CN113688959A - Plant disease and insect pest diagnosis method and system based on artificial intelligence - Google Patents

Plant disease and insect pest diagnosis method and system based on artificial intelligence Download PDF

Info

Publication number
CN113688959A
CN113688959A CN202111249226.1A CN202111249226A CN113688959A CN 113688959 A CN113688959 A CN 113688959A CN 202111249226 A CN202111249226 A CN 202111249226A CN 113688959 A CN113688959 A CN 113688959A
Authority
CN
China
Prior art keywords
plant
plant sample
labeled
sample image
image
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
Application number
CN202111249226.1A
Other languages
Chinese (zh)
Other versions
CN113688959B (en
Inventor
赵建忠
刘璇
赵志高
夏国卿
赵建荣
夏桂华
夏桂云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shouguang Defeng Ecological Agriculture Co ltd
Original Assignee
Shouguang Defeng Ecological Agriculture Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shouguang Defeng Ecological Agriculture Co ltd filed Critical Shouguang Defeng Ecological Agriculture Co ltd
Priority to CN202111249226.1A priority Critical patent/CN113688959B/en
Publication of CN113688959A publication Critical patent/CN113688959A/en
Application granted granted Critical
Publication of CN113688959B publication Critical patent/CN113688959B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biophysics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

The invention relates to a plant disease and insect pest diagnosis method and system based on artificial intelligence, and belongs to the technical field of plant disease and insect pest diagnosis. The method comprises the following steps: obtaining disease style characteristics of each plant sample image with the label according to the color texture distribution map; training a variational self-encoder by utilizing each plant sample image with a label to obtain a Gaussian model corresponding to each plant sample image to be labeled; according to the Gaussian model, distributing labels for the plant sample images to be labeled to obtain label data corresponding to the plant sample images to be labeled; training a target network according to the plant sample images with labels, the label data corresponding to the plant sample images with labels, the plant sample images to be labeled and the label data corresponding to the plant sample images to be labeled; and inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image. The invention can improve the accuracy of diagnosing plant infection diseases and insect pests by the target network.

Description

Plant disease and insect pest diagnosis method and system based on artificial intelligence
Technical Field
The invention relates to the technical field of plant disease and insect pest diagnosis, in particular to a plant disease and insect pest diagnosis method and system based on artificial intelligence.
Background
Plant diseases and insect pests refer to diseases and insect pests infected during the growth of plants, the diseases and insect pests of the plants are mainly fungal diseases, bacterial diseases and the like, the insect pests of the plants are mainly insects, mites, snails and the like, and the infected plant diseases and insect pests of the plants seriously affect the normal growth and results of the plants and may also cause the death of the plants.
The existing plant disease and insect pest diagnosis method is generally used for diagnosing whether plant diseases and insect pests occur or not based on manual or neural networks, the mode based on manual diagnosis is complicated and omission may occur, and the mode based on neural network diagnosis needs a professional specially knowing plant diseases and insect pests to label, because the labeling cost and energy are considered, the scale of the finally obtained data set with labels is not large, and the accuracy of the mode for diagnosing plant diseases and insect pests based on neural network is low.
Disclosure of Invention
The invention provides a plant disease and insect pest diagnosis method and system based on artificial intelligence, which are used for solving the problem that plant disease and insect pest cannot be accurately diagnosed in the prior art, and adopt the following technical scheme:
in a first aspect, an embodiment of the present invention provides an artificial intelligence-based plant disease and pest diagnosis method and system, comprising the following steps:
acquiring a plant sample image set with a label, a plant sample image set to be labeled and a target plant image; the labeled plant sample image set comprises plant sample images of different state categories with labels; the different status categories comprise different pest categories and pest-free categories;
obtaining a color texture distribution diagram corresponding to each plant sample image with the label according to each plant sample image with the label; obtaining disease style characteristics of each plant sample image with the label according to the color texture distribution map;
training a variational self-encoder by utilizing the plant sample images with the labels to obtain Gaussian models corresponding to the plant sample images to be labeled; the loss function of the variation self-encoder is constructed by the disease style characteristics of each plant sample image with labels and a Gaussian model; the variation self-encoder is used for extracting disease style characteristics of each plant sample image with a label and distributing the label for the plant sample image to be labeled;
according to the Gaussian model, distributing labels for the plant sample images to be labeled to obtain label data corresponding to the plant sample images to be labeled;
training a target network according to each plant sample image with a label, label data corresponding to each plant sample image with the label, each plant sample image to be labeled and the label data corresponding to each plant sample image to be labeled, wherein the target network is used for diagnosing whether the plant is infected with the plant diseases and insect pests and the corresponding specific plant disease and insect pest type when the plant is infected with the plant diseases and insect pests;
and inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image.
The invention also provides a plant disease and insect pest diagnosis system based on artificial intelligence, which comprises a memory and a processor, wherein the processor executes a computer program stored in the memory so as to realize the plant disease and insect pest diagnosis method based on artificial intelligence.
According to the color texture distribution map, acquiring disease style characteristics of each plant sample image with labels; training a variational self-encoder by utilizing each plant sample image with a label to obtain a Gaussian model corresponding to each plant sample image to be labeled; according to the Gaussian model, distributing labels for the plant sample images to be labeled to obtain label data corresponding to the plant sample images to be labeled; training a target network according to the plant sample images with labels, the label data corresponding to the plant sample images with labels, the plant sample images to be labeled and the label data corresponding to the plant sample images to be labeled; and inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image. According to the invention, each plant sample image with a label is used as a basis for training the variational self-encoder, the trained variational self-encoder is used as a basis for obtaining label data corresponding to each plant sample image to be labeled, and each plant sample image with a label, the label data corresponding to each plant sample image with a label, each plant sample image to be labeled and the label data corresponding to each plant sample image to be labeled are used as a basis for training the target network.
Preferably, the method for obtaining the color texture distribution map corresponding to each labeled plant sample image according to each labeled plant sample image comprises:
removing the illumination information and the texture information on each plant sample image with the label to obtain an RGB image without illumination and texture corresponding to each plant sample image with the label;
segmenting the RGB image without illumination and texture by utilizing a superpixel segmentation algorithm to obtain a color distribution map corresponding to each labeled plant sample image;
carrying out edge detection on the plant sample images with the labels by using an edge detection algorithm to obtain edge distribution maps corresponding to the plant sample images with the labels;
and obtaining a color texture distribution diagram corresponding to each labeled plant sample image according to the color distribution diagram and the edge gray level distribution diagram.
Preferably, the method for obtaining the disease style characteristics of each labeled plant sample image according to the color texture distribution map comprises the following steps:
processing the color texture distribution map by using a style migration algorithm to obtain a Gram matrix corresponding to each color texture distribution map;
and taking the Gram matrix as a disease style characteristic corresponding to each plant sample image with the label.
Preferably, the method for training the variational self-encoder by using the labeled plant sample images comprises the following steps:
and respectively inputting the plant sample images with the labels into a variation self-encoder, and training the variation self-encoder by utilizing a reconstruction loss function, an intra-class disease style loss function, an inter-class disease style loss function and a global loss function.
Preferably, the reconstruction loss function is:
Figure 81753DEST_PATH_IMAGE002
wherein,
Figure DEST_PATH_IMAGE003
is as follows
Figure 923195DEST_PATH_IMAGE004
Tagged second in individual status category
Figure DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the reconstruction loss function of the encoder,
Figure 640615DEST_PATH_IMAGE006
is corresponding to plant
Figure 504666DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 37147DEST_PATH_IMAGE005
A Gaussian model corresponding to the image of each plant sample,
Figure DEST_PATH_IMAGE007
is corresponding to plant
Figure 797293DEST_PATH_IMAGE004
The gaussian model corresponding to each state class,
Figure 329905DEST_PATH_IMAGE008
is corresponding to plant
Figure 997647DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 119055DEST_PATH_IMAGE005
An image of a sample of an individual plant,
Figure DEST_PATH_IMAGE009
corresponding to plants output from the decoder of the encoder for variation
Figure 846840DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 70011DEST_PATH_IMAGE005
A reconstructed image corresponding to the image of the plant sample,
Figure 275864DEST_PATH_IMAGE010
is composed of
Figure 517359DEST_PATH_IMAGE006
And
Figure 150465DEST_PATH_IMAGE007
is/are as follows
Figure DEST_PATH_IMAGE011
Divergence.
Preferably, the intra-category-disease style loss function is:
Figure DEST_PATH_IMAGE013
wherein,
Figure 595353DEST_PATH_IMAGE014
is as follows
Figure 788918DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 901231DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the intra-category disease style loss function of the encoder,
Figure DEST_PATH_IMAGE015
is corresponding to plant
Figure 970818DEST_PATH_IMAGE004
The number of labeled plant specimen images in each status category,
Figure 417849DEST_PATH_IMAGE016
is corresponding to plant
Figure 965505DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 666745DEST_PATH_IMAGE016
An image of a sample of an individual plant,
Figure 907233DEST_PATH_IMAGE005
is corresponding to plant
Figure 123451DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 724065DEST_PATH_IMAGE005
An image of a sample of an individual plant,
Figure DEST_PATH_IMAGE017
is corresponding to plant
Figure 748653DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 956781DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 863557DEST_PATH_IMAGE016
The degree of fusion of disease style characteristics of individual plant sample images,
Figure 736704DEST_PATH_IMAGE018
is as follows
Figure 678115DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 57144DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 185637DEST_PATH_IMAGE016
The degree of gaussian model fusion between individual plant sample images,
Figure DEST_PATH_IMAGE019
。
preferably, the inter-class disorder style loss function is:
Figure DEST_PATH_IMAGE021
wherein,
Figure 865404DEST_PATH_IMAGE022
is as follows
Figure 395743DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 148935DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the inter-class symptom style loss function of the encoder,
Figure DEST_PATH_IMAGE023
the number of state categories corresponding to the plants,
Figure 482833DEST_PATH_IMAGE024
is corresponding to plant
Figure 776411DEST_PATH_IMAGE024
The status of each of the plurality of status categories,
Figure 426835DEST_PATH_IMAGE004
is corresponding to plant
Figure 819771DEST_PATH_IMAGE004
The status of each of the plurality of status categories,
Figure DEST_PATH_IMAGE025
，
Figure 454014DEST_PATH_IMAGE026
is corresponding to plant
Figure 738234DEST_PATH_IMAGE004
Status class and
Figure 508744DEST_PATH_IMAGE024
the difficulty of gaussian model discrimination between individual state classes,
Figure DEST_PATH_IMAGE027
is corresponding to plant
Figure 807001DEST_PATH_IMAGE004
Status class and
Figure 912229DEST_PATH_IMAGE024
the difficulty of distinguishing the disease style characteristics between individual status categories.
Preferably, the global penalty function is:
Figure DEST_PATH_IMAGE029
wherein,
Figure 219714DEST_PATH_IMAGE022
is corresponding to plant
Figure 313572DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 91385DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the global loss function of the encoder,
Figure 903483DEST_PATH_IMAGE007
is corresponding to plant
Figure 545817DEST_PATH_IMAGE004
The gaussian model corresponding to each state class,
Figure 822078DEST_PATH_IMAGE030
is a standard Gaussian model and is used as a model,
Figure DEST_PATH_IMAGE031
is composed of
Figure 711405DEST_PATH_IMAGE007
And
Figure 276379DEST_PATH_IMAGE030
is/are as follows
Figure 253562DEST_PATH_IMAGE011
Divergence.
Preferably, the method for obtaining the label data corresponding to the plant sample image to be labeled comprises:
inputting each plant sample image to be labeled into a trained variational self-encoder to obtain a Gaussian model corresponding to each plant sample image to be labeled;
calculating the probability that each plant sample image to be labeled belongs to each state category corresponding to the plant according to the Gaussian model of each plant sample image to be labeled;
and taking the probability that each plant sample image to be labeled belongs to each state type corresponding to the plant as label data corresponding to the plant sample image to be labeled.
Drawings
To more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the following description will be made
While the drawings necessary for describing the embodiments or prior art are briefly described, it should be apparent that the drawings in the following description are merely examples of the invention and that other drawings may be derived from those drawings by those of ordinary skill in the art without inventive step.
FIG. 1 is a flow chart of the plant disease and insect pest diagnosis method based on artificial intelligence.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments, and all other embodiments obtained by those skilled in the art based on the embodiments of the present invention belong to the protection scope of the embodiments of the present invention.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The embodiment provides a plant disease and insect pest diagnosis method based on artificial intelligence, which is described in detail as follows:
as shown in figure 1, the plant disease and insect pest diagnosis method based on artificial intelligence comprises the following steps:
s001, acquiring a labeled plant sample image set, a plant sample image set to be labeled and a target plant image; the labeled plant sample image set comprises plant sample images of different state categories with labels; the different status categories include different pest categories and a pest-free category.
In this embodiment, the unmanned aerial vehicle carries the RGB high-definition camera to acquire images of plants in different growing periods in different areas, the plant images acquired by the unmanned aerial vehicle carrying the RGB high-definition camera in a historical time period are recorded as a plant sample image set, and the plant images acquired by the unmanned aerial vehicle carrying the RGB high-definition camera in real time are recorded as target plant images.
In the embodiment, when the unmanned aerial vehicle carries an RGB high-definition camera to acquire images, the unmanned aerial vehicle flies in low altitude; in this embodiment, the length of the historical time period and the frame rate of image acquisition are set according to actual conditions, but the acquired plant sample image set is large-scale and the plant sample image set includes all plant diseases and insect pests.
In this embodiment, a part of plant sample images in the collected plant sample image set is labeled manually, so that a labeled plant sample image set is obtained, and the remaining part of plant sample images in the collected plant sample image set is marked as a plant sample image set to be labeled; the labeled plant sample image set comprises plant sample images of different state categories with labels; the different status categories comprise different pest categories and pest-free categories; in this embodiment, the same label data is labeled on the plant sample images infected with the same pest and disease category, and the same label data is labeled on all the plant sample images which are normal, i.e., not infected with pest and disease, for example, the plant sample images infected with yellow leaf disease are all labeled as 1; the types of pests and diseases contained in the labeled plant sample image set are complete.
S002, obtaining a color texture distribution map corresponding to each plant sample image with the label according to each plant sample image with the label; and obtaining the disease style characteristics of each plant sample image with the label according to the color texture distribution map.
In the embodiment, the illumination information on each plant sample image with the label is removed by processing each plant sample image with the label, so that the influence of illumination on subsequent analysis is avoided; the specific method comprises the following steps: inputting each plant sample image with a label into a RetinexNet network to obtain a non-illuminated RGB image corresponding to each plant sample image with the label, wherein the RetinexNet network is a known prior art, and therefore, the embodiment is not described in detail.
In this embodiment, a 5 × 5 gaussian kernel is used to perform blurring processing on the non-illuminated RGB images corresponding to the labeled plant sample images, and texture information on the non-illuminated RGB images corresponding to the labeled plant sample images is removed to obtain non-illuminated non-textured RGB images corresponding to the labeled plant sample images; then, segmenting the RGB image without illumination and texture corresponding to each plant sample image with the label according to an SLIC superpixel segmentation algorithm to obtain a plurality of regions, wherein the color in each region is similar; in this embodiment, the pixel value of each pixel point in the same region is set as the average value of the summed pixel values in the region, so that the RGB image without illumination and texture corresponding to each labeled plant sample image containing color patches of different colors can be obtained; and marking the RGB image without illumination and texture corresponding to each labeled plant sample image containing different color patches as a color distribution diagram corresponding to each labeled plant sample image, wherein the color distribution diagram can indicate the color distribution condition of the plant after the plant is infected with the plant diseases and insect pests.
In this embodiment, the fuzzy processing and SLIC superpixel segmentation algorithms are well-known technologies, and therefore this embodiment is not described in detail; as another embodiment, other ways of blurring the non-illuminated RGB images corresponding to the labeled plant sample images or segmenting the non-illuminated non-textured RGB images corresponding to the labeled plant sample images by using other segmentation algorithms may be selected according to actual conditions.
In this embodiment, graying is performed on each labeled plant sample image to obtain a gray image of each labeled plant sample, then a Canny edge detection algorithm is used to perform edge extraction on the obtained gray image of each labeled plant sample to obtain an edge distribution map corresponding to each labeled plant sample gray image, and the edge distribution map can indicate texture distribution conditions after plant infection by plant diseases and insect pests; in this embodiment, the Canny edge detection algorithm is a known technique, and therefore this embodiment is not described in detail; as other embodiments, other algorithms can be used to perform edge extraction on the surface image of the injection molded part according to different requirements, such as a Sobel edge detection algorithm or a Roberts edge detection algorithm.
In this embodiment, the color distribution map corresponding to each plant sample image with a label and the edge distribution map corresponding to each plant sample gray-scale image with a label are fused to obtain the color texture distribution map corresponding to each plant sample image with a label; the specific fusion method comprises the following steps: copying three marks from the edge distribution diagram corresponding to the gray scale image of each plant sample with the single channel and carrying out weighting summation on the edge distribution diagram corresponding to the three channels of each plant sample image with the label and the color distribution diagram corresponding to each plant sample image with the label, and marking the image after weighting summation as the color texture distribution diagram corresponding to each plant sample image with the label.
In this embodiment, the weight of the three channels of the edge distribution map corresponding to each labeled plant sample image in the weighted summation process is 0.4, and the weight of the color distribution map corresponding to each labeled plant sample image is 0.6; as another embodiment, different weights may be assigned to the edge distribution map of the three channels corresponding to the labeled plant sample images and the color distribution map corresponding to the labeled plant sample images according to the actual situation, but the sum of the weights may be 1.
In this embodiment, the obtained color texture distribution map corresponding to each plant sample image with a label is analyzed to obtain a disease style characteristic corresponding to each plant sample image with a label, and the obtained disease style characteristic is used as a condition for constructing a loss function in the subsequent training of a variational self-encoder; the specific analysis method comprises the following steps: processing the obtained color texture distribution map corresponding to each plant sample image with the label by using a style migration algorithm to obtain a Gram matrix of the color texture distribution map corresponding to each plant sample image with the label, and taking the Gram matrix of the color texture distribution map corresponding to each plant sample image with the label as a disease style characteristic corresponding to each plant sample image with the label; the style migration algorithm is a well-known technique and thus the embodiment will not be described in detail. In this embodiment, the color texture distribution map obtained by the above process eliminates the mottle on each plant sample image with a label and enhances the disease style characteristics of each plant sample image with a label, as compared with each plant sample image with a label.
Step S003, training a variational self-encoder by utilizing the plant sample images with the labels to obtain Gaussian models corresponding to the plant sample images to be labeled; the loss function of the variation self-encoder is constructed by the disease style characteristics of each plant sample image with labels and a Gaussian model; the variation self-encoder is used for extracting disease style characteristics of each labeled plant sample image and distributing a label for the plant sample image to be labeled.
In the embodiment, the trained variational self-encoder is used as a basis for subsequently distributing labels for each plant sample image to be labeled; in this embodiment, a variational self-encoder is first constructed, and the structure of the variational self-encoder is a known technology, so this embodiment is not described in detail; the working principle of the variational self-encoder is as follows: inputting an image into a variational self-encoder, outputting a corresponding output from the encoder of the variational self-encoder for the inputted image
Figure 587592DEST_PATH_IMAGE032
The matrix is a matrix of a plurality of matrices,
Figure 929711DEST_PATH_IMAGE032
the matrix is represented as
Figure DEST_PATH_IMAGE033
High dimensional of dimensional space
Figure 700090DEST_PATH_IMAGE034
And is and
Figure 215385DEST_PATH_IMAGE032
first row representation of the matrix
Figure 669500DEST_PATH_IMAGE033
Mean of the dimensional gaussian model, the second row representing the elements on the diagonal of the gaussian covariance matrix; sampling one from the corresponding Gaussian model of the image
Figure 448100DEST_PATH_IMAGE033
One of the space
Figure 971354DEST_PATH_IMAGE033
Dimension vector of the
Figure 493603DEST_PATH_IMAGE033
The dimension vector input variation obtains a reconstructed image corresponding to the image from a decoder of an encoder.
In this embodiment, in the variational self-encoder
Figure 802224DEST_PATH_IMAGE033
128, as another embodiment, may be used according to the actual situation
Figure 751726DEST_PATH_IMAGE033
A different value is set, which may be 256, for example.
In this embodiment, each labeled plant sample image corresponding to each state category is input into a variation self-encoder, and the variation self-encoder is supervised and iteratively trained by using a conventional loss function, an intra-class disease style loss function, an inter-class disease style loss function, and a global loss function, for example, for the second plant corresponding to the plant
Figure 309746DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 887882DEST_PATH_IMAGE005
The conventional loss function for individual plant sample images is:
Figure DEST_PATH_IMAGE035
wherein,
Figure 519852DEST_PATH_IMAGE003
is as follows
Figure 374675DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 138101DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the reconstruction loss function of the encoder,
Figure 2152DEST_PATH_IMAGE006
is corresponding to plant
Figure 19786DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 373407DEST_PATH_IMAGE005
A Gaussian model corresponding to the image of each plant sample,
Figure 561812DEST_PATH_IMAGE007
is corresponding to plant
Figure 26291DEST_PATH_IMAGE004
The gaussian model corresponding to each state class,
Figure 898432DEST_PATH_IMAGE008
is corresponding to plant
Figure 360638DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 583809DEST_PATH_IMAGE005
An image of a sample of an individual plant,
Figure 38930DEST_PATH_IMAGE009
corresponding to plants output from the decoder of the encoder for variation
Figure 31156DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 726580DEST_PATH_IMAGE005
A reconstructed image corresponding to the image of the plant sample,
Figure 171468DEST_PATH_IMAGE010
is composed of
Figure 181012DEST_PATH_IMAGE006
And
Figure 539663DEST_PATH_IMAGE007
is/are as follows
Figure 343670DEST_PATH_IMAGE011
Divergence;
Figure 275854DEST_PATH_IMAGE036
is composed of
Figure DEST_PATH_IMAGE037
The L2 norm; this embodiment requires that the input of the variational autocoder be close to the output, i.e.
Figure 89090DEST_PATH_IMAGE008
Approach to
Figure 305176DEST_PATH_IMAGE009
Then, then
Figure 14506DEST_PATH_IMAGE038
The requirement approaches 0.
In the present embodiment, the first and second electrodes are,by corresponding of the plant
Figure 699565DEST_PATH_IMAGE004
Gaussian model corresponding to each state class
Figure 50912DEST_PATH_IMAGE007
The mean and covariance matrix of (a), analyzing the plant's corresponding second
Figure 668976DEST_PATH_IMAGE004
Gaussian model corresponding to each state class
Figure 329633DEST_PATH_IMAGE007
(ii) a Corresponding to plants
Figure 236409DEST_PATH_IMAGE004
Gaussian model corresponding to each state class
Figure 125868DEST_PATH_IMAGE007
The mean and covariance matrices of (a) are:
Figure 67279DEST_PATH_IMAGE040
Figure 633258DEST_PATH_IMAGE042
wherein,
Figure DEST_PATH_IMAGE043
is corresponding to plant
Figure 27330DEST_PATH_IMAGE004
Gaussian model corresponding to each state class
Figure 720480DEST_PATH_IMAGE007
The average value of (a) of (b),
Figure 250818DEST_PATH_IMAGE015
is corresponding to plant
Figure 256208DEST_PATH_IMAGE004
The number of labeled plant specimen images in each status category,
Figure 137576DEST_PATH_IMAGE044
is as follows
Figure 165575DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 815999DEST_PATH_IMAGE005
Gaussian model corresponding to plant sample image
Figure 474514DEST_PATH_IMAGE006
The mean value of (a);
Figure DEST_PATH_IMAGE045
is corresponding to plant
Figure 561287DEST_PATH_IMAGE004
The covariance matrix corresponding to each state class,
Figure 596240DEST_PATH_IMAGE046
is as follows
Figure 897908DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 727324DEST_PATH_IMAGE005
Gaussian model corresponding to plant sample image
Figure 832552DEST_PATH_IMAGE006
The covariance matrix of (2).
In this embodiment, the process can be used to obtain the second plant
Figure 202353DEST_PATH_IMAGE004
A state class pairCorresponding Gaussian model
Figure 827370DEST_PATH_IMAGE007
Is the mean value of
Figure 827687DEST_PATH_IMAGE004
Mean value of Gaussian model means corresponding to all labeled plant sample images in each state category, and mean value of Gaussian model means corresponding to plant
Figure 905364DEST_PATH_IMAGE004
The covariance matrix corresponding to each state class is
Figure 344436DEST_PATH_IMAGE004
Mean of covariance matrices corresponding to all labeled plant sample images in each state class, hence
Figure 73226DEST_PATH_IMAGE004
Gaussian model corresponding to each state class
Figure 510024DEST_PATH_IMAGE007
Is as follows
Figure 74997DEST_PATH_IMAGE004
And (3) the distribution characteristics of all labeled plant sample images in the state category in a high-dimensional space.
In this example, the plants correspond to
Figure 724284DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 593668DEST_PATH_IMAGE005
The intra-category-style loss function for individual plant sample images is:
Figure 201367DEST_PATH_IMAGE013
wherein,
Figure 50375DEST_PATH_IMAGE014
is as follows
Figure 768932DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 957468DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the intra-category disease style loss function of the encoder,
Figure 985335DEST_PATH_IMAGE015
is corresponding to plant
Figure 321639DEST_PATH_IMAGE004
The number of labeled plant specimen images in each status category,
Figure 578308DEST_PATH_IMAGE016
is corresponding to plant
Figure 886929DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 836431DEST_PATH_IMAGE016
An image of a sample of an individual plant,
Figure 394451DEST_PATH_IMAGE005
is corresponding to plant
Figure 969658DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 867207DEST_PATH_IMAGE005
An image of a sample of an individual plant,
Figure 987610DEST_PATH_IMAGE017
is corresponding to plant
Figure 501767DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 162556DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 429458DEST_PATH_IMAGE016
The degree of fusion of disease style characteristics of individual plant sample images,
Figure 720762DEST_PATH_IMAGE018
is as follows
Figure 456637DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 655537DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 793257DEST_PATH_IMAGE016
The degree of gaussian model fusion between individual plant sample images,
Figure 773239DEST_PATH_IMAGE019
。
in this embodiment, the following
Figure 465252DEST_PATH_IMAGE005
Image of plant sample and
Figure 733422DEST_PATH_IMAGE016
similarity of disease style characteristics between individual plant sample images is obtained
Figure 725649DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 93176DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 52911DEST_PATH_IMAGE016
The degree of fusion of disease pattern characteristics of individual plant sample images, and
Figure 859193DEST_PATH_IMAGE005
image of plant sample and
Figure 705926DEST_PATH_IMAGE016
similarity of disease style characteristics between individual plant sample images and the first
Figure 775513DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 442118DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 504621DEST_PATH_IMAGE016
The disease style characteristic fusion degrees of the plant sample images form a positive correlation; and calculating the corresponding second of the plants according to the following formula
Figure 471440DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 243087DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 662567DEST_PATH_IMAGE016
Disease style characteristic fusion degree of individual plant sample image
Figure 13914DEST_PATH_IMAGE017
：
Figure DEST_PATH_IMAGE047
Wherein,
Figure 350086DEST_PATH_IMAGE017
is corresponding to plant
Figure 230317DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 137093DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 88869DEST_PATH_IMAGE016
The degree of fusion of disease style characteristics between individual plant sample images,
Figure 764701DEST_PATH_IMAGE048
is as follows
Figure 593330DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 721823DEST_PATH_IMAGE005
The disease style characteristics corresponding to the images of the individual plant samples,
Figure DEST_PATH_IMAGE049
is as follows
Figure 680552DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 476469DEST_PATH_IMAGE016
Disease style characteristics corresponding to the plant sample images;
Figure 213350DEST_PATH_IMAGE050
is as follows
Figure 829139DEST_PATH_IMAGE005
Image of plant sample and
Figure 122717DEST_PATH_IMAGE016
the difference in the disease style characteristics between individual plant sample images,
Figure DEST_PATH_IMAGE051
is composed of
Figure 976404DEST_PATH_IMAGE050
The L2 norm of (a),
Figure 149765DEST_PATH_IMAGE052
is as follows
Figure 518429DEST_PATH_IMAGE005
Image of plant sample and
Figure 350119DEST_PATH_IMAGE016
similarity of disease style characteristics between individual plant sample images; when in use
Figure 855050DEST_PATH_IMAGE050
The smaller the value of (a) is,
Figure DEST_PATH_IMAGE053
the smaller the value of the L2 norm, the corresponding
Figure 402575DEST_PATH_IMAGE052
The greater the value of (A), i.e. the first
Figure 789694DEST_PATH_IMAGE005
Image of plant sample and
Figure 628337DEST_PATH_IMAGE016
the higher the similarity of disease style characteristics among the plant sample images;
Figure 987774DEST_PATH_IMAGE054
is at the first
Figure 253670DEST_PATH_IMAGE004
In the individual state class
Figure 52386DEST_PATH_IMAGE005
An image of a plant sample andthe sum of the similarity of the disease style characteristics of all the labeled plant sample images,
Figure DEST_PATH_IMAGE055
is a normalization coefficient.
In the present embodiment, the first and second electrodes are,
Figure 694720DEST_PATH_IMAGE017
the larger the value of (A) is, the
Figure 908664DEST_PATH_IMAGE004
The first in the individual state class
Figure 594729DEST_PATH_IMAGE005
Image of plant sample and
Figure 956440DEST_PATH_IMAGE016
the greater the degree of fusion of disease pattern characteristics of the individual plant sample images, and the requirements of this embodiment are on the second place
Figure 136886DEST_PATH_IMAGE004
In the individual state class
Figure 470915DEST_PATH_IMAGE005
Gaussian model of plant sample image and
Figure 78614DEST_PATH_IMAGE016
the Gaussian models of the plant sample images are relatively close to each other, namely the corresponding Gaussian model fusion degree is larger, so that for the same state class, the variational self-encoder can classify the two disease symptoms with large style feature similarity into one state class;
Figure 662042DEST_PATH_IMAGE017
smaller value of (A) indicates that
Figure 895446DEST_PATH_IMAGE004
The first in the individual state class
Figure 349561DEST_PATH_IMAGE005
Image of plant sample and
Figure 128162DEST_PATH_IMAGE016
the smaller the degree of fusion of disease pattern characteristics of the individual plant sample images, and the embodiment requires
Figure 402148DEST_PATH_IMAGE005
Gaussian model of plant sample image and
Figure 908085DEST_PATH_IMAGE016
the Gaussian models of the plant sample images are relatively far away from each other, namely the corresponding Gaussian models are less in fusion degree, so that for the same state class, the variational self-encoder can classify the two disease style characteristics with small similarity into the same state class.
In the present embodiment, the method is as follows
Figure 216706DEST_PATH_IMAGE004
In the individual state class
Figure 962945DEST_PATH_IMAGE005
Image of plant sample and
Figure 989807DEST_PATH_IMAGE016
the similarity of Gaussian models between the images of the individual plant samples is obtained
Figure 784588DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 459633DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 376773DEST_PATH_IMAGE016
The degree of Gaussian model fusion between individual plant sample images; according to the following formula
Figure 625352DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 223824DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 507038DEST_PATH_IMAGE016
The degree of gaussian model fusion between individual plant sample images:
Figure 47609DEST_PATH_IMAGE056
wherein,
Figure DEST_PATH_IMAGE057
is as follows
Figure 252326DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 920067DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 854525DEST_PATH_IMAGE016
The degree of gaussian model fusion between individual plant sample images,
Figure 565998DEST_PATH_IMAGE058
is as follows
Figure 523590DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 995023DEST_PATH_IMAGE016
The mean value of the gaussian model corresponding to each plant sample image,
Figure 783987DEST_PATH_IMAGE044
is as follows
Figure 417094DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 111249DEST_PATH_IMAGE005
The mean value of the gaussian model corresponding to each plant sample image,
Figure DEST_PATH_IMAGE059
is composed of
Figure 386373DEST_PATH_IMAGE060
The L2 norm of (a),
Figure DEST_PATH_IMAGE061
is composed of
Figure 701947DEST_PATH_IMAGE058
And
Figure 23732DEST_PATH_IMAGE058
the similarity of the gaussian models of (1),
Figure 955916DEST_PATH_IMAGE060
the smaller the norm of L2 of (a),
Figure 503572DEST_PATH_IMAGE058
and
Figure 204812DEST_PATH_IMAGE058
the greater the similarity of the gaussian models of (c),
Figure 242038DEST_PATH_IMAGE062
is at the first
Figure 910785DEST_PATH_IMAGE004
In the individual state class
Figure 996553DEST_PATH_IMAGE005
The sum of the similarity of the Gaussian models of the plant sample images and all the labeled plant sample images,
Figure DEST_PATH_IMAGE063
is a normalization coefficient.
In the present embodiment, the first and second electrodes are,
Figure 83458DEST_PATH_IMAGE057
greater values of (A) indicate in
Figure 494847DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 385312DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 540350DEST_PATH_IMAGE016
The more the Gaussian model fusion degree among the plant sample images is, the closer the Gaussian models are, and the distribution consistency of the Gaussian models is large; if it is
Figure 216182DEST_PATH_IMAGE057
Smaller value of (A) indicates that
Figure 798473DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 989283DEST_PATH_IMAGE005
The plant sample image and the labeled second
Figure 931700DEST_PATH_IMAGE016
The smaller the Gaussian model fusion degree among the plant sample images is, the closer the Gaussian models are, and the distribution consistency of the Gaussian models is small; required in this example
Figure 727617DEST_PATH_IMAGE064
Approaching 0, i.e.
Figure DEST_PATH_IMAGE065
Approaching 0, i.e.
Figure 418493DEST_PATH_IMAGE014
The value of (a) is close to 0, so that the distribution consistency of the Gaussian models of the labeled plant sample images with large disease style feature fusion degree is large, the distribution consistency of the Gaussian models of the labeled plant sample images with small disease style feature fusion degree is small, and the variational self-encoder can have stronger feature extraction and characterization capabilities.
In this example, the plants correspond to
Figure 96599DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 839777DEST_PATH_IMAGE005
The inter-class symptom style loss function for each plant sample image is:
Figure DEST_PATH_IMAGE067
wherein,
Figure 693464DEST_PATH_IMAGE022
is as follows
Figure 414295DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 517380DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the inter-class symptom style loss function of the encoder,
Figure 801600DEST_PATH_IMAGE023
the number of state categories corresponding to the plants,
Figure 306531DEST_PATH_IMAGE024
is corresponding to plant
Figure 401526DEST_PATH_IMAGE024
The status of each of the plurality of status categories,
Figure 788645DEST_PATH_IMAGE004
is corresponding to plant
Figure 627288DEST_PATH_IMAGE004
The status of each of the plurality of status categories,
Figure 235992DEST_PATH_IMAGE025
，
Figure 501889DEST_PATH_IMAGE026
is corresponding to plant
Figure 579566DEST_PATH_IMAGE004
Status class and
Figure 753058DEST_PATH_IMAGE024
the difficulty of gaussian model discrimination between individual state classes,
Figure 232581DEST_PATH_IMAGE027
is corresponding to plant
Figure 918646DEST_PATH_IMAGE004
Status class and
Figure 483620DEST_PATH_IMAGE024
the difficulty of distinguishing the disease style characteristics between individual status categories.
In this embodiment, the first step is calculated
Figure 398486DEST_PATH_IMAGE004
Status class and
Figure 732516DEST_PATH_IMAGE024
difficulty in differentiating disease style characteristics between individual status categories
Figure 136952DEST_PATH_IMAGE027
The specific method comprises the following steps: first, calculate the
Figure 441419DEST_PATH_IMAGE004
Image of each plant sample with label in individual state category and its second
Figure 425556DEST_PATH_IMAGE024
Similarity of disease style characteristics between the plant sample images with labels in the individual state category is obtained
Figure 348512DEST_PATH_IMAGE004
Status class and
Figure 923850DEST_PATH_IMAGE024
the similarity set of the disease style characteristics among the individual state categories takes the maximum value in the similarity set of the disease style characteristics as the first
Figure 463416DEST_PATH_IMAGE004
Status class and
Figure 234932DEST_PATH_IMAGE024
difficulty in differentiating disease style characteristics between individual status categories
Figure 277974DEST_PATH_IMAGE027
(ii) a When in use
Figure 227476DEST_PATH_IMAGE027
The larger the value of (A) indicates that the similarity of the disease style characteristics corresponding to the two labeled plant sample images corresponding to the maximum value in the disease style characteristic similarity set is higher, the higher the similarity of the disease style characteristics corresponding to the two labeled plant sample images is, the
Figure 254337DEST_PATH_IMAGE004
Status class and
Figure 111435DEST_PATH_IMAGE024
the greater the difficulty of distinguishing between the individual status categories, and when the disease style characteristics of the two labeled plant specimen images are too similar, this can result in a difficulty for the variational self-encoder to distinguish the two labeled plant specimen images into two pest types.
In this embodiment, in order to enable the variational self-encoder to distinguish the labeled plant sample images with too similar disease symptoms style characteristics between different state types, it is required that the gaussian models corresponding to the labeled plant sample images with too similar disease symptoms style characteristics between different state types have a relatively large differentiability. Namely, the difficulty in distinguishing the Gaussian models corresponding to the labeled plant sample images with too similar disease style characteristics among different state types is small, so that the variational self-encoder can extract different characteristics of the two labels, and the distinguishing capability of the variational self-encoder is improved.
In this embodiment, the following
Figure 523831DEST_PATH_IMAGE004
Status class and
Figure 909813DEST_PATH_IMAGE024
the difference degree of the Gaussian models among the individual state categories is obtained to obtain the corresponding second state of the plant
Figure 892812DEST_PATH_IMAGE004
Status class and
Figure 756863DEST_PATH_IMAGE024
difficulty of gaussian model discrimination between individual state classes; according to the following formula
Figure 836814DEST_PATH_IMAGE004
Status class and
Figure 377386DEST_PATH_IMAGE024
the difficulty of gaussian model discrimination between individual state classes:
Figure 847682DEST_PATH_IMAGE068
wherein,
Figure 515423DEST_PATH_IMAGE026
is as follows
Figure 653144DEST_PATH_IMAGE004
Status class and
Figure 912087DEST_PATH_IMAGE024
the difficulty of gaussian model discrimination between individual state classes,
Figure 405033DEST_PATH_IMAGE043
is as follows
Figure 610886DEST_PATH_IMAGE004
The mean of the gaussian models corresponding to each state class,
Figure DEST_PATH_IMAGE069
is as follows
Figure 603113DEST_PATH_IMAGE024
The mean of the gaussian models corresponding to each state class,
Figure 501799DEST_PATH_IMAGE070
is composed of
Figure DEST_PATH_IMAGE071
The L2 norm of (a),
Figure 664796DEST_PATH_IMAGE072
is as follows
Figure 408761DEST_PATH_IMAGE004
Status class and
Figure 521074DEST_PATH_IMAGE024
the degree of gaussian model difference between individual state classes,
Figure 121819DEST_PATH_IMAGE072
the larger the description is
Figure 568850DEST_PATH_IMAGE004
Status class and
Figure 850927DEST_PATH_IMAGE024
the less difficult the Gaussian model corresponding to each state class is to distinguish, and
Figure DEST_PATH_IMAGE073
is a normalized coefficient.
In this example, the requirements
Figure 286587DEST_PATH_IMAGE074
Approaching to 0, and making the Gaussian model corresponding to the labeled plant sample image with too similar disease style characteristics among different state types have smaller distinguishing difficulty.
In this example, the plants correspond to
Figure 323813DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 992561DEST_PATH_IMAGE005
The global loss function for each plant sample image is:
Figure DEST_PATH_IMAGE075
wherein,
Figure 547170DEST_PATH_IMAGE076
is as follows
Figure 165233DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 576623DEST_PATH_IMAGE005
The variation corresponding to each plant sample image is from the global loss function of the encoder,
Figure 735596DEST_PATH_IMAGE007
is corresponding to plant
Figure 625055DEST_PATH_IMAGE004
The gaussian model corresponding to each state class,
Figure 300887DEST_PATH_IMAGE030
is a standard Gaussian model and is used as a model,
Figure 883178DEST_PATH_IMAGE031
is composed of
Figure 73988DEST_PATH_IMAGE007
And
Figure 16405DEST_PATH_IMAGE030
is/are as follows
Figure 281164DEST_PATH_IMAGE011
Divergence.
In the present embodiment, the first and second electrodes are,
Figure 34357DEST_PATH_IMAGE030
is a standard Gaussian model with the mean value of 0 vector and all diagonal elements of the covariance matrix of 1,
Figure 915725DEST_PATH_IMAGE031
is composed of
Figure 209303DEST_PATH_IMAGE007
And
Figure 312257DEST_PATH_IMAGE030
is/are as follows
Figure 767509DEST_PATH_IMAGE011
Divergence, this example requires the corresponding second of the plant
Figure 136174DEST_PATH_IMAGE004
And the Gaussian models corresponding to the state classes are distributed on the standard Gaussian model.
In this embodiment, the following
Figure 171126DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 925324DEST_PATH_IMAGE005
The reconstruction loss function of the variational self-encoder corresponding to each plant sample image
Figure 754740DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 407438DEST_PATH_IMAGE005
Variation self-encoder intra-disease-like style loss function corresponding to individual plant sample image
Figure 246081DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 605518DEST_PATH_IMAGE005
Inter-class disease style loss function of variation self-encoder corresponding to plant sample image and second class disease style loss function corresponding to plant
Figure 117752DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 929851DEST_PATH_IMAGE005
Obtaining the global loss function of the variational self-encoder corresponding to the plant sample image to obtain the second loss function corresponding to the plant
Figure 572185DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 848445DEST_PATH_IMAGE005
Loss values corresponding to individual plant sample images; calculating the corresponding second of the plants according to the following formula
Figure 285243DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 99484DEST_PATH_IMAGE005
Loss values corresponding to individual plant sample images:
Figure 14350DEST_PATH_IMAGE078
wherein,
Figure DEST_PATH_IMAGE079
is corresponding to plant
Figure 817221DEST_PATH_IMAGE004
Tagged second in individual status category
Figure 221658DEST_PATH_IMAGE005
Loss values corresponding to individual plant sample images.
In this embodiment, each plant sample image with a label according to the above process may obtain a corresponding loss value, and then the parameters of the variational self-encoder are updated by using a random gradient descent algorithm until convergence, thereby completing the training process of the variational self-encoder.
And step S004, distributing labels to the plant sample images to be labeled according to the Gaussian model to obtain label data corresponding to the plant sample images to be labeled.
In this embodiment, a plant sample image set to be labeled is input into a trained variational self-encoder, a gaussian model corresponding to each plant sample image to be labeled in the plant sample image set to be labeled is obtained, then, according to the corresponding gaussian model similarity between each plant sample image to be labeled and each state category corresponding to the plant sample image, the probability that each plant sample image to be labeled belongs to each state category corresponding to the plant is obtained, and the probability that each plant sample image to be labeled belongs to each state category corresponding to the plant is calculated according to the following formula:
Figure DEST_PATH_IMAGE081
wherein,
Figure 726457DEST_PATH_IMAGE082
is as follows
Figure DEST_PATH_IMAGE083
The plant sample image to be labeled belongs to the second image corresponding to the plant
Figure 710594DEST_PATH_IMAGE004
The probability of the individual state class,
Figure 164709DEST_PATH_IMAGE084
is as follows
Figure 926997DEST_PATH_IMAGE083
A Gaussian model corresponding to the plant sample graph to be labeled,
Figure 466563DEST_PATH_IMAGE007
is as follows
Figure 988811DEST_PATH_IMAGE004
The gaussian model corresponding to each state class,
Figure DEST_PATH_IMAGE085
is composed of
Figure 297433DEST_PATH_IMAGE084
And
Figure 499132DEST_PATH_IMAGE007
is/are as follows
Figure 260414DEST_PATH_IMAGE011
The divergence of the light beam is measured by the light source,
Figure 383091DEST_PATH_IMAGE085
the smaller
Figure 546219DEST_PATH_IMAGE084
And
Figure 666622DEST_PATH_IMAGE007
the higher the corresponding gaussian model similarity,
Figure 164468DEST_PATH_IMAGE086
is composed of
Figure 28519DEST_PATH_IMAGE084
And
Figure 46153DEST_PATH_IMAGE007
corresponding gaussian model similarity; when in use
Figure 603037DEST_PATH_IMAGE086
The larger the value of (A), the corresponding second
Figure 135649DEST_PATH_IMAGE083
The plant sample image to be labeled belongs to the second image corresponding to the plant
Figure 52659DEST_PATH_IMAGE004
The greater the probability of a state class, where
Figure DEST_PATH_IMAGE087
Is a normalized coefficient.
In this embodiment, the probability that each plant sample image to be labeled belongs to each state category corresponding to the plant is used as the label data corresponding to the plant sample image to be labeled; therefore, through the process, the label can be distributed to each plant sample image to be labeled, and label data corresponding to each plant sample image to be labeled is obtained.
Step S005, training a target network according to the plant sample images with the labels, the label data corresponding to the plant sample images with the labels, the plant sample images to be labeled and the label data corresponding to the plant sample images to be labeled, wherein the target network is used for diagnosing whether the plants are infected with the plant diseases and insect pests and the corresponding specific plant disease types when the plants are infected with the plant diseases and insect pests.
In the embodiment, a target network is constructed, the target network is a fully-connected network, the labeled plant sample images, the label data corresponding to the labeled plant sample images, the plant sample images to be labeled and the label data corresponding to the plant sample images to be labeled are input into the target network for training, and the target network is trained by using a cross entropy loss function and a random gradient descent algorithm; the target network is used for diagnosing whether the plant is infected by the plant diseases and insect pests and the corresponding specific plant disease and insect pest types when the plant is infected by the plant diseases and insect pests.
And S006, inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image.
In this embodiment, the target plant image is input to the trained target network, the output result is the probability corresponding to each label, and the state class corresponding to the maximum probability value that the plant sample image to be labeled belongs to each state class corresponding to the plant is used as the state class of the plant sample image to be labeled.
According to the color texture distribution map, acquiring disease style characteristics of each plant sample image with labels; training a variational self-encoder by utilizing each plant sample image with a label to obtain a Gaussian model corresponding to each plant sample image to be labeled; according to the Gaussian model, distributing labels for the plant sample images to be labeled to obtain label data corresponding to the plant sample images to be labeled; training a target network according to the plant sample images with labels, the label data corresponding to the plant sample images with labels, the plant sample images to be labeled and the label data corresponding to the plant sample images to be labeled; and inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image. In this embodiment, each plant sample image with a label is used as a basis for training the variational self-encoder, the trained variational self-encoder is used as a basis for obtaining label data corresponding to each plant sample image to be labeled, and each plant sample image with a label, the label data corresponding to each plant sample image with a label, each plant sample image to be labeled and the label data corresponding to each plant sample image to be labeled are used as a basis for training the target network.
The plant disease and insect pest diagnosis system based on artificial intelligence comprises a memory and a processor, wherein the processor executes a computer program stored in the memory to realize the plant disease and insect pest diagnosis method based on artificial intelligence.
It should be noted that the order of the above-mentioned embodiments of the present invention is merely for description and does not represent the merits of the embodiments, and in some cases, actions or steps recited in the claims may be executed in an order different from the order of the embodiments and still achieve desirable results.

Claims (10)

1. A plant disease and insect pest diagnosis method based on artificial intelligence is characterized by comprising the following steps:
acquiring a plant sample image set with a label, a plant sample image set to be labeled and a target plant image; the labeled plant sample image set comprises plant sample images of different state categories with labels; the different status categories comprise different pest categories and pest-free categories;
obtaining a color texture distribution diagram corresponding to each plant sample image with the label according to each plant sample image with the label; obtaining disease style characteristics of each plant sample image with the label according to the color texture distribution map;
training a variational self-encoder by utilizing the plant sample images with the labels to obtain Gaussian models corresponding to the plant sample images to be labeled; the loss function of the variation self-encoder is constructed by the disease style characteristics of each plant sample image with labels and a Gaussian model; the variation self-encoder is used for extracting disease style characteristics of each plant sample image with a label and distributing the label for the plant sample image to be labeled;
according to the Gaussian model, distributing labels for the plant sample images to be labeled to obtain label data corresponding to the plant sample images to be labeled;
training a target network according to each plant sample image with a label, label data corresponding to each plant sample image with the label, each plant sample image to be labeled and the label data corresponding to each plant sample image to be labeled, wherein the target network is used for diagnosing whether the plant is infected with the plant diseases and insect pests and the corresponding specific plant disease and insect pest type when the plant is infected with the plant diseases and insect pests;
and inputting the target plant image into the trained target network to obtain a diagnosis result corresponding to the target plant image.
2. A method according to claim 1, wherein the method for obtaining a color texture distribution map corresponding to each labeled plant sample image from each labeled plant sample image comprises:
removing the illumination information and the texture information on each plant sample image with the label to obtain an RGB image without illumination and texture corresponding to each plant sample image with the label;
segmenting the RGB image without illumination and texture by utilizing a superpixel segmentation algorithm to obtain a color distribution map corresponding to each labeled plant sample image;
carrying out edge detection on the plant sample images with the labels by using an edge detection algorithm to obtain edge distribution maps corresponding to the plant sample images with the labels;
and obtaining a color texture distribution diagram corresponding to each labeled plant sample image according to the color distribution diagram and the edge gray level distribution diagram.
3. An artificial intelligence based plant pest diagnosis method according to claim 1, wherein the method for obtaining disease style characteristics of each labeled plant sample image according to the color texture distribution map comprises:
processing the color texture distribution map by using a style migration algorithm to obtain a Gram matrix corresponding to each color texture distribution map;
and taking the Gram matrix as a disease style characteristic corresponding to each plant sample image with the label.
4. An artificial intelligence based plant pest diagnosis method as claimed in claim 1, wherein said method of training a variational self-encoder using said labeled plant sample images comprises:
and respectively inputting the plant sample images with the labels into a variation self-encoder, and training the variation self-encoder by utilizing a reconstruction loss function, an intra-class disease style loss function, an inter-class disease style loss function and a global loss function.
5. An artificial intelligence based plant pest diagnosis method according to claim 4 wherein said reconfiguration loss function is:
Figure DEST_PATH_IMAGE002A
wherein,
Figure 369502DEST_PATH_IMAGE004
is as follows
Figure 831707DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 54878DEST_PATH_IMAGE008
The variation corresponding to each plant sample image is from the reconstruction loss function of the encoder,
Figure 244420DEST_PATH_IMAGE010
is corresponding to plant
Figure 236647DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 869753DEST_PATH_IMAGE008
A Gaussian model corresponding to the image of each plant sample,
Figure 563909DEST_PATH_IMAGE012
is corresponding to plant
Figure 307874DEST_PATH_IMAGE006
The gaussian model corresponding to each state class,
Figure 154607DEST_PATH_IMAGE014
is corresponding to plant
Figure 942303DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 140066DEST_PATH_IMAGE008
An image of a sample of an individual plant,
Figure 422143DEST_PATH_IMAGE016
corresponding to plants output from the decoder of the encoder for variation
Figure 372651DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 347560DEST_PATH_IMAGE008
A reconstructed image corresponding to the image of the plant sample,
Figure 767040DEST_PATH_IMAGE018
is composed of
Figure 102075DEST_PATH_IMAGE010
And
Figure 923401DEST_PATH_IMAGE012
is/are as follows
Figure 69211DEST_PATH_IMAGE020
Divergence.
6. An artificial intelligence based plant pest diagnosis method according to claim 4 wherein said intra-category disease style loss function is:
Figure DEST_PATH_IMAGE022A
wherein,
Figure 631779DEST_PATH_IMAGE024
is as follows
Figure 504926DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 180758DEST_PATH_IMAGE008
The variation corresponding to each plant sample image is from the intra-category disease style loss function of the encoder,
Figure 497470DEST_PATH_IMAGE026
is corresponding to plant
Figure 875231DEST_PATH_IMAGE006
The number of labeled plant specimen images in each status category,
Figure 302801DEST_PATH_IMAGE028
is corresponding to plant
Figure 82407DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 570020DEST_PATH_IMAGE028
An image of a sample of an individual plant,
Figure 451389DEST_PATH_IMAGE008
is corresponding to plant
Figure 666338DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 51183DEST_PATH_IMAGE008
An image of a sample of an individual plant,
Figure 709698DEST_PATH_IMAGE030
is corresponding to plant
Figure 62050DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 831423DEST_PATH_IMAGE008
The plant sample image and the labeled second
Figure 70775DEST_PATH_IMAGE028
The degree of fusion of disease style characteristics of individual plant sample images,
Figure 149458DEST_PATH_IMAGE032
is as follows
Figure 739839DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 296592DEST_PATH_IMAGE008
The plant sample image and the labeled second
Figure 921608DEST_PATH_IMAGE028
Gaussian model between plant sample imagesThe degree of fusion is such that,
Figure 921925DEST_PATH_IMAGE034
。
7. an artificial intelligence based plant pest diagnosis method according to claim 4 wherein said inter-class condition style loss function is:
Figure DEST_PATH_IMAGE036A
wherein,
Figure 655395DEST_PATH_IMAGE038
is as follows
Figure 281417DEST_PATH_IMAGE006
Tagged second in individual status category
Figure 495361DEST_PATH_IMAGE008
The variation corresponding to each plant sample image is from the inter-class symptom style loss function of the encoder,
Figure 932158DEST_PATH_IMAGE040
the number of state categories corresponding to the plants,
Figure 480820DEST_PATH_IMAGE042
is corresponding to plant
Figure 130107DEST_PATH_IMAGE042
The status of each of the plurality of status categories,
Figure 729716DEST_PATH_IMAGE006
is corresponding to plant
Figure 321103DEST_PATH_IMAGE006
The status of each of the plurality of status categories,
Figure 373373DEST_PATH_IMAGE044
，
Figure 560771DEST_PATH_IMAGE046
is corresponding to plant
Figure 264154DEST_PATH_IMAGE006
Status class and
Figure 777175DEST_PATH_IMAGE042
the difficulty of gaussian model discrimination between individual state classes,
Figure 316741DEST_PATH_IMAGE048
is corresponding to plant
Figure 822677DEST_PATH_IMAGE006
Status class and
Figure 865720DEST_PATH_IMAGE042
the difficulty of distinguishing the disease style characteristics between individual status categories.
8. An artificial intelligence based plant pest diagnosis method according to claim 4 wherein the global loss function is:
Figure DEST_PATH_IMAGE050A
wherein,
Figure 736593DEST_PATH_IMAGE052
is corresponding to plant
Figure 232296DEST_PATH_IMAGE054
Tagged second in individual status category
Figure 541924DEST_PATH_IMAGE056
The variation corresponding to each plant sample image is from the global loss function of the encoder,
Figure DEST_PATH_IMAGE058
is corresponding to plant
Figure 908314DEST_PATH_IMAGE054
The gaussian model corresponding to each state class,
Figure DEST_PATH_IMAGE060
is a standard Gaussian model and is used as a model,
Figure DEST_PATH_IMAGE062
is composed of
Figure 340301DEST_PATH_IMAGE058
And
Figure 854459DEST_PATH_IMAGE060
is/are as follows
Figure DEST_PATH_IMAGE064
Divergence.
9. A plant disease and pest diagnosis method based on artificial intelligence as claimed in claim 1, wherein the method for obtaining label data corresponding to the plant sample image to be labeled comprises:
inputting each plant sample image to be labeled into a trained variational self-encoder to obtain a Gaussian model corresponding to each plant sample image to be labeled;
calculating the probability that each plant sample image to be labeled belongs to each state category corresponding to the plant according to the Gaussian model of each plant sample image to be labeled;
and taking the probability that each plant sample image to be labeled belongs to each state type corresponding to the plant as label data corresponding to the plant sample image to be labeled.
10. An artificial intelligence based plant pest diagnosis system comprising a memory and a processor, wherein the processor executes a computer program stored in the memory to implement an artificial intelligence based plant pest diagnosis method according to any one of claims 1 to 9.
CN202111249226.1A 2021-10-26 2021-10-26 Plant disease and insect pest diagnosis method and system based on artificial intelligence Active CN113688959B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111249226.1A CN113688959B (en) 2021-10-26 2021-10-26 Plant disease and insect pest diagnosis method and system based on artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111249226.1A CN113688959B (en) 2021-10-26 2021-10-26 Plant disease and insect pest diagnosis method and system based on artificial intelligence

Publications (2)

Publication Number Publication Date
CN113688959A true CN113688959A (en) 2021-11-23
CN113688959B CN113688959B (en) 2022-02-18

Family

ID=78588163

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111249226.1A Active CN113688959B (en) 2021-10-26 2021-10-26 Plant disease and insect pest diagnosis method and system based on artificial intelligence

Country Status (1)

Country Link
CN (1) CN113688959B (en)

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102930249A (en) * 2012-10-23 2013-02-13 四川农业大学 Method for identifying and counting farmland pests based on colors and models
JP2016168046A (en) * 2015-03-09 2016-09-23 学校法人法政大学 Plant disease diagnostic system, plant disease diagnostic method, and program
CN106373136A (en) * 2015-07-22 2017-02-01 崔胡晋 Color and feature-based pest identifying and counting method
CN109308697A (en) * 2018-09-18 2019-02-05 安徽工业大学 A kind of leaf disease recognition method based on machine learning algorithm
CN110188824A (en) * 2019-05-31 2019-08-30 重庆大学 A kind of small sample plant disease recognition methods and system
CN111507419A (en) * 2020-04-22 2020-08-07 腾讯科技（深圳）有限公司 Training method and device of image classification model
CN112036335A (en) * 2020-09-03 2020-12-04 南京农业大学 Deconvolution-guided semi-supervised plant leaf disease identification and segmentation method
CN112560964A (en) * 2020-12-18 2021-03-26 深圳赛安特技术服务有限公司 Method and system for training Chinese herbal medicine pest and disease identification model based on semi-supervised learning
CN113468984A (en) * 2021-06-16 2021-10-01 哈尔滨理工大学 Crop pest and disease leaf identification system, identification method and pest and disease prevention method

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102930249A (en) * 2012-10-23 2013-02-13 四川农业大学 Method for identifying and counting farmland pests based on colors and models
JP2016168046A (en) * 2015-03-09 2016-09-23 学校法人法政大学 Plant disease diagnostic system, plant disease diagnostic method, and program
CN106373136A (en) * 2015-07-22 2017-02-01 崔胡晋 Color and feature-based pest identifying and counting method
CN109308697A (en) * 2018-09-18 2019-02-05 安徽工业大学 A kind of leaf disease recognition method based on machine learning algorithm
CN110188824A (en) * 2019-05-31 2019-08-30 重庆大学 A kind of small sample plant disease recognition methods and system
CN111507419A (en) * 2020-04-22 2020-08-07 腾讯科技（深圳）有限公司 Training method and device of image classification model
CN112036335A (en) * 2020-09-03 2020-12-04 南京农业大学 Deconvolution-guided semi-supervised plant leaf disease identification and segmentation method
CN112560964A (en) * 2020-12-18 2021-03-26 深圳赛安特技术服务有限公司 Method and system for training Chinese herbal medicine pest and disease identification model based on semi-supervised learning
CN113468984A (en) * 2021-06-16 2021-10-01 哈尔滨理工大学 Crop pest and disease leaf identification system, identification method and pest and disease prevention method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
G. MADHULATHA: "Recognition of Plant Diseases using Convolutional Neural Network", 《2020 FOURTH INTERNATIONAL CONFERENCE ON I-SMAC (IOT IN SOCIAL, MOBILE, ANALYTICS AND CLOUD) (I-SMAC)》 *
涂智潇: "基于计算机视觉的农作物病虫害检测系统", 《中国优秀硕士学位论文全文数据库 农业科技辑》 *
蔡高勇: "基于深度卷积生成对抗网络的小样本植物病害识别问题研究", 《中国优秀硕士学位论文全文数据库 农业科技辑》 *

Also Published As

Publication number Publication date
CN113688959B (en) 2022-02-18

Similar Documents

Publication Publication Date Title
Aich et al. Deepwheat: Estimating phenotypic traits from crop images with deep learning
CN111985536B (en) Based on weak supervised learning gastroscopic pathology image Classification method
Al Bashish et al. Detection and classification of leaf diseases using K-means-based segmentation and
CN114341948A (en) System and method for plant species identification
CN109325431B (en) Method and device for detecting vegetation coverage in feeding path of grassland grazing sheep
Ennadifi et al. Wheat diseases classification and localization using convolutional neural networks and GradCAM visualization
Al-Badri et al. Classification of weed using machine learning techniques: a review—challenges, current and future potential techniques
Liang et al. Low-cost weed identification system using drones
Kirola et al. Plants diseases prediction framework: A image-based system using deep learning
CN108629289A (en) The recognition methods in farmland and system, applied to the unmanned plane of agricultural
CN115205833A (en) Method and device for classifying growth states of cotton with few samples
Tosawadi et al. On the use of class activation map on rice blast disease identification and localization
CN113688959B (en) Plant disease and insect pest diagnosis method and system based on artificial intelligence
Balram et al. Crop field monitoring and disease detection of plants in smart agriculture using internet of things
Joshi et al. Detection and classification of plant diseases using soft computing techniques
CN115063602A (en) Crop pest and disease identification method based on improved YOLOX-S network
CN115170987A (en) Method for detecting diseases of grapes based on image segmentation and registration fusion
Cao et al. Plant leaf segmentation and phenotypic analysis based on fully convolutional neural network
Attada et al. A methodology for automatic detection and classification of pests using optimized SVM in greenhouse crops
Pajares et al. Combining classifiers through fuzzy cognitive maps in natural images
Bharti et al. Detection and classification of plant diseases
GSR et al. COTTON LEAF DISEASE DETECTION USING ARTIFICIAL NEURAL NETWORK.
Schurischuster Image analysis approaches for parasite detection on honeybees
CN118172676B (en) Farmland pest detection method based on quantum deep learning
Chaudhari et al. Identification of Banana Disease Using Color and Texture Feature

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