be angle where pinkish red tone.7 be the direction of rotation tone (H).8 be saturation degree (S) extending direction.9 be the extension side brightness (V)
To.Crop map picture can be subjected to classification processing according to the combination of 7 (tones), 8 (saturation degrees), 9 (brightness).
Fig. 4 is the double-deck Propagation Neural Network schematic diagram of the invention.1 is the temperature gradient value of current measuring point n；2 be current
The moist gradient value of measuring point n；3 be the crop map of current measuring point n as Time Domain Processing value；4 be the crop map picture of current measuring point n
Frequency domain processing costs；5,6 be the two kinds of results judged (whether being)；7 be input neuron node, and 8 be first layer nerve net node,
9 be second layer nerve net node.Measuring point data carries out preferably, carrying out weighed value adjusting from 1,2,3,4 nodes, and optimal value enters first
Layer nerve net；It also passes through preferably, carries out weighed value adjusting, optimal value enters second layer nerve net, finally on TensorFlows
Show whether measuring point occurs pest and disease damage.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (4)
1. a kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain based on TensorFlow, including based on technology of Internet of things
OV7725 camera child node acquisition unit and LoRa transmission child node transmission unit, soil entropy sense signals node acquisition list
The hardware platforms such as member, stm32 endpoint processing unit, are transmitted to monitoring center for the image of acquisition and agricultural land information；In monitoring
The heart carries out the data statistical analysis method based on TensorFlow, and crop map picture is analyzed and determined with agricultural land information, is obtained
Accurate pest and disease damage identification judgement out.
2. a kind of operating method of the multiple dimensioned identifying system of pest and disease damage time-frequency domain based on TensorFlow, includes the following steps:
The first step finds out the temperature field in crops farmland and the gradient distribution of moisture field；
Second step, farmland crop map is as multi-scale transform；
Third step, farmland crop map are converted as time domain；
4th step, the transformation of farmland crops image frequency domain；
5th step, data fusion, the agricultural land soil temperature and humidity that the first step, third step and the 4th step are obtained, crop map as when
Frequency domain information carries out data statistic analysis using the method for the double-deck Propagation Neural Network and obtains accurate pest and disease damage identification judgement.
3. further, the TensorFlow method is the mathematical statistics analysis and processing method based on adaptive change.
4. further, the gradient distribution in the temperature field for finding out crops farmland and moisture field, concrete operations are as follows:
Diseases and pests of agronomic crop identifies that agricultural land soil temperature acquisition treatment process is as follows:
Obtain the temperature value W (t) of the current measuring point n of current time tnWith the temperature value W (t) of upper measuring point n-1n-1Carry out gradient fortune
It calculates, calculation formula are as follows:
Tt=W (t)n-W(t)n-1
W(t)nFor the temperature value of the current measuring point n of current time t；W(t)n-1For the temperature of the upper measuring point n-1 of current time t
Value；TtFor temperature gradient value.By TtAfter temperature gradient value calculates, the gradient value for carrying out each measuring point is described, and makes corresponding points
Temperature gradient field distribution；
Diseases and pests of agronomic crop identifies that agricultural land soil humidity collection treatment process is as follows:
Obtain the humidity value H (t) of the current measuring point n of current time tnWith the humidity value H (t) of upper measuring point n-1n-1Carry out gradient fortune
It calculates, calculation formula are as follows:
Dt=H (t)n-H(t)n-1
H(t)nFor the humidity value of the current measuring point n of current time t；H(t)n-1For the humidity of the upper measuring point n-1 of current time t
Value；DtFor moist gradient value.By DtAfter moist gradient value calculates, the gradient value for carrying out each measuring point is described, and makes corresponding points
Moist gradient field distribution.
Further, the farmland crop map operates as follows as multi-scale transform:
Diseases and pests of agronomic crop identifies that image dividing processing process is as follows:
The image R for occupying entire area of space is obtained, R is segmented the image into1、R2、R3、R4Four regions, process can indicate
Are as follows:
R1+R2+R3+R4=R
R is the entire area of space that image occupies；R1For the upper left image block of image segmentation, R2For the upper right image of image segmentation
Block, R3For the lower-left image block of image segmentation, R4For the bottom right image block of image segmentation.This time image is divided twice, point
Figure layer after cutting amplifies processing；
Diseases and pests of agronomic crop identifies that image amplification transformation treatment process is as follows:
The one-dimensional abscissa function B (x) for obtaining original image shortens one times by the one-dimensional abscissa function that processing obtains original image
The function B (2x-1) that obtained function B (x), B (2x) are obtained to one unit of right translation, calculation formula are as follows:
A (x)=2 [B (2x)+B (2x-1)]
A (x) is twice of magnification function on the one-dimensional abscissa of image, and B (x) is the one-dimensional abscissa function of original image, and B (2x) is
The one-dimensional abscissa function of original image is shortened into one times of obtained function, B (2x-1) is to obtain B (2x) to one unit of right translation
The function arrived, one-dimensional ordinate function can finally be enlarged into image original four times also according to the processing of this method；
Multi-scale transform has image segmentation twice and image enhanced processing, uses first image segmentation, then image amplification, again
Image segmentation, the process that image amplifies again, the image finally obtained is that four times are exaggerated in original image length and width, is put on area
It is 16 times big, Image Multiscale transformation is just completed since then.
Further, for the farmland crop map as time domain transformation, concrete operations are as follows:
It obtains and represents the coordinate (r, g, b) of diseases and pests of agronomic crop image red, green, blue, r, the maximum max and r in g and b,
A smallest min, calculation formula are as follows in g, b:
S=((max-min)/max) * 100/255
V=max*100/255
H is the tone size on the red, green, blue coordinate (r, g, b) of image, and S is on the red, green, blue coordinate (r, g, b) of image
Saturation dimension, V are the brightness size on the red, green, blue coordinate (r, g, b) of image.H is obtained, after S, V, after quantization
The row pixel and column pixel of image carry out complementation, are matched with 72 color regions, finally extract the color characteristic of image simultaneously
It is matched with color libraries, has obtained no disease and pests harm.
Further, the farmland crops image frequency domain transformation.Specific operation process is as follows:
Obtain Gabor filter direction and scale μ and ν, image slices vegetarian refreshments coordinate (x, y) indicated with z, frequency domain Gaussian envelope
σ controls the width, concussion part wavelength and the function k in direction of Gaussian windowμ,ν.Calculation formula is as follows:
Ψμ,vIt (z) is Gabor core amplitude characteristic；μ and ν respectively indicates direction and the scale of Gabor filter；Z representative image picture
Vegetarian refreshments coordinate (x, y)；| | | | indicate norm；σ is Gaussian envelope；kμ,νControl Gaussian window width, concussion part wavelength and
Direction is defined as
kν=kmax/fυFor filter sample frequency, kmaxFor maximum frequency, fυFor limit frequency domain Kernel Function distance interval because
Son.
Obtain Gabor core amplitude characteristic Ψμ,v(z) after, convolution is carried out with original image, apparent characteristics of image can be obtained, thus
Pest and disease damage judgement is carried out,
Further, the step 5 is that farmland temperature field is respectively obtained by above step, moisture field, Time Domain Processing value, frequency domain
After processing costs, data statistic analysis is carried out using the double-deck Propagation Neural Network, to input vector XiThe normalized done calculates
Formula is as follows:
XiFor the input vector of each neuron, the i.e. soil moisture fields of the invention, temperature field, time-domain calculation value, frequency-domain calculations
Value, Xi *For to input vector XiThe normalized done；
Data can be passed to first layer nerve net by not all input neuron node, and only optimal value just can be with,
Calculation formula is as follows:
Xi *For to current measuring point n from input neuron to first layer nerve net input vector XiThe normalized done.For
Current measuring point n is from input neuron to first layer nerve net input vector XiThe normalization connection weighted value done, method for normalizing
With XiMethod for normalizing is identical.SjIt (n) is Xi *WithThe maximum value namely first layer optimal value of sum of products；
First layer optimal value in order to obtain needs repeatedly to adjust weighted value, current measuring point n nerve from input neuron to first layer
The connection weighted value modified computing formulae of net is as follows:
Normalization for current measuring point n from input neuron to 4 input values of first layer nerve net connects weighted value, just
Beginning assignment is respectively 0.1,0.2,0.3,0.4.α is learning rate, is the value between one 0-1, just as training constantly reduces
Beginning assignment 0.3.Xi *For to current measuring point n from input neuron to first layer nerve net input vector XiThe normalized done.Connection weighted value for revised current measuring point n from input neuron to first layer nerve net.Amendment power repeatedly
Value, until weight is constant, i.e.,It is equal toAt this timeEqual to zero；
Equally, data can be passed to second layer nerve net by not all first layer nerve net neuron node, only
Optimal value just can be with calculation formula is as follows:
VjkConnection weighted value for current measuring point n from first layer nerve net neuron node to second layer nerve net.SjIt (n) is Xi *
WithThe maximum value namely first layer nerve net optimal value of sum of products.LjIt (n) is VjkWith Sj(n) maximum value of sum of products,
Namely second layer nerve net optimal value；
Second layer optimal value in order to obtain needs repeatedly to adjust weighted value, and first layer nerve net is between second layer nerve net
Weighted value modified computing formulae is as follows:
Vjk(n+1)=Vjk(n)+βbj[(yk-Dk]
VjkIt (n) is connection weighted value of the current measuring point n from first layer nerve net to second layer nerve net, initial assignment is all to be
0.5,0.5.β is learning rate, is the value between one 0-1, initial assignment 0.4 as training constantly reduces.bjFor competition layer
Two-value output vector, work as Lj(n) when getting, bj=1；No person bj=0.DkFor the anticipated output value of second layer nerve net.Vjk(n
It+1) is connection weighted value of the revised current measuring point n from first layer nerve net to second layer nerve net.ykFor second layer nerve
The real output value of net, calculation formula are as follows:
VkjFor the connection weighted value of first layer nerve net to second layer nerve net, Vkj *For first layer nerve net to second layer nerve
The connection weighted value V of netkjWith Sj(n) product, works as Lj(n) when getting, i.e., first layer nerve net optimal value and the second layer are neural
When net connection weighted value sum of products maximum, b at this timej=1, Vkj *For Lj(n)；No person bj=0, yk=0.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910277867.4A CN110084145B (en) | 2019-04-08 | 2019-04-08 | TensorFlow-based pest and disease damage time-frequency domain multi-scale identification system and operation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910277867.4A CN110084145B (en) | 2019-04-08 | 2019-04-08 | TensorFlow-based pest and disease damage time-frequency domain multi-scale identification system and operation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110084145A true CN110084145A (en) | 2019-08-02 |
| CN110084145B CN110084145B (en) | 2023-05-12 |
Family
ID=67414498
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910277867.4A Active CN110084145B (en) | 2019-04-08 | 2019-04-08 | TensorFlow-based pest and disease damage time-frequency domain multi-scale identification system and operation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110084145B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111107530A (en) * | 2019-12-06 | 2020-05-05 | 深圳大学 | Agricultural disease and pest control system based on LoRa technology |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102084794A (en) * | 2010-10-22 | 2011-06-08 | 华南农业大学 | Method and device for early detecting crop pests based on multisensor information fusion |
| CN102523953A (en) * | 2011-12-07 | 2012-07-04 | 北京农业信息技术研究中心 | Crop information fusion method and disease monitoring system |
| CN102945376A (en) * | 2012-09-28 | 2013-02-27 | 北京农业信息技术研究中心 | Method for diagnosing crops diseases |
| CN103489006A (en) * | 2013-10-11 | 2014-01-01 | 河南城建学院 | Computer vision-based rice disease, pest and weed diagnostic method |
| CN105825177A (en) * | 2016-03-09 | 2016-08-03 | 西安科技大学 | Remote-sensing crop disease identification method based on time phase and spectrum information and habitat condition |
| CN106202489A (en) * | 2016-07-20 | 2016-12-07 | 青岛云智环境数据管理有限公司 | A kind of agricultural pest intelligent diagnosis system based on big data |
| US20170351933A1 (en) * | 2016-06-01 | 2017-12-07 | Intel Corporation | Vision enhanced drones for precision farming |
| CN107862687A (en) * | 2017-11-07 | 2018-03-30 | 潘柏霖 | A kind of early warning system for being used to monitor agricultural pest |
| CN108009936A (en) * | 2017-10-31 | 2018-05-08 | 四川农业大学 | Pest and disease monitoring system based on internet of things |
| CN108960310A (en) * | 2018-06-25 | 2018-12-07 | 北京普惠三农科技有限公司 | A kind of agricultural pest recognition methods based on artificial intelligence |
| CN109470299A (en) * | 2018-10-19 | 2019-03-15 | 江苏大学 | A kind of plant growth information monitoring system and method based on Internet of Things |
-
2019
- 2019-04-08 CN CN201910277867.4A patent/CN110084145B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102084794A (en) * | 2010-10-22 | 2011-06-08 | 华南农业大学 | Method and device for early detecting crop pests based on multisensor information fusion |
| CN102523953A (en) * | 2011-12-07 | 2012-07-04 | 北京农业信息技术研究中心 | Crop information fusion method and disease monitoring system |
| CN102945376A (en) * | 2012-09-28 | 2013-02-27 | 北京农业信息技术研究中心 | Method for diagnosing crops diseases |
| CN103489006A (en) * | 2013-10-11 | 2014-01-01 | 河南城建学院 | Computer vision-based rice disease, pest and weed diagnostic method |
| CN105825177A (en) * | 2016-03-09 | 2016-08-03 | 西安科技大学 | Remote-sensing crop disease identification method based on time phase and spectrum information and habitat condition |
| US20170351933A1 (en) * | 2016-06-01 | 2017-12-07 | Intel Corporation | Vision enhanced drones for precision farming |
| CN106202489A (en) * | 2016-07-20 | 2016-12-07 | 青岛云智环境数据管理有限公司 | A kind of agricultural pest intelligent diagnosis system based on big data |
| CN108009936A (en) * | 2017-10-31 | 2018-05-08 | 四川农业大学 | Pest and disease monitoring system based on internet of things |
| CN107862687A (en) * | 2017-11-07 | 2018-03-30 | 潘柏霖 | A kind of early warning system for being used to monitor agricultural pest |
| CN108960310A (en) * | 2018-06-25 | 2018-12-07 | 北京普惠三农科技有限公司 | A kind of agricultural pest recognition methods based on artificial intelligence |
| CN109470299A (en) * | 2018-10-19 | 2019-03-15 | 江苏大学 | A kind of plant growth information monitoring system and method based on Internet of Things |
Non-Patent Citations (2)
| Title |
|---|
| HYEON PARK等: "《Crops Disease Diagnosing Using Image-Based Deep Learning Mechanism》", 《2018 INTERNATIONAL CONFERENCE ON COMPUTING AND NETWORK COMMUNICATIONS 》 * |
| 宋凯: "《基于计算机视觉的农作物病害识别方法的研究》", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111107530A (en) * | 2019-12-06 | 2020-05-05 | 深圳大学 | Agricultural disease and pest control system based on LoRa technology |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110084145B (en) | 2023-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sarangdhar et al. | Machine learning regression technique for cotton leaf disease detection and controlling using IoT | |
| Joshi et al. | Monitoring and controlling rice diseases using Image processing techniques | |
| Seng et al. | Computer vision and machine learning for viticulture technology | |
| Ramesh et al. | Rice blast disease detection and classification using machine learning algorithm | |
| Kiani et al. | Identification of plant disease infection using soft-computing: Application to modern botany | |
| Shah et al. | Detection of disease in cotton leaf using artificial neural network | |
| Vamsidhar et al. | Plant disease identification and classification using image processing | |
| Roldán-Serrato et al. | Automatic pest detection on bean and potato crops by applying neural classifiers | |
| Devi et al. | Web enabled paddy disease detection using Compressed Sensing | |
| Patel et al. | A survey on the plant leaf disease detection techniques | |
| Gining et al. | Harumanis mango leaf disease recognition system using image processing technique | |
| Kumar et al. | An identification of crop disease using image segmentation | |
| Machha et al. | Crop leaf disease diagnosis using convolutional neural network | |
| CN117496356A (en) | Agricultural artificial intelligent crop detection method and system | |
| CN110084145A (en) | A kind of multiple dimensioned identifying system of pest and disease damage time-frequency domain and operating method based on TensorFlow | |
| Kambale et al. | A survey paper on crop disease identification and classification using pattern recognition and digital image processing techniques | |
| Kumar et al. | Detection of Sugarcane Disease and Classification using Image Processing | |
| Radhika et al. | Machine learning approach-based plant disease detection and pest detection system | |
| Singh et al. | A review on recognition of plant disease using intelligent image retrieval techniques | |
| Tripathy et al. | Image processing techniques aiding smart agriculture | |
| Deisy et al. | Image segmentation for feature extraction: A study on disease diagnosis in agricultural plants | |
| Pawar et al. | Farmer Buddy-Plant Leaf Disease Detection on Android Phone | |
| Sujatha et al. | Disease identification in mango leaf using image processing | |
| Attada et al. | A methodology for automatic detection and classification of pests using optimized SVM in greenhouse crops | |
| Thorat et al. | Early disease detection and monitoring large field of crop by using IoT |
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 |