CN109977901A - A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing - Google Patents
A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing Download PDFInfo
- Publication number
- CN109977901A CN109977901A CN201910267513.1A CN201910267513A CN109977901A CN 109977901 A CN109977901 A CN 109977901A CN 201910267513 A CN201910267513 A CN 201910267513A CN 109977901 A CN109977901 A CN 109977901A
- Authority
- CN
- China
- Prior art keywords
- corn
- weeds
- model
- image
- aerial vehicle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 240000008042 Zea mays Species 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 46
- 235000002017 Zea mays subsp mays Nutrition 0.000 claims abstract description 36
- 241000196324 Embryophyta Species 0.000 claims abstract description 35
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 claims abstract description 33
- 235000005822 corn Nutrition 0.000 claims abstract description 33
- 238000000513 principal component analysis Methods 0.000 claims abstract description 23
- 230000009467 reduction Effects 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 238000003709 image segmentation Methods 0.000 claims abstract description 5
- 238000003066 decision tree Methods 0.000 claims description 40
- 238000007637 random forest analysis Methods 0.000 claims description 27
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 238000011156 evaluation Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 235000016383 Zea mays subsp huehuetenangensis Nutrition 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 235000009973 maize Nutrition 0.000 claims description 3
- 238000012360 testing method Methods 0.000 abstract description 14
- 238000005070 sampling Methods 0.000 abstract description 6
- 230000012010 growth Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 239000002689 soil Substances 0.000 description 4
- 238000013145 classification model Methods 0.000 description 3
- 230000006378 damage Effects 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 244000025254 Cannabis sativa Species 0.000 description 2
- 240000007594 Oryza sativa Species 0.000 description 2
- 235000007164 Oryza sativa Nutrition 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000000686 essence Substances 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000010977 jade Substances 0.000 description 2
- 230000008635 plant growth Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 235000009566 rice Nutrition 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000017105 transposition Effects 0.000 description 2
- 244000131316 Panax pseudoginseng Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 238000012271 agricultural production Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 235000013339 cereals Nutrition 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 235000015097 nutrients Nutrition 0.000 description 1
- 239000000575 pesticide Substances 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- VMXUWOKSQNHOCA-UKTHLTGXSA-N ranitidine Chemical compound [O-][N+](=O)\C=C(/NC)NCCSCC1=CC=C(CN(C)C)O1 VMXUWOKSQNHOCA-UKTHLTGXSA-N 0.000 description 1
- 238000007634 remodeling Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000031068 symbiosis, encompassing mutualism through parasitism Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000009333 weeding Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Image Processing (AREA)
Abstract
A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing, belongs to technical field of image processing.It is characterized by comprising following steps: step 1, the image data in corn planting region is obtained by unmanned plane, obtains the multispectral image in corn planting region；Step 2, image segmentation is carried out to the multispectral image in corn planting region；Step 3, texture feature extraction is carried out to the pseudo color image for converting band combination after principal component analysis processing, obtains textural characteristics parameter；Step 4, dimension-reduction treatment is carried out to textural characteristics parameter obtained in step 3, the textural characteristics parameter after obtaining dimensionality reduction；Step 5, classification monitor model is established, and classification monitor model is trained, monitor model identifies the weeds in corn planting region by classifying.By the corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing, the uniformly distributed sampling of test area, while discrimination with higher are realized.
Description
Technical field
A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing, belongs to technical field of image processing.
Background technique
In agricultural production process, weeds in field is considered as seriously affecting one of factor of plant growth, and China is because of weeds
Caused cereal crops production loss reaches the 10% of total output.Weeds are since its growth cycle is short, the speed of growth is fast, easily transmitted
The features such as, it is easy to concentrate large-scale outbreak, in the plant growths institute such as field and crop seedling contention soil nutrient, moisture, illumination
The resource needed leads to the harm such as crop seedling growth and development is slow, yield reduces, it is therefore necessary to add in time to weeds in field growth
With control.Chemical weed control practices is due to its high efficiency, it has also become main weeding means both domestic and external, but blanket type removed using chemistry
Careless agent not only causes certain destruction to farmland and surrounding ecological environment, and the medicament remained on crop leaf also pacifies food
It causes damages entirely.Therefore, accurate, quick crop weed identification method is studied, variable accurate for pesticide sprays the hair of technology
Exhibition is of great significance.
In recent years, carrying out weed identification using image recognition is an important research direction, current weed identification research
In, it is different according to image information acquisition means and image characteristics extraction mode, can be divided into based on spectral signature, machine vision and
The weed identification method of light spectrum image-forming, what is be widely used at present is the recognition methods based on machine vision, for the crop of acquisition
And weed images, the features such as shape, texture, color shown using crop and weeds in computer analysis image are extracted
The characteristic parameter to differ greatly carries out recognition detection to target.But the generally existing identification of weed identification method in the prior art
The low problem of rate affects the development of Technique for Identification of Weed, while traditional ground image acquisition is difficult to realize test area
Uniformly distributed sampling, therefore be easy to be influenced by sample close quarters, the method for increasing sample are mostly to rotate original image or again
Field image acquisition, it is time-consuming and laborious.
Summary of the invention
The technical problem to be solved by the present invention is overcoming the deficiencies of the prior art and provide a kind of test area of realizing
Uniformly distributed sampling, while there is the corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing of higher discrimination.
The technical solution adopted by the present invention to solve the technical problems is: should the corn weeds in field based on unmanned aerial vehicle remote sensing
Recognition methods, characterized by the following steps:
Step 1, in corn planting region, the image data in corn planting region is obtained by unmanned plane, to image data
It is pre-processed, obtains the multispectral image in corn planting region；
Step 2, image segmentation is carried out using multispectral image of the Principal Component Analysis to corn planting region；
Step 3, band group is converted after handling as texture analysis method principal component analysis using gray level co-occurrence matrixes method
The pseudo color image of conjunction carries out texture feature extraction, obtains textural characteristics parameter；
Step 4, dimension-reduction treatment is carried out to textural characteristics parameter obtained in step 3, the textural characteristics ginseng after obtaining dimensionality reduction
Number；
Step 5, classification monitor model is established, and classification monitor model is trained, by classification monitor model to jade
Weeds in rice planting area are identified.
Preferably, pretreated process described in step 1 are as follows: radiant correction, geometric correction are carried out to single image first,
Then same place is found according to flight POS data, generates point cloud model by empty three survey calculation raw videos, finally adds line
It manages and single channel image group is combined into multispectral image, obtain the multispectral image in corn planting region.
Preferably, textural characteristics parameter described in step 3 include: selection mean value, it is variance, concertedness, contrast, different
Property, comentropy, second moment, correlation.
Preferably, in step 1, blue and green light in corn planting region, feux rouges, infrared, close red is obtained by unmanned plane
The image data in outer five channels.
Preferably, the textural characteristics parameter after dimensionality reduction described in step 4 includes: red variance, blue variance, red Bian Fang
Difference, blue contrast, near-infrared variance, near-infrared contrast, red side contrast, blue diversity.
Preferably, in step 3 into using Relief algorithm or Principal Component Analysis row dimension-reduction treatment.
Preferably, the classification monitor model established in steps of 5 is Random Forest model or C4.5 model.
Preferably, the classification monitor model is Random Forest model, and the accuracy of identification of base decision tree quantity 5 is made
For the accuracy of identification of Random Forest model.
Preferably, after executing step 5, interpretation of result and precision are carried out to the classification monitor model established in step 5
Evaluation, and using precision ratio P, recall ratio R and F1 value as precision evaluation index, precision ratio P, recall ratio R and F1 value
Calculation formula is respectively as follows:
Wherein, TP indicates real example, and FN indicates false counter-example, and FP indicates false positive example, and TN indicates true counter-example.
Compared with prior art, the present invention has the beneficial effects that
1, it by the corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing, realizes being evenly distributed with for test area and adopts
Sample, while discrimination with higher.
2, the low latitude multispectral image obtained using unmanned aerial vehicle remote sensing, the RGB image that blue, green, red wave band synthesizes can be with
Effectively enhance the heterochromia of vegetation and exposed soil, shade by principal component analysis, realizes the segmentation of vegetation and background.
3, C4.5 decision-tree model and Random Forest model can effectively identify corn, weeds, accuracy of identification
Respectively up to 99.0%, 96.501%, compared to more traditional weed identification method, discrimination is greatly improved.
4, Principal Component Analysis Algorithm and Relief algorithm can be effectively to initial data dimensionality reductions, and PCA algorithm lays particular emphasis on list
One data identify accuracy, and Relief algorithm lays particular emphasis on the accuracy of identification of entire data set.C4.5 model and RF model treatment
The characteristic used time of Relief algorithms selection is respectively shortened less than the used time for handling PCA algorithm principal component, operation time
51.76%, 3.57%, efficiency with higher.
Detailed description of the invention
Fig. 1 is the corn weeds in field recognition methods flow chart based on unmanned aerial vehicle remote sensing.
Specific embodiment
Fig. 1 is highly preferred embodiment of the present invention, and 1 the present invention will be further described with reference to the accompanying drawing.
Embodiment 1:
As shown in Figure 1, a kind of corn weeds in field recognition methods (hereinafter referred to as weed identification side based on unmanned aerial vehicle remote sensing
Method), include the following steps:
Step 1, the multispectral image in corn planting region is obtained；
In corn planting region, image data is obtained by the Red Edge-M multispectral camera that unmanned plane is carried,
The sensor includes blue and green light, feux rouges, infrared, five channels of near-infrared, and unmanned plane is 30m, flying speed in flying height
Under the state of flight of 3m/s, image is acquired using vertical mode, five single channel images is obtained and amounts to 2630 width.It will be in maize seed
The image that growing area obtains carries out the pretreatment operations such as calibration, splicing by Agisoft PhotoScan software, first to single width figure
As carrying out radiant correction, geometric correction, same place is then found according to flight POS data, passes through the original shadow of empty three survey calculations
As generating point cloud model, finally adds texture and single channel image group is combined into multispectral image, obtain corn planting region
Multispectral image.
Step 2, image segmentation is carried out to the multispectral image in corn planting region；
Using Principal Component Analysis (Principal Component Analysis, PCA) to the more of corn planting region
Spectrum picture carries out image segmentation.When using Principal Component Analysis, the data in current orthogonal coordinate system are become by matrix
It changes operation and is mapped to a new coordinate system, what first new reference axis selected is the maximum direction of variance in initial data,
Second new reference axis selection is orthogonal with the first reference axis and has the direction of maximum variance, and reference axis selection repeats to sit until newly
Parameter number is equal with number of features in initial data.
Redundancy between each wave band of multispectral image can be effectively removed using Principal Component Analysis, reduce between each wave band
Correlation, by the several effective conversion wave bands of image information boil down to.The information that transformed each principal component of new wave band is included
The trend that tapers off is measured, first principal component includes maximum covariance information, and the information content of second and third principal component is successively reduced, until most
Several principal components, information content are almost nil afterwards.
Principal component analysis is carried out to the multispectral image of corn-growing regions, obtained the first, second and third principal component difference
Containing 92.57%, 5.99%, 1.43% information content, using first three most conversion wave band containing information content as red, green, blue
Wave band synthesis shows pseudo color image, and the pixel color between different atural objects shows biggish difference, and exposed soil, shade and vegetation
Pixel radiation brightness value is obvious in green light band difference, extracts exposed soil, shadows pixels region by setting up threshold value, is made as
Exposure mask file is applied to multispectral image, realizes the segmentation of vegetation and background.
Step 3, texture feature extraction；
Corn and weeds have bright in the pseudo color image for converting band combination after the processing of the Principal Component Analysis of step 2
Aobvious heterochromia, is found by on-site inspection, and part weeds fail to show color with maize leaf in pseudo color image
Difference, and weeds pixel is similar in each wave band to corn pixel radiance value, only in accordance with radiance value using threshold method without
Method distinguishes corn and weeds, therefore the textural characteristics of corn in selective extraction multispectral image, grass cutting blade are classified.
In this step, using gray level co-occurrence matrixes method as texture analysis method to converted wave after principal component analysis processing
The pseudo color image of Duan Zuhe carries out texture feature extraction.Gray level co-occurrence matrixes method is the second-order statistics measurement of image grayscale, instead
Image grayscale has been reflected in the integrated information of direction, neighborhood and amplitude of variation.In practical applications, it mostly uses based on gray scale symbiosis square
Characteristic parameter of the calculated statistic of battle array as texture recognition selects mean value, variance, collaboration in herbaceous weed recognition methods
8 textural characteristics parameters such as property, contrast, diversity, comentropy, second moment, correlation are analyzed.
Using ENVI software Co-occurrence-Measures function, corn, grass cutting blade to corn planting region
Textural characteristics extract, and set 64 for gray scale quality-class, filter window size is 5 × 5, spatial coherence matrix X, Y's
Variable quantity is 1, calculates separately above-mentioned 8 textural characteristics parameters under blue, green, red, red side, each wave band of near-infrared, is obtained
40 kinds of characteristic parameters.
Step 4, dimensionality reduction is carried out to textural characteristics parameter；
By step 3 it is found that if calculating separately selection mean value, side under five wave bands such as blue, green, red, red side, near-infrared
8 textural characteristics parameters such as difference, concertedness, contrast, diversity, comentropy, second moment, correlation can be obtained altogether 40 kinds
Characteristic parameter can be led if directly using input item of the 40 above-mentioned characteristic parameters as supervised classification model because of redundancy
Model efficiency decline is caused, operation time is increased.
In herbaceous weed recognition methods, dimension-reduction treatment is carried out to sample data using Relief algorithm.Relief algorithm is
A kind of filtering type feature selection approach, is mainly used for the feature selecting of two classification problems.This method uses ASSOCIATE STATISTICS metric
Feature importance, ASSOCIATE STATISTICS amount are vector, and component respectively corresponds an initial characteristics, bigger its corresponding spy of proof of component
It is stronger to levy classification capacity.
Sample data contains 55.39% through the first to the 8th principal component that principal component analysis obtains respectively, 17.28%,
12.60%, 3.83%, 3.00%, 1.58%, 1.01%, 0.93% information content is accumulated contribution rate 95.62%, can be solved
Most of variable in characteristic parameter matrix is released, therefore effectively replaces original 40 kinds of characteristic parameters.It is realized using Python
Relief feature selecting sorts to 40 kinds of characteristic parameters according to ASSOCIATE STATISTICS amount component size, then takes preceding 12 features: red
Variance, blue variance, it is red while variance, blue contrast, near-infrared variance, near-infrared contrast, it is red while contrast, blue it is different
Property, green variance, red vs' degree, green correlation, green contrast, it is preferred that first 8 in above-mentioned 12 features of selection
Feature realizes the dimensionality reduction to textural characteristics parameter.
Step 5, classification monitor model is established, weed identification is carried out；
In the present embodiment, it chooses Random Forest model (RandomForest, abbreviation RF model) and is used as supervised classification mould
Type carries out disaggregated model training to Random Forest model, carries out weeds knowledge in corn planting region by Random Forest model
Not.
Carrying out when establishing of Random Forest model, it is assumed that given sample set D={ xi,yi(i=1,2 ... it m) include m
A sample randomly selects T and an equal amount of sampling set D of original sample intersection D by autonomous sampling method with putting back tot=(t
=1,2 ... T), based on sampling set DtT base decision tree is trained, and the optimal dividing of each base decision tree node is by the knot
The subset comprising k attribute selected at random in point attribute set d determines, is referred to as k=log2d.T base decision tree is collected
At resulting assembled classifier, as random forest is learnt, the categorised decision of output uses the ballot method:
Wherein, H (x) indicates Random Forest model, htIt (x) is base decision-tree model, Y indicates output variable, and I () shows
Property function.
Step 6, interpretation of result and precision evaluation；
When carrying out interpretation of result and precision evaluation, precision evaluation is carried out based on Random Forest model classification results, is used
Precision ratio (P), recall ratio (R), each branch of F1 value 3 rely on index as precision:
In above formula, TP indicates real example, and FN indicates false counter-example, and FP indicates false positive example, and TN indicates true counter-example, and use is macro
Average value integrated survey accuracy rate and recall rate:
In formula, n indicates that cross validation number, Pi indicate that i-th verifies accuracy rate, and Ri indicates that i-th verifies recall rate.
In herbaceous weed recognition methods, using the corn of sample areas at 480, weeds as training sample, sample area at 120
Domain is as test sample.8 kinds of textural characteristics of each 5 wave bands of image are extracted, 40 kinds of characteristic parameters are amounted to, finally obtain 480 ×
40 characteristic parameter matrixes, then with preceding 8 textural characteristics data of Relief algorithm screening are denoted as A group data, Relief algorithm
Preceding 12 textural characteristics data of screening are denoted as B group data.Random Forest model is constructed using Python, random forest base is taken to determine
Plan tree quantity is 1,5,10,15, and respectively obtains the accuracy of identification of two groups of data of A, B, as shown in table 1:
| Data A/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| Random forest | 97.667% | 1.98 | 98.000% | 8.85 | 98.500% | 18.00 | 98.667% | 28.22 |
| Data B/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| Random forest | 97.500% | 2.98 | 98.000% | 13.62 | 98.833% | 27.79 | 99.000% | 41.16 |
1 test sample discrimination of table
By 1 data of table it is found that RF model accuracy of identification increases with the increase of base decision tree quantity, when base decision tree
When quantity is promoted to 10 by 1, accuracy of identification increases obviously, and as base decision tree quantity continues growing, accuracy of identification increases gradually
Slow down, the RF model of 10 base decision trees and the RF model accuracy of identification gap of 15 base decision trees are minimum, and with base decision tree
Quantity increases the growth of model running time.Comprehensively consider precision and operation time, using the accuracy of identification of base decision tree quantity 5 as
The accuracy of identification of RF model.
Two groups of data accuracy of identification discoveries of A, B are compared, are increased with B group data characteristics quantity, RF model is in base decision tree number
Accuracy of identification does not increase when amount is 1,5, and accuracy of identification slightly increases when base decision tree quantity is 10,15, shows the spy of B group addition
Parameter information redundancy is levied, fails to provide enough classification informations for two kinds of models, the first eight feature energy of Relief algorithm picks
40 features in enough effectively substitution initial data.
2 test sample precision evaluation of table
From the data in table 2, it can be seen that the precision ratio of Random Forest model, recall ratio be 96% or more, based on precision ratio with look into
The harmonic-mean F1 of full rate is 97% or more, therefore Random Forest model knows the classification of corn, weeds test sample collection
It does not work well.Wherein, when the base decision tree quantity of RF model is 10, highest recall ratio reaches when handling A group data
99.355%.
Embodiment 2:
The present embodiment difference from example 1 is that: in the present embodiment, carry out step 4 to textural characteristics
Parameter completed when dimensionality reduction using Principal Component Analysis, and when step 6 carries out interpretation of result and precision evaluation, will be using master
The table 1 in 1 and 2 data of table obtain according to being denoted as C group data, and in conjunction with the embodiments for preceding 8 number of principal components that componential analysis obtains
4 data of table 3 and table as follows:
| Data A/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| Random forest | 97.667% | 1.98 | 98.000% | 8.85 | 98.500% | 18.00 | 98.667% | 28.22 |
| Data B/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| Random forest | 97.500% | 2.98 | 98.000% | 13.62 | 98.833% | 27.79 | 99.000% | 41.16 |
| Data C/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| Random forest | 97.500% | 2.07 | 97.667% | 9.42 | 98.400% | 18.88 | 98.833% | 28.79 |
3 test sample discrimination of table
From the data in table 3, it can be seen that other than the conclusion of available 1 data of table, it is longitudinal to compare two groups of data identification essences of A, C
Degree discovery, the RF model accuracy of identification that base decision tree quantity is 1,5 when handling A group data do not improve compared with component C
0.167%, 0.333%, and the overall used time shortens 3.57% compared with C group.Show middle Relief algorithm employed in embodiment 1
Principal Component Analysis Algorithm is better than to the dimension-reduction treatment effect of initial data, but by Principal Component Analysis to the drop of initial data
The effect of weed identification method finally obtained when dimension processing compares the weed identification method of the prior art, still has larger excellent
Gesture.
4 test sample precision evaluation of table
From the data in table 4, it can be seen that other than the conclusion of available 2 data of table, when the base decision tree quantity of RF model is
When 15, there is highest precision ratio when handling C group data, up to 99.298%, has highest F1 when handling A group data, up to 99.018%；RF
When the base decision tree quantity of model is 10, there is highest recall ratio when handling A group data, up to 99.355%, shows through principal component point
The data of analysis algorithm process are conducive to improve model to the accuracy of corn, weed identification.
Embodiment 3:
The present embodiment the difference from embodiment 1 is that: in the present embodiment, when carrying out step 5, using C4.5 decision tree
Model is as supervised classification model, and during constructing decision tree, ID decision Tree algorithms are using information gain as Attribute transposition
Foundation.Assuming that the i-th class sample proportion is p in current sample intersection Di(i=1,2 ... m), then comentropy Info (D) is public
Formula are as follows:
Assuming that attribute A has j (j=1,2 ... n) a possible values { a1, a2 ... aj }, if drawing set D using attribute A
It is divided into j different classes, then jth class is a comprising all values on attribute A in set DnSample, be denoted as Dj.Then information increases
Beneficial Gain (D, a) formula are as follows:
Information gain criterion tend to can the more attribute of value, in order to reduce its for can the less attribute of value may
Classification influence, C4.5 decision Tree algorithms are improved on the basis of ID3 decision Tree algorithms, and selection is non-straight using ratio of profit increase
It connects use information gain and carries out Attribute transposition, ratio of profit increase formula are as follows:
Wherein:IV (a) is known as the eigenvalue of attribute A, as attribute A can value
When j is bigger, the value of IV (a) also be will increase.
In the present embodiment, after using C4.5 decision-tree model as supervised classification model, interpretation of result is carried out in step 6
When with precision evaluation, by preceding 8 number of principal components obtained using Principal Component Analysis according to being denoted as C group data, and in conjunction with the embodiments 1
In table 1 and 4 data of table 3 and table in 2 data of table and embodiment 2, obtain 6 data of table 5 and table as follows:
| Data A/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| C4.5 decision tree | 95.837% | 4.10 | / | / | / | / | / | / |
| Random forest | 97.667% | 1.98 | 98.000% | 8.85 | 98.500% | 18.00 | 98.667% | 28.22 |
| Data B/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| C4.5 decision tree | 95.669% | 8.74 | / | / | / | / | / | / |
| Random forest | 97.500% | 2.98 | 98.000% | 13.62 | 98.833% | 27.79 | 99.000% | 41.16 |
| Data C/ base decision tree quantity | 1 | Used time | 5 | Used time | 10 | Used time | 15 | Used time |
| C4.5 decision tree | 96.501% | 8.50 | / | / | / | / | / | / |
| Random forest | 97.500% | 2.07 | 97.677% | 9.42 | 98.440% | 18.88 | 98.833% | 28.79 |
5 test sample discrimination of table
As shown in Table 5, it other than the related conclusions of available table 1 and table 3, can also obtain as drawn a conclusion: RF model
Overall recognition accuracy it is high compared with C4.5 model, accuracy of identification reaches as high as 99.000%, shows that integrated study can learn in base
Accuracy of identification is further promoted on the basis of device.
Though accuracy of identification of the C4.5 model when handling 3 groups of data is lower than RF model, reach 95% or more, up to
96.501%.The C4.5 model calculation used time compares RF model and greatly reduces, it is contemplated that this can be used in the requirement quickly identified
C4.5 decision tree precision is as accuracy of identification result.Two groups of data accuracy of identification discoveries of longitudinal comparison A, C, C4.5 model are longitudinally right
Than A, C two groups of data accuracy of identification discovery, C4.5 model accuracy of identification when handling A group data compared with C group reduces 0.764%,
But operation time shortens 51.76% compared with C group.
Comprehensively consider accuracy of identification and operation use time, the recognition effect of C4.5 model and RF model when handling A group data
Better than C group, show that Relief algorithm is better than Principal Component Analysis Algorithm to the dimension-reduction treatment effect of initial data in this test.It is right
Than two groups of data accuracy of identification discoveries of A, B, increase with B group data characteristics quantity, C4.5 model accuracy of identification does not increase, RF model
When base decision tree quantity is 1,5, accuracy of identification does not increase, and accuracy of identification slightly increases when base decision tree quantity is 10,15, and two
The operation use time of kind model increases 113.17%, 49.80% compared with A group respectively, shows that the characteristic parameter information of B group addition is superfluous
It is remaining, fail to provide enough classification informations for two kinds of models, the first eight feature of Relief algorithm picks can effectively substitute original
40 features in beginning data.
6 test sample precision evaluation of table
From the data in table 6, it can be seen that other than the conclusion of available table 2 and table 4, C4.5 model and RF model look into standard
Rate, recall ratio are 94% or more, and the harmonic-mean F1 based on precision ratio and recall ratio is in 95%, two kind of model to jade
Rice, weeds test sample collection Classification and Identification work well.Precision ratio, recall ratio, the F1 of RF model are totally higher than C4.5 model.
Precision ratio when two kinds of model treatment A group data is below C group data, and recall ratio is above C group data, show through
The data of Principal Component Analysis Algorithm processing are conducive to improve model to the accuracy of corn, weed identification, at Relief algorithm
The data of reason are then conducive to improve model to the accuracy of identification of data entirety.
The above described is only a preferred embodiment of the present invention, being not that the invention has other forms of limitations, appoint
What those skilled in the art changed or be modified as possibly also with the technology contents of the disclosure above equivalent variations etc.
Imitate embodiment.But without departing from the technical solutions of the present invention, according to the technical essence of the invention to above embodiments institute
Any simple modification, equivalent variations and the remodeling made, still fall within the protection scope of technical solution of the present invention.
Claims (9)
1. a kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing, characterized by the following steps:
Step 1, in corn planting region, the image data in corn planting region is obtained by unmanned plane, and image data is carried out
Pretreatment, obtains the multispectral image in corn planting region；
Step 2, image segmentation is carried out using multispectral image of the Principal Component Analysis to corn planting region；
Step 3, using gray level co-occurrence matrixes method as texture analysis method to conversion band combination after principal component analysis processing
Pseudo color image carries out texture feature extraction, obtains textural characteristics parameter；
Step 4, dimension-reduction treatment is carried out to textural characteristics parameter obtained in step 3, the textural characteristics parameter after obtaining dimensionality reduction；
Step 5, classification monitor model is established, and classification monitor model is trained, by classification monitor model to maize seed
Weeds in growing area domain are identified.
2. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: step
Pretreated process described in 1 are as follows: radiant correction, geometric correction are carried out to single image first, then according to flight POS data
Same place is found, generates point cloud model by empty three survey calculation raw videos, finally adds texture and by single channel image group
It is combined into multispectral image, obtains the multispectral image in corn planting region.
3. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: step
Textural characteristics parameter described in 3 includes: selection mean value, variance, concertedness, contrast, diversity, comentropy, second moment, phase
Guan Xing.
4. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: in step
In rapid 1, blue and green light in corn planting region, feux rouges, infrared, five channels of near-infrared picture number are obtained by unmanned plane
According to.
5. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: step
Textural characteristics parameter after dimensionality reduction described in 4 includes: red variance, blue variance, red side variance, blue contrast, near-infrared
Variance, near-infrared contrast, red side contrast, blue diversity.
6. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: in step
Into using Relief algorithm or Principal Component Analysis row dimension-reduction treatment in rapid 3.
7. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: in step
The classification monitor model established in rapid 5 is Random Forest model or C4.5 model.
8. the corn weeds in field recognition methods according to claim 7 based on unmanned aerial vehicle remote sensing, it is characterised in that: described
Classification monitor model be Random Forest model, and using the accuracy of identification of base decision tree quantity 5 as the knowledge of Random Forest model
Other precision.
9. the corn weeds in field recognition methods according to claim 1 based on unmanned aerial vehicle remote sensing, it is characterised in that: holding
After row step 5, the progress interpretation of result of classification monitor model and precision evaluation to being established in step 5, and use precision ratio P,
Recall ratio R and F1 value as precision evaluation index, precision ratio P, recall ratio R and F1 value calculation formula be respectively as follows:
Wherein, TP indicates real example, and FN indicates false counter-example, and FP indicates false positive example, and TN indicates true counter-example.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910267513.1A CN109977901A (en) | 2019-04-03 | 2019-04-03 | A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910267513.1A CN109977901A (en) | 2019-04-03 | 2019-04-03 | A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109977901A true CN109977901A (en) | 2019-07-05 |
Family
ID=67082956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910267513.1A Pending CN109977901A (en) | 2019-04-03 | 2019-04-03 | A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109977901A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110726675A (en) * | 2019-09-03 | 2020-01-24 | 宿州山海经测地理信息科技有限公司 | Unmanned aerial vehicle remote sensing detection system and remote sensing detection method |
| CN114758256A (en) * | 2022-04-18 | 2022-07-15 | 云南这里信息技术有限公司 | Tobacco field weed identification method, device, equipment and storage medium |
| CN114820578A (en) * | 2022-05-26 | 2022-07-29 | 云南省烟草公司红河州公司 | Unmanned aerial vehicle remote sensing monitoring equipment for tobacco planting area |
| CN117115667A (en) * | 2023-10-23 | 2023-11-24 | 广东碧然美景观艺术有限公司 | Weed data acquisition method based on remote sensing image |
| CN117576560A (en) * | 2023-11-17 | 2024-02-20 | 中化现代农业有限公司 | Method, device, equipment and medium for identifying field weeds of northern spring corns |
| CN117788351A (en) * | 2024-02-27 | 2024-03-29 | 杨凌职业技术学院 | Agricultural remote sensing image correction method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170076438A1 (en) * | 2015-08-31 | 2017-03-16 | Cape Analytics, Inc. | Systems and methods for analyzing remote sensing imagery |
| CN107850567A (en) * | 2015-07-01 | 2018-03-27 | 株式会社岛津制作所 | Data processing equipment |
| CN108021890A (en) * | 2017-12-05 | 2018-05-11 | 武汉大学 | A kind of high score remote sensing image harbour detection method based on PLSA and BOW |
| CN109325431A (en) * | 2018-09-12 | 2019-02-12 | 内蒙古大学 | The detection method and its device of vegetation coverage in Crazing in grassland sheep feeding path |
| CN109522929A (en) * | 2018-10-16 | 2019-03-26 | 南京农业大学 | It is a kind of based on multi-source image fusion natural environment under wheat weeds recognition methods |
-
2019
- 2019-04-03 CN CN201910267513.1A patent/CN109977901A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107850567A (en) * | 2015-07-01 | 2018-03-27 | 株式会社岛津制作所 | Data processing equipment |
| US20170076438A1 (en) * | 2015-08-31 | 2017-03-16 | Cape Analytics, Inc. | Systems and methods for analyzing remote sensing imagery |
| CN108021890A (en) * | 2017-12-05 | 2018-05-11 | 武汉大学 | A kind of high score remote sensing image harbour detection method based on PLSA and BOW |
| CN109325431A (en) * | 2018-09-12 | 2019-02-12 | 内蒙古大学 | The detection method and its device of vegetation coverage in Crazing in grassland sheep feeding path |
| CN109522929A (en) * | 2018-10-16 | 2019-03-26 | 南京农业大学 | It is a kind of based on multi-source image fusion natural environment under wheat weeds recognition methods |
Non-Patent Citations (3)
| Title |
|---|
| 乔永亮 等: "基于多光谱图像和SVM的玉米田间杂草识别", 《农机化研究》 * |
| 刘继元 等: "《云南省第一次全国地理国情普查实施指南》", 30 April 2016 * |
| 李哲 等: "《建筑低空摄影》", 30 September 2016 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110726675A (en) * | 2019-09-03 | 2020-01-24 | 宿州山海经测地理信息科技有限公司 | Unmanned aerial vehicle remote sensing detection system and remote sensing detection method |
| CN114758256A (en) * | 2022-04-18 | 2022-07-15 | 云南这里信息技术有限公司 | Tobacco field weed identification method, device, equipment and storage medium |
| CN114820578A (en) * | 2022-05-26 | 2022-07-29 | 云南省烟草公司红河州公司 | Unmanned aerial vehicle remote sensing monitoring equipment for tobacco planting area |
| CN114820578B (en) * | 2022-05-26 | 2024-05-28 | 云南省烟草公司红河州公司 | Unmanned aerial vehicle remote sensing monitoring equipment for tobacco planting area |
| CN117115667A (en) * | 2023-10-23 | 2023-11-24 | 广东碧然美景观艺术有限公司 | Weed data acquisition method based on remote sensing image |
| CN117576560A (en) * | 2023-11-17 | 2024-02-20 | 中化现代农业有限公司 | Method, device, equipment and medium for identifying field weeds of northern spring corns |
| CN117788351A (en) * | 2024-02-27 | 2024-03-29 | 杨凌职业技术学院 | Agricultural remote sensing image correction method and system |
| CN117788351B (en) * | 2024-02-27 | 2024-05-03 | 杨凌职业技术学院 | Agricultural remote sensing image correction method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977901A (en) | A kind of corn weeds in field recognition methods based on unmanned aerial vehicle remote sensing | |
| Abdullahi et al. | Convolution neural network in precision agriculture for plant image recognition and classification | |
| CN107103306B (en) | Winter wheat powdery mildew remote sensing monitoring method based on wavelet analysis and support vector machine | |
| Mishra et al. | A Deep Learning-Based Novel Approach for Weed Growth Estimation. | |
| CN105608474B (en) | Region adaptivity plant extraction method based on high resolution image | |
| Pujari et al. | Classification of fungal disease symptoms affected on cereals using color texture features | |
| CN114766041A (en) | System and method for determining crop damage | |
| Tits et al. | The potential and limitations of a clustering approach for the improved efficiency of multiple endmember spectral mixture analysis in plant production system monitoring | |
| Ji et al. | In-field automatic detection of maize tassels using computer vision | |
| CN110163101A (en) | The difference of Chinese medicine seed and grade quick discrimination method | |
| CN116543316B (en) | Method for identifying turf in paddy field by utilizing multi-time-phase high-resolution satellite image | |
| Kumar et al. | An identification of crop disease using image segmentation | |
| Li et al. | An automatic approach for detecting seedlings per hill of machine-transplanted hybrid rice utilizing machine vision | |
| CN117079152A (en) | Fine crop classification extraction method and system based on satellite remote sensing image | |
| CN116863345A (en) | High-resolution image farmland recognition method based on dual attention and scale fusion | |
| CN113657158A (en) | Google Earth Engine-based large-scale soybean planting region extraction algorithm | |
| Zhang et al. | Robust image segmentation method for cotton leaf under natural conditions based on immune algorithm and PCNN algorithm | |
| Long et al. | Recognition of drought stress state of tomato seedling based on chlorophyll fluorescence imaging | |
| Meng et al. | Fine hyperspectral classification of rice varieties based on attention module 3D-2DCNN | |
| Nandhini et al. | PREDICTION OF CROP GROWTH USING MACHINE LEARNING BASED ON SEED FEATURES. | |
| Yang et al. | Classification of sugar beets based on hyperspectral and extreme learning machine methods | |
| Jin et al. | Detection and coverage estimation of purple nutsedge in turf with image classification neural networks | |
| Ma et al. | Monitoring defoliation rate and boll-opening rate of machine-harvested cotton based on UAV RGB images | |
| CN117152609A (en) | Crop appearance characteristic detecting system | |
| Calma et al. | Cassava Disease Detection using MobileNetV3 Algorithm through Augmented Stem and Leaf Images |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |