CN113989176B - Construction method of tobacco freshness She Chengshou degree judgment model - Google Patents
Construction method of tobacco freshness She Chengshou degree judgment model Download PDFInfo
- Publication number
- CN113989176B CN113989176B CN202010662576.XA CN202010662576A CN113989176B CN 113989176 B CN113989176 B CN 113989176B CN 202010662576 A CN202010662576 A CN 202010662576A CN 113989176 B CN113989176 B CN 113989176B
- Authority
- CN
- China
- Prior art keywords
- leaf
- image
- tobacco
- blade
- parameters
- 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.)
- Active
Links
- 235000002637 Nicotiana tabacum Nutrition 0.000 title claims abstract description 74
- 241000208125 Nicotiana Species 0.000 title claims abstract description 69
- 238000010276 construction Methods 0.000 title claims abstract description 12
- 238000000034 method Methods 0.000 claims abstract description 31
- 238000000605 extraction Methods 0.000 claims description 20
- 238000009826 distribution Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 17
- 238000013528 artificial neural network Methods 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 12
- 238000010219 correlation analysis Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 6
- 244000061176 Nicotiana tabacum Species 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 5
- 238000003062 neural network model Methods 0.000 claims description 5
- 238000005286 illumination Methods 0.000 claims description 4
- 238000000540 analysis of variance Methods 0.000 claims description 3
- 238000005520 cutting process Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 210000003462 vein Anatomy 0.000 claims description 3
- 238000009434 installation Methods 0.000 claims description 2
- 238000012417 linear regression Methods 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 8
- 210000002569 neuron Anatomy 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 241000196324 Embryophyta Species 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000010150 least significant difference test Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 241001270131 Agaricus moelleri Species 0.000 description 2
- 230000032683 aging Effects 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000035800 maturation Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000004576 sand Substances 0.000 description 2
- 239000000779 smoke Substances 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000003746 surface roughness Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 239000011800 void material Substances 0.000 description 2
- 241000238631 Hexapoda Species 0.000 description 1
- 241000607479 Yersinia pestis Species 0.000 description 1
- 229930002875 chlorophyll Natural products 0.000 description 1
- 235000019804 chlorophyll Nutrition 0.000 description 1
- ATNHDLDRLWWWCB-AENOIHSZSA-M chlorophyll a Chemical compound C1([C@@H](C(=O)OC)C(=O)C2=C3C)=C2N2C3=CC(C(CC)=C3C)=[N+]4C3=CC3=C(C=C)C(C)=C5N3[Mg-2]42[N+]2=C1[C@@H](CCC(=O)OC\C=C(/C)CCC[C@H](C)CCC[C@H](C)CCCC(C)C)[C@H](C)C2=C5 ATNHDLDRLWWWCB-AENOIHSZSA-M 0.000 description 1
- 235000019504 cigarettes Nutrition 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 239000000428 dust Substances 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003306 harvesting Methods 0.000 description 1
- 238000003703 image analysis method Methods 0.000 description 1
- 238000000338 in vitro Methods 0.000 description 1
- 238000003874 inverse correlation nuclear magnetic resonance spectroscopy Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000000391 smoking effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000001521 two-tailed test Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
- G06T2207/30188—Vegetation; Agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Radiology & Medical Imaging (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Manufacture Of Tobacco Products (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a construction method of a tobacco fresh She Chengshou degree judging model, which comprises the steps of firstly dividing leaves into three groups of upper, middle and lower tobacco leaves according to national standards, wherein each group comprises three types of underripe, ripe and overripe, and the total number of the three types is 9; and then, respectively extracting leaf color bias parameters, HSV (hue, saturation, value) tone parameters, leaf texture parameters, leaf profile parameters and other leaf characteristic parameters of three color channels and gray images of 9 types of leaf RGB images Red, green, blue through a standard image acquisition flow, taking the leaf characteristic parameters as input factors, constructing a leaf characteristic parameter-tobacco leaf appearance maturity judging model by using a multiple regression and BP (back propagation) network neural method, and comparing the influence of different parameter systems and different modeling modes on the fitting goodness and judging accuracy of the fresh tobacco appearance maturity judging model to preferably obtain an optimal fresh tobacco appearance maturity quantitative judging model, thereby providing a new realization path and application foundation for intelligent judgment of fresh tobacco appearance maturity.
Description
Technical Field
The invention relates to the field of tobacco detection, in particular to a construction method of a tobacco fresh She Chengshou-degree judgment model.
Background
Tobacco is one of main cash crops, and the degree of maturity of tobacco leaves is a key factor affecting the baking quality of the tobacco leaves and further affecting the sale price of the baked tobacco leaves. The insufficient maturity of the tobacco leaves not only affects the appearance character, the physical character, the proportion of the superior tobacco and the like of the tobacco leaves after baking, but also affects the chemical and overall quality of the tobacco leaves after smoking. At present, tobacco leaves with insufficient harvesting maturity still have common problems in flue-cured tobacco production, and further improvement of tobacco leaf quality is limited to a certain extent.
The traditional quantitative determination method for the maturity of the tobacco leaves respectively utilizes physical parameters (such as included angles of stems and leaves, thickness of the leaves and the like) and chemical parameters (SPAD and the like) of the leaves as indexes for measuring the maturity of the tobacco leaves, and the methods all require a large amount of manpower, are time-consuming and labor-consuming, and are easy to generate subjective errors. SPAD values (relative chlorophyll values can be determined) are commonly used in the laboratory to quantitatively describe the physiological maturity of fresh smoke. Because SPAD value measurement consumes time and labor, tobacco leaves are easy to damage, and the method is limited in field production and use. In actual production at the present stage, the appearance maturity of fresh cigarettes is distinguished mainly by means of manual visual recognition, but a large gap exists between the appearance maturity and the physiological maturity, so that the manual discrimination accuracy and the stability are low.
With the increasing maturity of digital image technology and the popularization of high-resolution image pickup equipment, the research of qualitative and quantitative description of plant appearance phenotype characteristics by using digital images is increasing. The digital camera can record the spectrum information of the visible wave band, and has high resolution and low cost. In addition, digital static color images contain a large amount of plant morphological structure and color information and are therefore often used to identify fruit or leaf maturity. The commonly used digital color image analysis method is an RGB color model, but the existing RGB color model is based on analysis of data under normal distribution conditions, only has R, G, B color gradation mean value parameters of three channels, and a leaf color normal distribution parameter set N1 formed by various combination parameters is researched and proposed, only 16 parameters can be obtained at maximum, the problem of small information quantity exists, the physiological significance of describing leaf color change of the leaf color can not be solved, and the application range of the leaf color normal distribution parameter set N1 is greatly limited.
Disclosure of Invention
The invention aims to provide a construction method of a tobacco fresh She Chengshou degree judgment model, which is used for enriching the information quantity of RGB models and improving the judgment accuracy of tobacco maturity.
The technical problems solved by the invention can be realized by adopting the following technical scheme:
A construction method of a tobacco freshness She Chengshou degree judgment model comprises the following steps:
Step one: classifying samples; picking up the same number of leaves at the upper, middle and lower parts of the multi-sample tobacco plant, and classifying according to the parts and maturity;
Step two: blade image acquisition; placing the blade picked in the step one on an image acquisition platform, and acquiring images by using a high-resolution digital camera;
step three: preprocessing a picture; cutting the original image acquired in the second step by Adobe Photoshop CS software, and only keeping the fresh leaf blade part;
Step four: extracting blade image characteristic parameters; carrying out parameter extraction on the image preprocessed in the step three by adopting MATLAB software, wherein the parameter extraction comprises leaf color bias parameter extraction, HSV tone parameter extraction, leaf surface texture parameter extraction and leaf profile parameter extraction, and constructing a leaf color image characteristic parameter set N2;
Step five: analyzing variance; taking fresh tobacco maturity as a classification factor, adopting SPSS software, and performing variance analysis on the leaf image characteristic parameters extracted in the step four by adopting an LSD method and a Duncan method;
Step six: correlation analysis; selecting all samples, and performing Person correlation analysis on fresh She Chengshou ℃ and the blade image characteristic parameters extracted in the step four by using SPSS software;
Step seven: constructing a judgment model; adopting a traditional leaf color normal distribution parameter set N1 and a leaf color image characteristic parameter set N2 constructed in the step four to respectively construct F1-F4 fresh She Chengshou degree judgment models; comprising the following steps: ① Adopting SPSS software, taking fresh leaf maturity as a dependent variable, taking an N1 and N2 parameter set as independent variables, and adopting a stepwise regression mode based on a least square method to respectively establish linear regression models F1 and F2; ② Adopting Neural Network Toolbox of MATLAB, taking fresh leaf maturity as an output factor, respectively taking N1 and N2 parameter sets as input layers, carrying out normalization processing on data, and then training through a neural network to construct BP neural network models F3 and F4;
Step eight: determining a judgment model; and F1-F4 model constructed in the step seven is used for judging the maturity of the sample, the judgment accuracy is calculated, and the F4 fresh She Chengshou degree judgment model constructed by the leaf color image characteristic parameter set N2 is determined through model judgment accuracy comparison.
Preferably, the sorting in the first step is performed indoors, and the time for picking up the leaves from the tobacco plants and sending the leaves into the room for sorting is less than 10 minutes.
Preferably, the table top size of the image acquisition platform in the second step is 300cm multiplied by 200cm multiplied by 80cm, and the color of the bottom plate is an off-white matte sand grinding table top; illumination light sources are uniformly distributed above the image acquisition platform, wherein the illumination light sources are 2 20W strip-shaped white LED lamp tubes, and the color temperature is 5000K; the installation height of the digital camera is 100cm away from the platform surface, and the resolution is 3840 x 5120; the veins of the blade are vertical to the edge of the table top of the image acquisition platform.
Preferably, the image cut in the step three is stored as a JPG image format with a white background.
Preferably, the extracting the blade image feature parameters in the fourth step specifically includes: ① Respectively acquiring 20 parameters of the mean value, the median, the mode, the skewness and the kurtosis of three channels of the blade color image Red, green, blue and the gray level image by adopting MATLAB built-in functions; ② Converting a target image RGB color model into an HSV color model by adopting MATLAB, and then calculating H, S, V parameters of the image through a built-in function; ③ Extracting texture parameters of the image by adopting MATLAB, converting the color image into a gray image, acquiring GLCM (global coordinate system) of the color image by using a built-in function, and calculating to obtain ASM, CON, COR, IDM parameters of texture characteristic values after normalization; ④ Acquiring connected domain information of a target blade by adopting a MATLAB built-in function, recording the blade area LA and the blade perimeter LP, approximately representing the blade length LL by the long side of the MBR, approximately representing the blade width LW by the short side of the MBR, and calculating the MBR area S, the target blade width-to-blade length ratio WL and the target blade area ratio SS by a formula to obtain LA, LP, LL, LW, WL and SS 6 parameters; ⑤ And summarizing the parameters obtained by ①～④ to form 33 blade image characteristic parameters, and constructing a blade image characteristic parameter table.
Preferably, the significant level of analysis of variance in step five, α=0.05.
Preferably, the significance check of the step six correlation analysis employs a two-tailed test.
The invention utilizes the leaf color deviation parameters, HSV color parameters, leaf shape parameters and leaf surface texture differences of tobacco leaf images with different maturity to firstly divide the leaves into three groups of upper, middle and lower tobacco leaves according to national standards, wherein each group comprises three types of underripeness, maturity and overripeness, and the total number of the three types is 9. And then, respectively extracting leaf color bias parameters, HSV (hue, saturation, value) tone parameters, leaf texture parameters, leaf profile parameters and other leaf characteristic parameters of three types of leaf RGB images Red, green, blue and gray images through a standard image acquisition flow, taking the leaf characteristic parameters as independent variables (input factors), constructing a leaf characteristic parameter-tobacco appearance maturity judging model by using a multiple regression and BP (back propagation) network neural method, and comparing the influence of different parameter systems and different modeling modes on the fitting goodness and judging accuracy of the fresh tobacco appearance maturity judging model to preferably obtain an optimal fresh tobacco appearance maturity quantitative judging model, thereby providing a new realization path and application foundation for intelligent judgment of fresh tobacco appearance maturity.
Drawings
FIG. 1 is a flow chart of the steps of the preliminary separation method of the present invention.
Fig. 2 is a leaf profile parameter illustration.
FIG. 3 is a graph showing a comparison of the fitting value and the actual value of the training model of the BP neural network under different input factors.
Detailed Description
The present invention will be described in detail below with reference to the drawings and specific examples, but embodiments of the present invention are not limited thereto.
Examples:
fig. 1 is a flowchart of the invention, a method for constructing a tobacco fresh She Chengshou degree judgment model, comprising the following steps:
Step one: classifying samples; after the cloud tobacco 87 varieties are selected for planting, 100 tobacco plants which are normal in field growth and consistent in growth vigor and have no plant diseases and insect pests are selected; meanwhile, 3 leaves at the upper part, the middle part and the lower part of each plant are picked, after the damaged leaves are removed, the rest 821 fresh tobacco leaves are sent into a room for fresh tobacco maturity sorting (fresh leaf in vitro time is less than 10 min), and the specific sorting quantity is shown in table 1:
TABLE 1 classification and quantity of test samples
Step two: blade image acquisition; adopting a high-resolution digital camera to collect images;
The platform used for image acquisition is a tobacco leaf standardized grading platform, the size of the platform surface is 300cm long, 200cm wide and 80cm high, and the color of the desktop bottom plate is an off-white (RGB value is 225:225:225) matte sand grinding platform surface. The lighting source is 2 20W strip-shaped white LED lamp tubes, the color temperature is 5000K, and the hanging positions of the lamp tubes are positioned at 1/4 and 3/4 of the platform, so that the light on the surface of the platform is uniform; the digital camera is fixed at a position 100cm away from the platform surface, and the camera is a Canon EOS-550D model digital camera of Canon company, and the resolution of an original digital image is 3840 x 5120. After dust and dew on the surface of the fresh leaves with the sorted maturity are wiped by the water absorbing paper, the samples are placed in the center of the platform for shooting, and the veins of the fresh leaves are ensured to be vertical to the edge of the tabletop.
Step three: preprocessing a picture; and cutting the original image by Adobe Photoshop CS software, only preserving the fresh leaf blade part, and storing the cut image into a JPG image format with a white background.
Step four: extracting blade image characteristic parameters; the method comprises leaf color deviation parameter extraction, HSV tone parameter extraction, leaf surface texture parameter extraction and leaf profile parameter extraction, and a leaf image characteristic parameter table is constructed.
(1) Leaf color deviation parameter extraction
20 Parameters including mean value, median, mode, skewness, kurtosis and the like of three channels of a blade color image Red, green, blue and a gray level image are respectively obtained by adopting MATLAB 2016R software (hereinafter referred to as MATLAB) built-in function (RMean,RMedian,RMode,RSkewness,RKurtosis,GMean,GMedian,GMode,GSkewness,GKurtosi,BMean,BMedian,BMode,BSkewness,BKurtosis,YMean,YMedian,YMode,YSkewness,YKurtosis).
(2) HSV hue parameter extraction
And converting the RGB color model of the target image into an HSV color model by adopting MATLAB, and calculating H, S, V of the image through a built-in function. In the HSV tone model, H represents the tone of an image, the range of values is [0,1], the image is displayed as yellow when H is 0.167, and the image is displayed as green when H is 0.333. S represents the saturation of the image color, the value range is [0,1], and the larger the value of S is, the darker the image color is. V represents the brightness of the image color, the value range is [0,1], and the larger the value of V is, the brighter the image color is.
(3) Leaf surface texture parameter extraction
The texture parameters of the image are extracted by MATLAB, and the specific operation is as follows: and converting the color image into a gray image, acquiring GLCM (global coordinate system) of the gray image through a built-in function, and calculating a texture characteristic value ASM, CON, COR, IDM of the gray image after normalization.
The object surface texture is typically represented by a Gray-level Co-occurrence Matrix, GLCM. GLCM refers to a common method of describing textures by studying the spatial correlation characteristics of gray scales, which describes the joint distribution of gray scales of two pixels with a certain spatial positional relationship. GLCM is denoted as P (i, j), and if the gray level is L, P (i, j) is an l×l matrix. The texture condition of an object surface is often quantitatively described by the following four parameters:
The energy (Angular Second Moment, ASM) is the sum of squares of all elements in the GLCM, with a range of values of 0, 1. ASM is used to represent texture thickness, with thicker textures having greater ASM values. The calculation formula is as follows:
the Contrast (CON) reflects the brightness of a pixel versus its neighbors. CON is used to indicate the degree of texture void depth, with the deeper texture void the greater the CON value. The calculation formula is as follows:
the autocorrelation (COR) reflects the Correlation between a pixel and its neighboring pixels, and the range of values is [ -1,1]. COR is used to denote the consistency of textures, the more consistent the texture, the greater the COR value. The calculation formula is as follows:
In the method, in the process of the invention,
The inverse distance difference (INVERSE DIFFERENT movement, IDM) reflects the homogeneity of the texture, measures the local variation of the image texture, and the value range is [0,1]. The larger the IDM value, the lack of variation between different regions of the image texture, and the locally very uniform. The calculation formula is as follows:
(4) Leaf profile parameter extraction
And acquiring connected domain information of the target blade by adopting an MATLAB built-in function, recording blade Area (LA) and blade circumference (LEAF PERIMETER, LP), and acquiring MBR related parameter information by adopting the MATLAB built-in function.
As shown in fig. 2, the minimum bounding rectangle (Minimum Bounding Rectangle, MBR) is a rectangle determined by taking four parameter values of the maximum abscissa, the minimum abscissa, the maximum ordinate and the minimum ordinate of each vertex of a given two-dimensional shape as boundary points, and is used for representing the maximum range of the two-dimensional shape. Since the blade images are all perpendicular to the edge of the desktop, the long side of the MBR is approximately used for representing the blade length (LEAF LENGTH, LL), the short side of the MBR is approximately used for representing the blade width (LEAF WIDTH, LW), and the MBR area S, the target blade width-to-blade length ratio WL and the target blade area ratio SS are calculated by the following formula:
S＝LL×LW
WL＝LW÷LL
SS＝LA÷S
(5) Construction of blade image characteristic parameter table
According to the steps (1) - (4) of the fourth step, a blade image characteristic parameter table is constructed to form 33 blade image characteristic parameters (namely, a blade color image characteristic parameter set N2), and the specific table is shown in table 2:
TABLE 2 blade image characteristic parameter Table
Step five: analyzing variance; in this embodiment, all 821 samples are selected, fresh smoke maturity is used as a classification factor, SPSS software is used, and an LSD (Least significant difference test) method and a Duncan (Duncan multiple coefficient detection) method are used to perform variance analysis on leaf image characteristic parameters, wherein the significant level α=0.05.
In the leaf color bias parameters, the mean value, the median and the mode reflect the shade of the leaf color, the bias reflects the deviation of the leaf color, and the kurtosis reflects the concentration of the leaf color distribution. As can be seen from table 3, tobacco RMean, RMedian, RMode, GMean, GMedian, GMode, BMean, BMedian, BMode, BSkewness, YMean, YMedian, YMode at the same position exhibits a significant increase with increasing maturity, reflected as gradual green loss of leaf color; RSKEWNESS, GKURTOSI, YSKEWNESS decreases significantly with increasing maturation, the degree of negative bias increases progressively, the leaf color tends progressively to yellow, but BSkewness behaves inversely. In the aspect of kurtosis, RKurtosis, GKurtosi, YKurtosis is firstly declined and then ascended along with the growth of leaf age, and is reflected on the change of leaf color, so that the whole aging process of changing the leaf color from bright green to yellow green (at the moment, the leaf color has the highest discrete degree and the smallest kurtosis) and finally changing the leaf color into yellow is corresponding.
As shown in table 3, in the leaf color HSV parameter, the chroma value (H) of the tobacco leaves at the same position is significantly reduced with the increase of the maturity of the tobacco leaves, which indicates that the leaf color of the tobacco leaves is gradually changed from green to yellow in the maturity process; meanwhile, the brightness value (V) is obviously increased along with the increase of the maturity of the tobacco leaves, which shows that the color of the tobacco leaves at the same part is gradually changed from deep to light in the maturity process. The saturation (S) is irregular, and no obvious difference exists between XUL and XOL and between CUL and CRL.
As shown in table 3, in the leaf surface texture parameters, except XRL, the energy (ASM) and Contrast (CON) of the tobacco leaves at the same position were significantly reduced with increasing maturity, indicating that the leaf surface roughness was gradually reduced and the texture grooves were gradually shallower with increasing maturity. Except XRL, the inverse distance difference (IDM) of tobacco leaves at the same part synchronously rises along with the increase of the maturity, which shows that the change of different areas of the leaf surface texture tends to be uniform along with the increase of the maturity, but no obvious difference exists between XUL and XOL and between BRL and BOL. The auto-Correlation (COR) of different maturity of tobacco leaves at the same part has no obvious difference, which indicates that the consistency of leaf surface textures of different maturity has no obvious change.
As shown in table 3, in the leaf profile parameters, the Leaf Area (LA), the Leaf Width (LW) and the leaf length-width ratio (WL) of the tobacco leaves among different positions are significantly different, and particularly, the Leaf Area (LA), the Leaf Width (LW) and the leaf length-width ratio (WL) among different maturity of the lower tobacco leaves are also significantly different, while the difference among different maturity of the middle and upper tobacco leaves is smaller.
Table 3 analysis of variance of blade image characteristic parameters for different fresh leaf maturity (n=821)
Step six: correlation analysis; all 821 samples are selected, and the SPSS software is used for carrying out Person related analysis on the fresh leaf maturity and the leaf image characteristic parameters, and the significance inspection adopts double-tail inspection.
As can be seen from table 4, among the 33 leaf image characteristic parameters, 27 parameters other than the 6 parameters of GMode, BMedian, BMode, BSkewness, BKurtosis, ASM were all significantly or extremely significantly correlated with the fresh leaf maturity of tobacco. In the category of parameters, among the leaf color bias parameters, the five parameters of the R channel have the highest correlation coefficient, the gray image parameters are represented the next time, and the B channel has the worst representation; except the mode of the B channel, the average value, the median and the mode of the R, G, B three channels and the gray level image are positively correlated with the maturity; among the HSV parameters, hue (H) parameter correlation coefficient is highest, and saturation (S) is poor; among the leaf texture parameters, the Contrast (CON) and inverse distance difference (IDM) have the highest correlation coefficient, and the energy (ASM) is too small to perform correlation analysis; among the leaf profile parameters, the correlation coefficients of Leaf Area (LA), leaf Width (LW), leaf width-to-leaf length ratio (WL) and fresh leaf maturity are respectively-0.832, -0.850 and-0.849, and are three items which are the best in all the parameters.
From the correlation coefficient, GMean, GMedian, GMode, BMean, BMedian, BMode, BSkewness, YMean, YMedian, YMode shows remarkable improvement along with the increase of the maturity, and reflects that the color of the leaf gradually loses green; RSKEWNESS, GKURTOSI, YSKEWNESS decreases significantly with increasing maturation, the degree of negative bias increases progressively, the leaf color tends progressively to yellow, but BSkewness behaves inversely. In the aspect of kurtosis, RKurtosis, GKurtosi, YKurtosis is firstly declined and then ascended along with the growth of leaf age, and is reflected on the change of leaf color, so that the whole aging process of changing the leaf color from bright green to yellow green (at the moment, the leaf color has the highest discrete degree and the smallest kurtosis) and finally changing the leaf color into yellow is corresponding.
As shown in table 4, in the HSV hue parameter, the chroma value (H) of the tobacco leaves at the same position is significantly reduced with the increase of the maturity of the tobacco leaves, which indicates that the tobacco leaf color is gradually changed from green to yellow in the maturity process; meanwhile, the brightness value (V) is obviously increased along with the increase of the maturity of the tobacco leaves, which shows that the color of the tobacco leaves at the same part is gradually changed from deep to light in the maturity process. The saturation (S) is irregular, and no obvious difference exists between XUL and XOL and between CUL and CRL.
As shown in table 4, in the leaf surface texture parameters, except XRL, the energy (ASM) and Contrast (CON) of the tobacco leaves at the same position were significantly reduced with increasing maturity, indicating that the leaf surface roughness was gradually reduced and the texture grooves were gradually shallower with increasing maturity. Except XRL, the inverse distance difference (IDM) of tobacco leaves at the same part synchronously rises along with the increase of the maturity, which shows that the change of different areas of the leaf surface texture tends to be uniform along with the increase of the maturity, but no obvious difference exists between XUL and XOL and between BRL and BOL. The auto-Correlation (COR) of different maturity of tobacco leaves at the same part has no obvious difference, which indicates that the consistency of leaf surface textures of different maturity has no obvious change.
As shown in table 4, in the leaf profile parameters, the Leaf Area (LA), the Leaf Width (LW) and the leaf length-width ratio (WL) of the tobacco leaves among different positions are significantly different, and particularly, the Leaf Area (LA), the Leaf Width (LW) and the leaf length-width ratio (WL) among different maturity of the lower tobacco leaves are also significantly different, while the difference among different maturity of the middle and upper tobacco leaves is smaller.
Table 4 correlation of fresh leaf maturity and leaf image characteristic parameters (n=821)
Note that: 1. the numbers in the table are followed by the expression that there is a significant correlation at the α=0.05 level; 2. the numbers in the table are followed by the symbols indicated as having a very significant correlation at the α=0.01 level; 3. ASM is too small to perform correlation analysis.
Step seven: decision model construction
The conventional leaf color normal distribution parameter set N1 and the leaf color image characteristic parameter set N2 (also called as leaf color bias distribution parameter set N2) are adopted to respectively construct F1-F4 fresh She Chengshou degree judgment models, as shown in Table 5:
TABLE 5 judgment model summary table
(1) Establishing a decision model independent variable parameter table
The parameter set N1 of the conventional leaf color normal distribution and the parameter set N2 based on the leaf color deviation distribution (namely, the 33 leaf image characteristic parameters constructed in the fourth step, table 2) are adopted as independent variables (input factors), and the specific parameter sets are shown in table 6:
TABLE 6 determination model argument list
(2) Regression model construction
SPSS software is adopted, fresh leaf maturity is taken as a dependent variable, N1 and N2 parameter sets are taken as independent variables, and a stepwise regression mode based on a least square method is adopted to respectively establish linear models F1 and F2. Both F1 and F2 model parameters were set to Properformance of F-to-enter < = 0.050,Probability of F-to-remove > = 0.100.
The main explanatory variables of the F1 model are BMean, H, S, V, LW, LP and CON seven parameters; the F2 model is mainly explained by RSKEWNESS, GSKEWNESS, RKURTOSIS, YKURTOSIS, LW, LP and H seven parameters together;
(3) BP neural network model construction
And adopting Neural Network Toolbox of MATLAB, taking fresh leaf maturity as an output factor, and respectively taking N1 and N2 parameter sets as input layers to construct judging models F3 and F4. The comparison of the fitting value and the actual value of the BP neural network training model under the condition of different input factors is shown in fig. 3, the left graph of fig. 3 is a schematic diagram of the comparison of the fitting value and the actual value of the BP neural network training model when the input factors are the leaf color normal parameter set N1, and the right graph is a schematic diagram of the comparison of the fitting value and the actual value of the BP neural network training model when the input factors are the leaf color bias parameter set N2.
The BP neural network model topology includes three layers, an input layer (input), an hidden layer (HIDE LAYER), and an output layer (output layer). The hidden layer neuron number is determined by the following empirical formula, and the hidden layer neuron number with the minimum judgment result and actual error is obtained after training and comparison to be the optimal neural network structure.
Wherein Z, m, n are the neuron numbers of the hidden layer, the input layer and the output layer respectively, q is a constant between [1, 10], and the result is rounded upwards.
First, normalization processing of data is performed. As the parameters reflecting the maturity of fresh leaves are more and the level difference of the original data of each parameter is obvious, the original data is normalized before BP neural network construction is carried out, and the data is linearly compressed to the range of [ -1,1], so that the network in the training stage is easier to converge.
Neural network training is then performed. The network middle layer neuron transfer function adopts Logsig functions, the output layer neuron transfer function adopts a linear function Purelin, and the training function adopts Trainlm. 70% of the data were used for model training, 15% for predictive validation, and 15% for model testing. The convergence error of the model was set to 0.0001, the learning rate to 0.05, and the maximum training time to 1000.
When a fresh She Chengshou-degree judgment model is built by using the BP neural network, when m=16, n=1 and q=7, Z=12, and the F3 convergence effect by adopting N1 as an input factor is optimal, so that the F3 optimal BP neural network structure is determined to be 16-12-1; when N2 is used as an input factor, z=10 when m=33, n=1, q=4, and the F4 convergence effect is optimal, thereby determining that the F4 optimal BP neural network structure is 33-10-1.
(4) And (3) constructing a final judgment model:
The final decision model is shown in table 7:
TABLE 7 fresh leaf maturity determination model
By comparing R 2 of the four models, R 2 of the four models exceeds 0.8, and the four models can be better fitted with fresh She Chengshou degrees; from the modeling mode, F3 (R 2＝0.947)、F4(R2 =0.966) constructed by using the BP neural network is better than F1 (R 2＝0.828)、F2(R2 =0.858) constructed by using stepwise regression; from the input parameter category, under the condition of the same modeling mode, F2 and F4 adopting the bias distribution parameters are respectively better than F1 and F3; in general, the BP neural network model F4 constructed by taking the leaf color image characteristic parameter set N2 as an input factor has the optimal fitting effect; and the stepwise regression model F1 constructed by adopting the leaf color normal distribution parameter set N1 as an independent variable has the worst fitting effect.
Step eight: model judgment accuracy comparison, determination judgment model
Determining 821 samples maturity by using F1-F4 (wherein F1-F4 uses rounding method to round predicted value of each sample as determination result), and calculating determination accuracy, the formula is as follows:
After F1-F4 was used to determine the maturity of 821 samples, it can be seen that F4 using N2 as the input factor has the highest accuracy of 94.15%, 14.25% higher than F3 using the same modeling method but with the input factor N1, and 52.74% higher than F1 with the lowest accuracy, as shown in table 8. From different maturity, F4 is optimal in the judging accuracy of 9 different maturity, wherein the judging accuracy of F4 to XRL, CRL, COL, BUL, BRL is more than 95%, and particularly the judging accuracy of BRL is 98%; f4 improves the accuracy of determination of 9 different maturity levels (XUL, XRL, XOL, CUL, CRL, COL, BUL, BRL, BOL) by 11.43, 10.52, 7.4, 27.85, 14.64, 18.94, 10.31, 18, 3.45, and 14.25 percent, respectively.
TABLE 8 accuracy of determination of degree of maturity of fresh leaves by different models
In summary, since the tone level distribution of the RGB model of the blade is the bias distribution, the information of the RGB model is greatly enriched based on the bias distribution parameters (mean, median, mode, skewness, kurtosis, etc.) extracted by the method. The invention describes the change condition of the leaf color from two aspects of leaf color depth and distribution uniformity, and improves the accuracy of individual character description and model prediction.
The foregoing description is only a preferred embodiment of the present invention, and is not intended to limit the technical scope of the present invention, so any minor modifications, equivalent changes and modifications made to the above embodiments according to the technical spirit of the present invention still fall within the scope of the present invention.
Claims (7)
1. The construction method of the tobacco freshness She Chengshou degree judgment model is characterized by comprising the following steps of:
Step one: classifying samples; picking up the same number of leaves at the upper, middle and lower parts of the multi-sample tobacco plant, and classifying according to the parts and maturity;
Step two: blade image acquisition; placing the blade picked in the step one on an image acquisition platform, and acquiring images by using a high-resolution digital camera;
step three: preprocessing a picture; cutting the original image acquired in the second step by Adobe Photoshop CS software, and only keeping the fresh leaf blade part;
Step four: extracting blade image characteristic parameters; carrying out parameter extraction on the image preprocessed in the step three by adopting MATLAB software, wherein the parameter extraction comprises leaf color bias parameter extraction, HSV tone parameter extraction, leaf surface texture parameter extraction and leaf profile parameter extraction, and constructing a leaf color image characteristic parameter set N2;
Step five: analyzing variance; taking fresh tobacco maturity as a classification factor, adopting SPSS software, and performing variance analysis on the leaf image characteristic parameters extracted in the step four by adopting an LSD method and a Duncan method;
Step six: correlation analysis; selecting all samples, and performing Person correlation analysis on fresh She Chengshou ℃ and the blade image characteristic parameters extracted in the step four by using SPSS software;
Step seven: constructing a judgment model; adopting a traditional leaf color normal distribution parameter set N1 and a leaf color image characteristic parameter set N2 constructed in the step four to respectively construct F1-F4 fresh She Chengshou degree judgment models; comprising the following steps: ① Adopting SPSS software, taking fresh leaf maturity as a dependent variable, taking an N1 and N2 parameter set as independent variables, and adopting a stepwise regression mode based on a least square method to respectively establish linear regression models F1 and F2; ② Adopting Neural Network Toolbox of MATLAB, taking fresh leaf maturity as an output factor, respectively taking N1 and N2 parameter sets as input layers, carrying out normalization processing on data, and then training through a neural network to construct BP neural network models F3 and F4;
Step eight: determining a judgment model; and F1-F4 model constructed in the step seven is used for judging the maturity of the sample, the judgment accuracy is calculated, and the F4 fresh She Chengshou degree judgment model constructed by the leaf color image characteristic parameter set N2 is determined through model judgment accuracy comparison.
2. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: the classification in the first step is carried out indoors, and the time for picking the leaves from tobacco plants and sending the leaves into the room for classification is less than 10 minutes.
3. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: step two, the table top size of the image acquisition platform is 300cm multiplied by 200cm multiplied by 80cm, and the bottom plate is an off-white matte sanding table top; illumination light sources are uniformly distributed above the image acquisition platform, wherein the illumination light sources are 2 20W strip-shaped white LED lamp tubes, and the color temperature is 5000K; the installation height of the digital camera is 100cm away from the platform surface, and the resolution is 3840 x 5120; the veins of the blade are vertical to the edge of the table top of the image acquisition platform.
4. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: and step three, saving the cut image as a JPG image format with a white background.
5. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: the step four of extracting the blade image characteristic parameters specifically comprises the following steps: ① Respectively acquiring 20 parameters of the mean value, the median, the mode, the skewness and the kurtosis of three channels of the blade color image Red, green, blue and the gray level image by adopting MATLAB built-in functions; ② Converting a target image RGB color model into an HSV color model by adopting MATLAB, and then calculating H, S, V parameters of the image through a built-in function; ③ Extracting texture parameters of the image by adopting MATLAB, converting the color image into a gray image, acquiring GLCM (global coordinate system) of the color image by using a built-in function, and calculating to obtain ASM, CON, COR, IDM parameters of texture characteristic values after normalization; ④ Acquiring connected domain information of a target blade by adopting a MATLAB built-in function, recording the blade area LA and the blade perimeter LP, approximately representing the blade length LL by the long side of the MBR, approximately representing the blade width LW by the short side of the MBR, and calculating the MBR area S, the target blade width-to-blade length ratio WL and the target blade area ratio SS by a formula to obtain LA, LP, LL, LW, WL and SS 6 parameters; ⑤ And summarizing the parameters obtained by ①～④ to form 33 blade image characteristic parameters, and constructing a blade image characteristic parameter table.
6. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: significant level of analysis of variance α=0.05 in step five.
7. The method for constructing the tobacco fresh She Chengshou degree judgment model according to claim 1, which is characterized by comprising the following steps: the significance check of the correlation analysis in the step six adopts a two-tail test.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010662576.XA CN113989176B (en) | 2020-07-10 | 2020-07-10 | Construction method of tobacco freshness She Chengshou degree judgment model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010662576.XA CN113989176B (en) | 2020-07-10 | 2020-07-10 | Construction method of tobacco freshness She Chengshou degree judgment model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113989176A CN113989176A (en) | 2022-01-28 |
| CN113989176B true CN113989176B (en) | 2024-05-31 |
Family
ID=79731255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010662576.XA Active CN113989176B (en) | 2020-07-10 | 2020-07-10 | Construction method of tobacco freshness She Chengshou degree judgment model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113989176B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102323221A (en) * | 2011-05-31 | 2012-01-18 | 河南农业大学 | Tobacco maturity detection method and device |
| CN104899424A (en) * | 2015-05-07 | 2015-09-09 | 中国烟草总公司郑州烟草研究院 | Microscopic feature based method for objectively testing ripeness and leaf structure of flue-cured tobacco leaves |
| CN110415181A (en) * | 2019-06-12 | 2019-11-05 | 勤耕仁现代农业科技发展(淮安)有限责任公司 | Flue-cured tobacco RGB image intelligent recognition and grade determination method under a kind of open environment |
| CN110705655A (en) * | 2019-11-05 | 2020-01-17 | 云南省烟草农业科学研究院 | Tobacco leaf classification method based on coupling of spectrum and machine vision |
-
2020
- 2020-07-10 CN CN202010662576.XA patent/CN113989176B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102323221A (en) * | 2011-05-31 | 2012-01-18 | 河南农业大学 | Tobacco maturity detection method and device |
| CN104899424A (en) * | 2015-05-07 | 2015-09-09 | 中国烟草总公司郑州烟草研究院 | Microscopic feature based method for objectively testing ripeness and leaf structure of flue-cured tobacco leaves |
| CN110415181A (en) * | 2019-06-12 | 2019-11-05 | 勤耕仁现代农业科技发展(淮安)有限责任公司 | Flue-cured tobacco RGB image intelligent recognition and grade determination method under a kind of open environment |
| CN110705655A (en) * | 2019-11-05 | 2020-01-17 | 云南省烟草农业科学研究院 | Tobacco leaf classification method based on coupling of spectrum and machine vision |
Non-Patent Citations (1)
| Title |
|---|
| 基于机器视觉技术的烤烟鲜烟叶成熟度检测;史龙飞;宋朝鹏;贺帆;段史江;王涛;王梅;宫锦;宫长荣;;湖南农业大学学报(自然科学版);20120825(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113989176A (en) | 2022-01-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103076288B (en) | A kind of flesh of fish self-grading device based on computer vision and method | |
| AU2006264799B2 (en) | Method and device for analysing berries | |
| CN102147402A (en) | Machine vision technology based method for rapidly detecting egg freshness | |
| CN110705655A (en) | Tobacco leaf classification method based on coupling of spectrum and machine vision | |
| CN106530346B (en) | A kind of crown image analysis method of cupressaceae plant | |
| CN109100323A (en) | A kind of transmitted spectrum harmless quantitative evaluation method of apple water core | |
| CN109409579A (en) | The method of BP neural network prediction Raw material processing suitability | |
| CN112634237A (en) | Long bamboo strip surface defect detection method and system based on YOLOv3 improved network | |
| CN114677671A (en) | Automatic identifying method for old ribs of preserved szechuan pickle based on multispectral image and deep learning | |
| Rafiq et al. | Application of computer vision system in food processing | |
| CN109242341A (en) | Method based on apple feedstock specifications prediction fruit juice integrated quality | |
| CN115512123A (en) | Multi-period key growth characteristic extraction and time period classification method for hypsizygus marmoreus | |
| CN108460380A (en) | A kind of bamboo cane method for sorting colors and system based on domain color | |
| CN113518182B (en) | Cucumber phenotype characteristic measuring method based on raspberry pie | |
| CN114092839B (en) | Unmanned aerial vehicle remote sensing-based soybean harvest period maturity judging method | |
| Motonaga et al. | A method of making digital fruit color charts for cultivation management and quality control | |
| CN107703064A (en) | A kind of method of natural color-cotton color grading | |
| CN113989176B (en) | Construction method of tobacco freshness She Chengshou degree judgment model | |
| Steinmetz et al. | Sorting cut roses with machine vision | |
| Fu et al. | Color based classification for berries of Japanese Blue Honeysuckle | |
| CN116563714A (en) | Method and system for automatically distinguishing growth stage of rice | |
| Doerflinger et al. | Objective assessment of dried sultana grape quality using digital image analysis | |
| Surový et al. | Importance of automatic threshold for image segmentation for accurate measurement of fine roots of woody plants/Význam automatického prahovania na obrazovú segmentáciu pre presné merania jemných koreňov drevín | |
| CN111563536B (en) | Bamboo strip color self-adaptive classification method based on machine learning | |
| CN113673456A (en) | Street view image scoring method based on color distribution learning |
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 |