CN111860567A - Construction method of blade image recognition model - Google Patents
Construction method of blade image recognition model Download PDFInfo
- Publication number
- CN111860567A CN111860567A CN202010388794.9A CN202010388794A CN111860567A CN 111860567 A CN111860567 A CN 111860567A CN 202010388794 A CN202010388794 A CN 202010388794A CN 111860567 A CN111860567 A CN 111860567A
- Authority
- CN
- China
- Prior art keywords
- leaf
- neural network
- blade
- mathematical model
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000010276 construction Methods 0.000 title claims abstract description 6
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000013178 mathematical model Methods 0.000 claims abstract description 22
- 238000013139 quantization Methods 0.000 claims abstract description 17
- 238000005457 optimization Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 9
- 238000013528 artificial neural network Methods 0.000 claims description 60
- 239000013598 vector Substances 0.000 claims description 25
- 238000000605 extraction Methods 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 12
- 230000009467 reduction Effects 0.000 claims description 6
- 241000288673 Chiroptera Species 0.000 claims description 4
- 230000015572 biosynthetic process Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 238000003786 synthesis reaction Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 2
- 241000196324 Embryophyta Species 0.000 description 19
- 230000008569 process Effects 0.000 description 10
- 238000012549 training Methods 0.000 description 9
- 238000010586 diagram Methods 0.000 description 5
- 241000694364 Acer capillipes Species 0.000 description 3
- 230000002068 genetic effect Effects 0.000 description 3
- 241000208140 Acer Species 0.000 description 2
- 230000007797 corrosion Effects 0.000 description 2
- 238000005260 corrosion Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 241000931526 Acer campestre Species 0.000 description 1
- 240000007187 Acer circinatum Species 0.000 description 1
- 235000006794 Acer circinatum Nutrition 0.000 description 1
- 241001106462 Ulmus Species 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008961 swelling Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a construction method of a blade image recognition model, which is characterized by comprising the following steps of: a. carrying out contour quantization processing on the batch leaf pictures to extract the characteristic information and explaining a quantization index system of the extracted characteristic information; b. establishing a mathematical model for judging the type of the leaves according to the extracted characteristic information, identifying core indexes based on a quantization index system, and evaluating the performance of the mathematical model and the influence of the core indexes on the judging performance of the mathematical model; c. and improving the established mathematical model by using an optimization algorithm according to the leaf texture information and the core index in the leaf picture. The invention aims to provide a method for constructing a leaf image identification model aiming at the defects of the prior art, which is used for establishing a mathematical model for classifying plants by means of leaf image information by extracting effective information in plant leaf images.
Description
Technical Field
The invention relates to the technical field of modern botany, in particular to a construction method of a leaf image recognition model.
Background
Plants are of various kinds, and it is very important for people to scientifically classify the plants. For plants, although local characteristics of roots, stems, flowers, fruits, seeds and the like of the plants have certain value for plant classification, the collection and treatment processes are troublesome, leaves of the plants are convenient to classify through the leaves of the plants due to the diversity of the shapes, and classification bases are also diverse, but the genetic relationship among the plants and the status in the phylogeny are often ignored only by considering a few obvious shapes of the plants. Therefore, the plants can be classified accurately and reasonably, and the understanding of the mutual relations among the plants becomes an important content in the taxonomic classification of the plants. However, the prior art lacks a tool for rapidly classifying based on the plant leaf image.
Disclosure of Invention
The invention aims to provide a method for constructing a leaf image identification model aiming at the defects of the prior art, which is used for establishing a mathematical model for classifying plants by means of leaf image information by extracting effective information in plant leaf images.
The invention provides a construction method of a blade image recognition model, which is characterized by comprising the following steps of:
a. carrying out quantization processing on the batch of leaf pictures to extract leaf contour characteristic information and explaining a quantization index system of the extracted characteristic information;
b. establishing a mathematical model for judging the type of the leaves according to the extracted characteristic information, identifying core indexes based on a quantization index system, and evaluating the performance of the mathematical model and the influence of the core indexes on the judging performance of the mathematical model;
c. and optimizing the established mathematical model according to the leaf texture information and the core index in the leaf picture.
In the above technical scheme, in the step a, edge information, texture information and geometric feature information in the leaf image are extracted by combining the binarized leaf image with relevant standard quantities of plant taxonomy.
In the above technical solution, in the step b, after the mathematical model is established, the performance of the model and the influence of each index on the performance of the model are evaluated by analyzing the speed and the accuracy of the mathematical model algorithm; and analyzing the accuracy of each index on leaf classification identification through a quantitative index system to identify a core index.
In the above technical solution, in the step c, based on the established mathematical model, the extracted data is subjected to dimensionality reduction by combining the texture information of the leaf picture; and solving the optimal parameter value of the mathematical model through an optimization algorithm by combining the core indexes.
In the above technical solution, the step a specifically includes the following steps: extracting pixel point coordinates of the leaf contour by adopting a bwpherem command in MATLAB, and carrying out binarization processing on the leaf image by using an im2bw command to generate a quantization index system; sorting the coordinate coordinates of the pixel points of the leaf contour in a clockwise direction, and forming a target boundary function of the leaf contour according to the sorted pixel point coordinates; and solving the elliptic Fourier descriptor to generate a blade edge characteristic information extraction result.
In the above technical scheme, in the step a, the geometric feature quantities of the 7 traditional botanical leaves classified according to the perimeter, the area, the minimum circumscribed rectangle, the rectangle degree, the elongation degree, the circularity degree and the density of the leaves are calculated according to the extracted leaf edge profile.
In the above technical solution, the step b specifically includes the following steps:
establishing a BP neural network with multiple hidden layers and multiple nodes by using an MATLAB BP neural network toolbox, wherein the total number of the hidden layers is three; wherein, the first layer has 100 nodes, the second layer has 2 nodes, and the third layer has one node;
Randomly extracting leaf data which comprise geometric characteristic quantities of 7 traditional botanical leaf classifications and 10-dimensional elliptic Fourier descriptors; the BP neural network learns by using the leaf data and the corresponding types thereof as learning samples, trains the BP neural network and realizes the mapping of the internal relation of the sample set;
inputting a single geometric characteristic quantity of the blade and a corresponding blade type into a BP bible network for retraining, and using the trained BP neural network for testing to determine the blade perimeter, a 10-dimensional elliptic Fourier descriptor and a comprehensive geometric characteristic as core indexes for describing the blade geometric characteristic; wherein the comprehensive geometric characteristics refer to the comprehensive description of the minimum circumscribed rectangle, the elongation, the blade area, the rectangle degree, the circularity and the density.
In the above technical solution, the step c specifically includes the following steps:
performing dimension reduction processing on the leaf information by using skeleton extraction, and optimizing a quantization index system;
optimizing the weight and the threshold of the BP neural network by using a bat algorithm, wherein the weight and the threshold of the BP neural network correspond to the position vectors of the bats in the algorithm, namely the position vector of each bat corresponds to a network structure, each component of the position vector represents a weight or a threshold, and the dimension of the position vector is equal to the synthesis of the number of the weights and the thresholds in the network;
And after the optimal weight and threshold of the BP neural network are obtained by using a bat algorithm, the BP neural network is retrained by using an optimized quantitative index system, and a finally improved blade image classification and identification model is obtained.
The invention adopts a BP neural network model to construct a blade image recognition model, and ensures the speed and accuracy of model recognition. The method uses the bat algorithm which integrates most advantages of group intelligent algorithms such as genetic algorithm, particle swarm algorithm and the like to optimally solve the weight and the threshold of the BP neural network, and further improves the speed and the precision of model identification. The invention uses the elliptic Fourier descriptor to extract useful information on the picture as much as possible, and reasonably reduces the dimension of data of a quantization index system consisting of core indexes, thereby solving the problem of difficult convergence caused by overhigh dimension of the data under the condition of ensuring minimum information loss.
Drawings
FIG. 1 is a schematic flow chart of the present invention
FIG. 2 is a diagram of leaf contour extraction
FIG. 3 is a comparison graph of the contour convex hull
FIG. 4 is a schematic flow chart of the Clockwise algorithm
FIG. 5 is a topology diagram of a BP network
FIG. 6 is a graph of BP neural network accuracy versus training times
FIG. 7 is a BP neural network toolbox results diagram
FIG. 8 is a Euclidean distance map of SOM neurons
FIG. 9 is a SOM cluster map
FIG. 10 is a chaotic matrix
FIG. 11 is an error curve
FIG. 12 is a schematic drawing of Ulmus Bergma framework extraction
FIG. 13 is an algorithm flow diagram of BA optimized BP
FIG. 14 is a graph of BP accuracy versus training times before improvement
FIG. 15 is a graph of BP accuracy and training times trend after BA optimization
FIG. 16 is a model run before improvement
Fig. 17 is a modified model operation diagram.
Detailed Description
The invention will be further described in detail with reference to the following drawings and specific examples, which are not intended to limit the invention, but are for clear understanding.
As shown in fig. 1, the present invention provides a method for constructing a blade image recognition model, which specifically includes the following steps:
firstly, establishing a leaf feature data quantization index system based on a plant leaf classification standard and a leaf edge feature extraction model of an elliptic Fourier descriptor; extracting an edge contour of a leaf picture by using a bwplaim command in MATLAB, and carrying out binarization processing on a leaf image by using an im2bw command;
then selecting to extract the features of the leaf outline by an elliptic Fourier descriptor based on Gaussian multi-scale analysis: the method comprises the steps of extracting a target boundary function (solving a convex hull of a blade) and calculating an elliptic Fourier descriptor (in order to meet the requirement that the edge contour point coordinates must be arranged according to a Clockwise sequence in the process of solving the elliptic Fourier descriptor, a Clockwise coordinate sorting algorithm is written in the specific embodiment). And calculating the geometrical characteristic quantities of the 7 traditional plant leaf classifications of the leaf perimeter, the leaf area, the minimum circumscribed rectangle, the rectangle degree, the elongation degree, the circularity degree and the density degree according to the extracted leaf edge profile. The geometric characteristic quantity of the leaves and the elliptic Fourier descriptors of the leaves form a quantization index system for identifying the leaf types.
The fourier descriptor can relate the binarized picture information as much as possible, but the concept of the obtained feature vector may not be explained from the perspective of plant taxonomy, and the following feature indexes are extracted in the embodiment in consideration of the geometric features of the leaves.
(1) Perimeter ρ, where perimeter refers to the perimeter of the outline of the closed leaf.
(2) Area A of the blade0And refers to the area of the white area in the picture.
(3) Minimum circumscribed rectangle Am
(4) Rectangular degree R
R=A0/Am
(5) Elongation E
Wherein lLong and longLength of minimum bounding rectangle,/Short lengthRepresenting the width of the minimum bounding rectangle.
(6) Degree of circularity DR
DR=4πA0/ρ2
(7) Compactness C0
C0=ρ2/A0
In summary, the feature vector composed of the fourier descriptors and the 7 geometric feature indexes are selected to form a quantization index system extracted from the data information.
Secondly, establishing a BP neural network with a plurality of hidden layers and a plurality of nodes by using a BP neural network toolbox in MATLAB, wherein the total number of the hidden layers is three, the first layer is provided with 100 nodes, the second layer is provided with 2 nodes, and the third layer is provided with one node;
the BP neural network learns by using randomly extracted leaf data (including 7 geometric characteristic quantities of traditional botany leaf classification and 10 elliptic Fourier descriptors) and leaf corresponding types as learning samples, trains the BP neural network, realizes mapping of internal relations of a sample set, and successfully and preliminarily establishes a BP neural network leaf recognition model.
The basic steps of the BP algorithm are as follows:
step 1: and initializing a structure, and giving initial values to the number of network nodes, the connection weight between each layer and the network learning rate.
Step 2: providing input values x of samples to a BP neural networkpAnd ideal output value dp。
Step 3: and sequentially calculating the output of the hidden layer and the output of the output layer according to a related mathematical formula of the basic principle of the BP neural network.
Step4 calculates WijAnd wijAnd updating the weight value correction value according to a related mathematical formula of the basic principle of the BP neural network.
Step5 calculates the accumulated error E of the learning sample by using the updated weight value.
Step6, judging whether the network error E meets the precision requirement, if so, ending, otherwise, continuing.
And Step7, updating the iteration times, if the iteration times are less than the maximum iteration times, turning to Step2, and otherwise, continuing. Step8 ends.
Through inputting single characteristic index data of the blade and the corresponding blade type into a network for testing, the perimeter of the blade, a 10-dimensional elliptic Fourier descriptor and comprehensive geometric characteristics (comprehensive description of the minimum circumscribed rectangle, elongation, blade area, rectangle degree, circularity degree and density) are determined as core indexes for describing the geometric characteristics of the blade.
And thirdly, optimizing the weight and the threshold of the BP neural network by adopting a bat algorithm with the advantages of group intelligent optimization algorithms such as a set genetic algorithm, a particle swarm algorithm and the like.
If all the obtained core indexes are used as the characteristic quantities of the leaves and input into the BP neural network, the accuracy of classification and identification can be theoretically improved, but the dimensionality of an input sample is too high, so that the network is difficult to converge, and therefore, the skeleton extraction is used for performing dimensionality reduction on the leaf information.
Optimizing the weight and the threshold of the BP neural network by using a bat algorithm, and enabling the weight and the threshold of the BP neural network to correspond to the position vectors of the bats, namely enabling the position vector of each bat to correspond to a network structure, wherein each component of the position vector represents a weight or a threshold, and the dimension of the position vector is equal to the synthesis of the number of the weights and the thresholds in the network.
And after the optimal weight and threshold of the BP neural network are obtained by using a bat algorithm, the improved BP neural network is reestablished by using the blade information after dimension reduction, and the blade image classification and identification are carried out.
In the specific embodiment, it is assumed that the original image is not damaged in the process of extracting the contour edge of the picture, and the information loss is small. It is assumed that the processing of the blade data does not have a large impact on the amount of original information. The following table illustrates the symbols in this particular example:
Symbol | Description of the invention |
Ω | Contour profile |
ρ | Circumference length |
A0 | Area of blade |
FDk | Fourier descriptor |
C0 | Compactness degree |
wii | Connection weight |
Am | Minimum circumscribed rectangle |
DR | Degree of circularity |
η | Learning rate |
f | Frequency of |
θj、θk | Threshold value |
p | Number of learning samples |
n,q,m | Number of nodes in each layer |
In a specific embodiment, the leaf stalks of most leaves in a given 1600 pictures are removed or only a short part of the leaves is reserved, so that the interference degree of the leaf stalks on the feature extraction of the leaves is negligible. The leaf contour is extracted using the bwgrind command in MATLAB. The extraction results are shown in fig. 2.
The method comprises the steps of obtaining 1600 binarized pictures of 100 leaves in the specific embodiment that outline images of the leaves are all in a closed shape, selecting feature extraction on the outline shapes of the leaves based on an elliptic Fourier descriptor under Gaussian multi-scale analysis, analyzing and comparing similarity of the leaf images on the whole and the details from multiple angles by means of the Gaussian multi-scale analysis, and finally achieving the purpose of analyzing the leaf shapes from multiple angles.
In the first step, the complex coordinate function is adopted as the boundary function to derive the Fourier descriptor with better retrieval performance, so the complex coordinate function of the point on the one-dimensional boundary line is selected as the target boundary function. The boundary line is traced clockwise with a certain point on the boundary line as a starting point, and each point on the boundary is expressed by a complex representation of u + jv. Regarding the leaf as a closed boundary composed of N points, from any point, clockwise around the boundary one circle, a complex sequence shown in formula (1) can be obtained:
s(n)=u(n)+jv(n),n=0,1,…,N-1 (1);
And performing linear convolution on the complex coordinate function s (n) and the Gaussian kernel function g (n, sigma) with the scale sigma to obtain an evolutionary curve s (n, sigma) based on different scales sigma.
the gaussian function has good distribution characteristic and low-pass filtering characteristic, so the convolution process has certain noise filtering property.
Fig. 3 is a comparison graph of the blade profile and the convex hull.
Assuming that the initial positions of the closed leaf contours are the same and are considered as a period of the periodic function, the specific embodiment expands by the fourier series of infinite sine-cosine waves.
The parameter equation of the profile omega on the complex plane is set as follows:
c(t)=x(t)+jy(t),
wherein:
x(t)∈{x1,x2,…,xm},
y(t)∈{y1,y2,…,ym},
t∈(0,2π]
the fourier series of curve c (t) in the (x, y) direction:
wherein:
wherein m is the number of edge points contained in the profile omega; t is a period, and T is 2 pi/m; ω is frequency, ω 1.
Coefficient k of curve c (t)xk,bxk,ayk,bykThe combination of (a) constitutes an elliptic fourier descriptor with translation, rotation and scale invariance:
considering that the high-frequency coefficient is susceptible to interference, N low-frequency fourier descriptors are used herein to describe the outline shape of the leaf, and a feature vector of the leaf shape is obtained:
FDk=[FD1,FD2,...,FDk]
by assigning N differently a plurality of times, it is finally determined that N is 10.
In order to meet the requirement that the edge contour point coordinates must be arranged according to a Clockwise sequence in the process of solving the elliptic Fourier descriptor, the Clockwise coordinate ordering algorithm is written. The basic process is as follows:
Step1 solves the coordinates of the mean of all points on the edge profile, and records the point as C.
Step2 is centered at point C and is aligned parallel to the y-axis.
Step3, calculating the included angle between the connecting line of any point on the contour and the point C and the directrix in Step2i。
Step4 sorts the angle values in order of small to large.
In this embodiment, image information feature extraction is performed on 1600 pictures of 100 leaves by MATLAB R2018a, and in view of space limitation, only information extraction results of various indexes of the first picture in the Acer Campestre are shown here, as shown in the following table:
characteristic index | Characteristic data |
Circumference length | 67926.1012081795 |
Area of blade | 197657 |
Minimum circumscribed rectangle | 384335 |
Degree of rectangularity | 0.514283112 |
Degree of circularity | 538.3301301 |
Compactness degree | 23.3432422092 |
FD1 | 1.771719413 |
FD2 | 1.039062873 |
FD3 | 1.013947963 |
FD4 | 1.110916822 |
FD5 | 1.02360539 |
FD6 | 1.033783607 |
FD7 | 1.031314686 |
FD8 | 1.047610439 |
FD9 | 1.022866683 |
FD10 | 1.022448654 |
Note: fDiFeature component vector representing Fourier descriptor
In the second step, the BP neural network with multiple hidden layers and multiple nodes is selected in this embodiment, and there are three hidden layers in total, where the first layer is 100 nodes, the second layer is 2 nodes, and the third layer is 1 node. After the structure is determined, the network learns by using the learning sample, updates and corrects the connection weight and the threshold of the network, and realizes the mapping of the internal relation of the sample set. The learning process of the BP network is divided into two stages:
and in the mode forward propagation stage, a learning sample is provided for the network, and then the output of each layer of neurons is calculated layer by layer from the first layer backwards by using the designed network structure.
And in the error reverse propagation stage, the gradient vector of the network error to the weight or the threshold is calculated layer by layer from the output layer of the network forward, and then the weight and the threshold are corrected according to the negative gradient direction.
The two phases are repeatedly and alternately executed until the algorithm converges. This learning process of error back-propagation for BP networks improves the ability of the network to process information.
According to the method, 1500 training sample data are input through a BP neural network toolbox in MATLAB, the running time is 538s, and the recognition accuracy is 90%. The BP neural network result classification comparison table is as follows:
actual kind of | BP |
3 | 2 |
20 | 20 |
21 | 21 |
22 | 22 |
23 | 23 |
24 | 24 |
25 | 25 |
35 | 35 |
36 | 36 |
37 | 37 |
3 | 2 |
20 | 20 |
21 | 21 |
Note: for example, in Table 2, the leaf species Acer Capillipes
The required precision is achieved after the model runs for 800 times, the algorithm runs for 298s, and the model performance is better on the whole.
Three types of leaves are selected in this embodiment: acer Campesre, Acer Capillipes and Acer _ Circinatum are used as training samples, 7 classes are classified by adopting the SOM neural network classification result, obviously, the classification result is inaccurate, the accuracy is low, and the classification accuracy is 43%. Therefore, the method has low precision and is obviously compared with the BP neural network result.
In this embodiment, the two leaves are taken as an example, 50 training samples are calculated in total, the recognition accuracy of the leaf classification is calculated through the LVQ neural network, and the comparison table of the classification results of the LVQ neural network is as follows (2 leaves):
Actual kind of | LVQ |
2 | 1 |
2 | 1 |
2 | 1 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 1 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 2 |
2 | 2 |
Note: 2 represents Acer Capillipes, and 1 represents Acer Campesre
As can be seen from fig. 11, the LVQ neural network has a high error when classifying the leaves, and the accuracy of the determination is 73%, and the model is not used because the accuracy of the method does not meet the requirement. Although the time for identifying the BP neural network is longer compared with that of the LVQ and SOM neural networks, the correct identification rate of the classification result of the BP neural network is as high as 92%, and the correct rates of the LVQ and SOM neural networks are gradually reduced with the increase of training sample data, while the BP neural network shows better stability.
The sample data corresponding to each single index is used as a training set, whether the sample data is a core index of the system or not is judged through the information contribution rate, the perimeter and the 10-dimensional Fourier descriptor are both the core indexes, and although the independent contribution rate of the other indexes is small, the comprehensive information contribution rate of the minimum circumscribed rectangle, the elongation, the blade area, the rectangle degree, the circularity and the density reaches 70%. Therefore, in the embodiment, the perimeter and the elliptic Fourier descriptor are selected, and the comprehensive geometric characteristic index is used as a core index.
Contribution rate of each index information
In the third step, the selected core index data adopted in the embodiment is a 64-dimensional vector, and there are 100 leaves and 1600 pictures in total, and when all the data are input into the BP neural network, the problem of too high dimensionality of the input sample occurs, so that the neural network is difficult to converge, and therefore the problem is to perform preprocessing on the data according to the idea of an elliptic fourier descriptor, so as to achieve the purpose of reducing dimensionality.
The Matlab framework extraction algorithm operates on an image by using two basic methods of 'corrosion' and 'expansion' based on gray-scale image morphology, and the essence of the two methods is to use an operator to carry out convolution operation on a graph matrix. This operator is divided into two types in total, the Flat type and the Nonflat type. Both erosion and dilation operations can be performed using one of two operators. The core formula is as follows:
flat type etching:
(f-B)(x,y)=min{f(x+i,y+j)},(i,j)∈B
nonflat type corrosion:
(f-G)(x,y)=min{f(x+i,y+i)-G(i,j)},(i,j)∈G
flat type expansion:
swelling of the Nonflat type:
(Sk (x) is the result, the original binary image)
The batlization of the weight and the threshold of the BP neural network refers to that the weight and the threshold of the BP neural network correspond to the position vectors of the 'bats', namely the position vector of each 'bat' corresponds to a network structure, each component of the position vector represents a weight or a threshold, and the dimension of the position vector is equal to the synthesis of the number of the weights and the thresholds in the network. And if the number of nodes of each layer of the 3-layer BP network is n, q and m respectively, the network represented by the position vector of the ith bat is as follows:
x=(xi1,xi2,…,xid)
=(w11,…,w1q,wn1,…,w11,…w1m,wq1,…,wqm,θ1,…,θq,θ′1,…,θ′1)′.
Wherein d is nq + qm + q + m, wij(i-1, 2, …, n; j-1, 2, …, q) represents the connection weight between the input layer and the hidden layer, θj、θkIndicating a threshold between layers; after the weight and the threshold are batched, the objective function formula of BA is as follows:
wherein x isiPosition vector representing bat i, number of n-type learning samples, Oih, TihRespectively representing the network output and the ideal output of the h-th learning sample under the network structure determined by the bat i.
At this time, the update iteration of the bat position vector corresponds to the update of the weight and the threshold, and the optimal weight and the threshold of the BP neural network are obtained by utilizing the process of finding the optimal solution by the bat individual in the algorithm.
FIG. 15 shows training data obtained by using the second question data for the improved modelThe relationship between the exercise frequency and the precision is shown. From fig. 15, it can be known that, under the condition that the optimal weight and the threshold of the BP neural network are obtained through bat algorithm optimization, when the number of iterations is 972, the accuracy of the improved BP neural network model reaches 5.4793 × 10-9The recognition accuracy was 95%. The value is higher than the value of 90% before optimization, and the two program operation result graphs show that the convergence speed of the BP neural network is improved by using the optimization algorithm, the time required by the BP neural network for identifying the blade type is shortened by about 120s, and the performance of the optimized model is further improved.
Details not described in this specification are within the skill of the art that are well known to those skilled in the art.
Claims (8)
1. A construction method of a blade image recognition model is characterized by comprising the following steps:
a. carrying out quantization processing on the batch of leaf pictures to extract leaf contour characteristic information and explaining a quantization index system of the extracted characteristic information;
b. establishing a mathematical model for judging the type of the leaves according to the extracted characteristic information, identifying core indexes based on a quantization index system, and evaluating the performance of the mathematical model and the influence of the core indexes on the judging performance of the mathematical model;
c. and improving the established mathematical model by using an optimization algorithm according to the leaf texture information and the core index in the leaf picture.
2. The method for constructing the leaf image recognition model according to claim 1, wherein in the step a, edge information, texture information and geometric feature information in the leaf image are extracted by combining relevant standard quantities of plant taxonomy after the leaf image is binarized.
3. The method for constructing the blade image recognition model according to claim 2, wherein in the step b, after the mathematical model is established, the performance of the model and the influence of each index on the performance of the model are evaluated by analyzing the speed and the accuracy of the mathematical model algorithm; and identifying the core index by analyzing the accuracy of each index in the quantitative index system on leaf classification identification.
4. The method for constructing a leaf image recognition model according to claim 3, wherein in the step c, on the basis of the established mathematical model, the extracted core index data is subjected to dimension reduction processing by combining texture information of a leaf picture; and solving the optimal parameter value of the mathematical model through an optimization algorithm.
5. The method for constructing the blade image recognition model according to claim 2, wherein the step a specifically comprises the following steps: extracting pixel point coordinates of the leaf contour by adopting a bwpherem command in MATLAB, and carrying out binarization processing on the leaf image by using an im2bw command; sorting the coordinate coordinates of the pixel points of the leaf contour in a clockwise direction, and forming a target boundary function of the leaf contour according to the sorted pixel point coordinates; and solving the elliptic Fourier descriptor to generate a blade edge characteristic information extraction result.
6. The method for constructing a leaf image recognition model according to claim 5, wherein in the step a, the geometric feature quantities of 7 traditional botanical leaf classification including the leaf perimeter, the leaf area, the minimum circumscribed rectangle, the rectangle degree, the elongation degree, the circularity degree and the compactness degree are calculated according to the extracted leaf edge profile; the geometric characteristic quantity of the leaves and the elliptic Fourier descriptors of the leaves form a quantization index system for identifying the leaf types.
7. The method for constructing the blade image recognition model according to claim 3, wherein the step b specifically comprises the following steps:
establishing a BP neural network with multiple hidden layers and multiple nodes by using a BP neural network toolbox in MATLAB, wherein the total number of the hidden layers is three; wherein, the first layer has 100 nodes, the second layer has 2 nodes, and the third layer has one node;
randomly extracting leaf data which comprise geometric characteristic quantities of 7 traditional botanical leaf classifications and 10-dimensional elliptic Fourier descriptors; the BP neural network uses the leaf data and the corresponding types thereof as learning samples to learn, trains the BP neural network, realizes the mapping of the internal relation of the sample set, and initially establishes a BP neural network leaf recognition model;
inputting a single geometric characteristic quantity of the blade and a corresponding blade type into a BP neural network for retraining, and using the trained BP neural network for testing to determine the blade perimeter, a 10-dimensional elliptic Fourier descriptor and a comprehensive geometric characteristic index as core indexes for characterizing the blade; wherein the comprehensive geometric characteristic index refers to comprehensive description of minimum circumscribed rectangle, elongation, blade area, rectangle degree, circularity and density.
8. The method for constructing the blade image recognition model according to claim 4, wherein the step c specifically comprises the following steps:
performing dimensionality reduction on the leaf information by using skeleton extraction, and optimizing an original quantization index system;
optimizing the weight and the threshold of the BP neural network by using a bat algorithm, wherein the weight and the threshold of the BP neural network correspond to the position vectors of the bats in the algorithm, namely the position vector of each bat corresponds to a network structure, each component of the position vector represents a weight or a threshold, and the dimension of the position vector is equal to the synthesis of the number of the weights and the thresholds in the network;
and after the optimal weight and threshold of the BP neural network are obtained by using a bat algorithm, retraining the improved BP neural network by using an optimized quantitative index system, and obtaining a final improved blade image classification recognition model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010388794.9A CN111860567B (en) | 2020-05-09 | 2020-05-09 | Construction method of blade image recognition model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010388794.9A CN111860567B (en) | 2020-05-09 | 2020-05-09 | Construction method of blade image recognition model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111860567A true CN111860567A (en) | 2020-10-30 |
CN111860567B CN111860567B (en) | 2024-04-16 |
Family
ID=72985259
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010388794.9A Active CN111860567B (en) | 2020-05-09 | 2020-05-09 | Construction method of blade image recognition model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111860567B (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106022343A (en) * | 2016-05-19 | 2016-10-12 | 东华大学 | Fourier descriptor and BP neural network-based garment style identification method |
CN109543572A (en) * | 2018-11-07 | 2019-03-29 | 北京交通大学 | A kind of traveling pavement condition evaluation method |
CN110287882A (en) * | 2019-06-26 | 2019-09-27 | 北京林业大学 | A kind of big chrysanthemum kind image-recognizing method based on deep learning |
-
2020
- 2020-05-09 CN CN202010388794.9A patent/CN111860567B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106022343A (en) * | 2016-05-19 | 2016-10-12 | 东华大学 | Fourier descriptor and BP neural network-based garment style identification method |
CN109543572A (en) * | 2018-11-07 | 2019-03-29 | 北京交通大学 | A kind of traveling pavement condition evaluation method |
CN110287882A (en) * | 2019-06-26 | 2019-09-27 | 北京林业大学 | A kind of big chrysanthemum kind image-recognizing method based on deep learning |
Also Published As
Publication number | Publication date |
---|---|
CN111860567B (en) | 2024-04-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111178432B (en) | Weak supervision fine granularity image classification method of multi-branch neural network model | |
CN110321830B (en) | Chinese character string picture OCR recognition method based on neural network | |
CN106960214A (en) | Object identification method based on image | |
CN110348399B (en) | Hyperspectral intelligent classification method based on prototype learning mechanism and multidimensional residual error network | |
CN111414849B (en) | Face recognition method based on evolution convolutional neural network | |
US20050100209A1 (en) | Self-optimizing classifier | |
WO2018052587A1 (en) | Method and system for cell image segmentation using multi-stage convolutional neural networks | |
CN108875933B (en) | Over-limit learning machine classification method and system for unsupervised sparse parameter learning | |
CN110197205A (en) | A kind of image-recognizing method of multiple features source residual error network | |
CN111368885A (en) | Aero-engine gas circuit fault diagnosis method based on deep learning and information fusion | |
CN106845528A (en) | A kind of image classification algorithms based on K means Yu deep learning | |
CN115410088B (en) | Hyperspectral image field self-adaption method based on virtual classifier | |
CN111898621A (en) | Outline shape recognition method | |
CN116071560A (en) | Fruit identification method based on convolutional neural network | |
CN114782761A (en) | Intelligent storage material identification method and system based on deep learning | |
CN109740672B (en) | Multi-stream feature distance fusion system and fusion method | |
CN113191361B (en) | Shape recognition method | |
CN113128518B (en) | Sift mismatch detection method based on twin convolution network and feature mixing | |
CN112509017B (en) | Remote sensing image change detection method based on learnable differential algorithm | |
CN112990130A (en) | Data cleaning method and device for training sample and classifier | |
CN114220164A (en) | Gesture recognition method based on variational modal decomposition and support vector machine | |
CN111860567B (en) | Construction method of blade image recognition model | |
CN114067314B (en) | Neural network-based peanut mildew identification method and system | |
Li et al. | Weed identification based on shape features and ant colony optimization algorithm | |
CN115100694A (en) | Fingerprint quick retrieval method based on self-supervision neural network |
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 |