CN112613393B - Silkworm disease identification system - Google Patents
Silkworm disease identification system Download PDFInfo
- Publication number
- CN112613393B CN112613393B CN202011509810.1A CN202011509810A CN112613393B CN 112613393 B CN112613393 B CN 112613393B CN 202011509810 A CN202011509810 A CN 202011509810A CN 112613393 B CN112613393 B CN 112613393B
- Authority
- CN
- China
- Prior art keywords
- silkworm
- pictures
- image
- disease
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/12—Details of acquisition arrangements; Constructional details thereof
- G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
- G06V10/141—Control of illumination
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention discloses a silkworm disease identification method, which comprises the following steps: a, acquiring a plurality of silkworm pictures, and classifying and labeling; b, constructing a TensorFlow frame, taking a plurality of classified silkworm pictures as a training set, and training the TensorFlow frame by using the image feature vectors of the silkworm pictures of the training set and the corresponding classification labels to obtain a silkworm disease identification model; and c, extracting the image characteristic vector of the silkworm picture to be identified, inputting the image characteristic vector into the silkworm disease identification model, and outputting to obtain the classification label corresponding to the silkworm picture to be identified. The intelligent silkworm disease prevention and control system has the beneficial effects of conveniently acquiring and rapidly mastering silkworm disease information and realizing the intellectualization of silkworm disease prevention and control. Also disclosed is a silkworm disease identification system comprising: a silkworm image acquisition module; a picture processing module; a silkworm image identification module; and a display module. The method has the beneficial effects of conveniently acquiring and rapidly mastering silkworm disease information and assisting in silkworm disease prevention and control.
Description
Technical Field
The invention relates to the technical field of silkworm disease picture identification. More particularly, the present invention relates to a silkworm disease identification system.
Background
Silkworm breeding has been receiving much attention from ancient times as part of agriculture. From the famous meaning from Luo applied Xi sang, Cai Yuan Ming Tang in Mi Shang sang, to the famous meaning from spring silkworm to dead silkworm in Tang Dynasty Li Shang, it can be seen that the sericulture has been in a position of long-term origin and great significance in our country. At present, the Guangxi Zhuang autonomous region in China is more reputable that the world silkworm industry sees China and the China silkworm industry sees Guangxi, and the silkworm cocoon and raw silk yield is at the top of the world. According to investigation, subacute infectious diseases such as silkworm nuclear polyhedrosis and fungal parasitic diseases such as silkworm white muscardine are harmful to Guangxi silkworm industry, and it is necessary to improve targeted prevention and control measures and strengthen accurate help and support for silkworm farmers.
With the development of science and technology, the technology and the agriculture are precisely combined, and the method becomes a necessary trend for the modernization of the agriculture development in China. In order to better serve the broad masses of silkworm farmers, help the silkworm farmers to conveniently acquire and quickly master the silkworm disease prevention and control knowledge, the traditional expert telephone inquiry or online knowledge inquiry is replaced, the diagnosis and treatment of silkworm diseases are not limited by time and places any more, and the realization of the intellectualization of silkworm disease prevention and control is worthy of research.
Disclosure of Invention
An object of the present invention is to solve at least the above problems and to provide at least the advantages described later.
The invention also aims to provide a silkworm disease identification method, wherein a large number of silkworm pictures and silkworm disease classification labels are established, intelligent training is carried out through a TensorFlow frame, a mapping relation between the silkworm pictures and the silkworm disease classification labels is obtained, only the sick silkworms need to be photographed at the later stage, image feature vector representation is carried out on the photographed pictures, and then the pictures are input into a silkworm disease identification model, so that whether the silkworms are sick or not, which disease is sick, symptoms, causes and hazards of the disease and a common treatment method can be quickly known. And a silkworm disease expert does not need to visit the silkworm farmer on the spot, so that a large amount of time can be saved.
The silkworm disease identification system can help silkworm farmers to conveniently acquire and quickly master silkworm disease prevention and control knowledge, replaces the traditional expert telephone inquiry or online knowledge inquiry, enables diagnosis and treatment of silkworm diseases not to be limited by time and place, and realizes intellectualization of silkworm disease prevention and control.
To achieve these objects and other advantages in accordance with the present invention, there is provided a silkworm disease identification method comprising the steps of:
a, acquiring a plurality of silkworm pictures, and carrying out classification and labeling on the silkworm pictures, wherein the classification and labeling comprise silkworm disease names of the silkworm pictures;
b, constructing a TensorFlow frame, taking a plurality of classified silkworm pictures as a training set, extracting image characteristic vectors of the silkworm pictures in the training set, and training the TensorFlow frame by using the image characteristic vectors of the silkworm pictures in the training set and corresponding classification labels to obtain a silkworm disease identification model;
and c, extracting the image characteristic vector of the silkworm picture to be identified, inputting the image characteristic vector into the silkworm disease identification model, and outputting to obtain the classification label corresponding to the silkworm picture to be identified.
Preferably, the method further comprises identification verification, which comprises the following steps:
d, randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles, and classifying and labeling the silkworm pictures;
e, constructing a graph variation self-encoder, selecting a plurality of silkworm pictures classified and labeled as the same class as a training set, extracting image characteristic vectors of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vectors of the silkworm pictures of the training set to obtain the graph variation self-encoder of the classification label;
f, obtaining a plurality of classified and labeled graph variation autocoders according to the classified and labeled types and the method of the step e;
step g, randomly rotating the silkworm pictures to be identified to obtain a plurality of silkworm pictures with different rotation angles, extracting image feature vectors of the silkworm pictures, respectively inputting the image feature vectors into the graph variation self-encoders corresponding to the classification mark types output in the step c, and outputting to obtain loss values of a plurality of image features;
and h, comparing with a preset loss value threshold, and outputting the classification label obtained in the step c if the loss value threshold is within the range of the loss value threshold.
Preferably, the method further comprises the step of preprocessing the silkworm pictures before extracting the feature vectors from the silkworm pictures, wherein the preprocessing comprises the steps of unifying backgrounds of the silkworm pictures, unifying pixels of the silkworm pictures and unifying sizes of the silkworm pictures.
Preferably, in the step c, the image feature vector of the silkworm picture to be identified and the image feature vector of the silkworm disease picture in the training set are compared through the Euclidean distance, similarity matching retrieval is carried out, the similarity values are arranged according to the sequence from high to low, classification labels of the silkworm disease pictures in the first N training sets are selected and arranged, and the classification labels are output, wherein N is an integer larger than 0.
Preferably, the loss value is calculated using the sum of the reconstruction loss and the KL divergence.
Preferably, the method further comprises associating the silkworm disease name with information on symptoms, causes, and treatment methods corresponding to the silkworm disease name.
Preferably, the method for unifying the background of the silkworm pictures specifically comprises the following steps: selecting a silkworm leaf picture as a standard picture, calculating the average value of the brightness values of all pixel points of the standard picture to obtain first illumination intensity information, and calculating the average value of the gray scale values of all pixel points of the standard picture to obtain second illumination intensity information;
determining ambient light compensation data according to the first illumination intensity information and the second illumination intensity information;
and carrying out exposure compensation processing on the silkworm disease picture according to the ambient light compensation data to obtain the silkworm picture with a uniform background.
There is provided a silkworm disease recognition system including:
the silkworm image acquisition module is used for acquiring a silkworm image;
the image processing module is used for preprocessing the silkworm image and extracting an image characteristic vector from the silkworm image;
the silkworm image identification module is used for constructing a TensorFlow frame, training the TensorFlow frame by taking a plurality of silkworm images which are classified and labeled as a training set and taking the image characteristic vector of the silkworm images of the training set and the corresponding classification label to obtain a silkworm disease identification model, and outputting the image characteristic vector of the silkworm image to be identified as an input value to obtain the classification label corresponding to the silkworm image to be identified;
and the display module is used for displaying the silkworm pictures and the classification labels.
Preferably, the silkworm disease verification device further comprises a silkworm disease verification module, which comprises:
the processing unit is used for randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles;
the extraction unit is used for selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set and extracting the image feature vector of the silkworm pictures;
the judging unit is used for constructing the graph variation self-encoder, selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set, extracting the image characteristic vector of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vector of the silkworm pictures of the training set to obtain the graph variation self-encoder which is classified and labeled;
and the verification unit is used for respectively inputting the image characteristic vectors of a plurality of silkworm pictures to be identified with different rotation angles to the graph variation self-encoder corresponding to the classification marking types output by the silkworm picture identification module, outputting loss values of a plurality of image characteristics, comparing the loss values with a preset loss value threshold value, and outputting the classification marking obtained by the silkworm picture identification module if the loss values are within the loss value threshold value range.
Preferably, the picture processing module is used for unifying the background of the silkworm pictures.
The invention at least comprises the following beneficial effects:
first, can help silkworm raisers to conveniently acquire and rapidly master silkworm disease prevention and control knowledge, replace the traditional expert telephone inquiry or online knowledge inquiry, make the diagnosis and treatment of silkworm disease no longer be limited by time and place, realize the intellectuality of silkworm disease prevention and control, fill in the blank of the industry.
Secondly, through establishing a large amount of classification labels of silkworm pictures and silkworm diseases, and carrying out intelligent training through a TensorFlow frame, obtaining the mapping relation of the classification labels of the silkworm pictures and the silkworm diseases, only taking pictures of the diseased silkworms at the later stage, carrying out image feature vector representation on the taken pictures, and then inputting the pictures into a silkworm disease identification model, whether the silkworms are diseased or not can be quickly known, which disease is diseased, and the symptoms, the etiology and the harm of the disease, and a common treatment method are obtained. And a silkworm disease expert does not need to visit the silkworm farmer on the spot, so that a large amount of time can be saved.
Thirdly, in order to further improve the accuracy of silkworm disease identification, especially for some silkworm disease types with high morbidity, high fatality rate or difficult rash, a graph variational self-encoder can be independently established, so that further identification verification is carried out before final output of classification labels, and the loss of silkworm farmers due to misdiagnosis is avoided.
Fourth, silkworms are usually cultivated indoors and are cultivated layer by layer at intervals from top to bottom, and the condition of insufficient light or uneven light often appears in the taken silkworm pictures. Therefore, a mode of selecting a standard picture is adopted, the silkworm picture is compensated according to the illumination intensity of the standard picture, the standard picture is preferably a silkworm leaf picture, the similarity of the silkworm leaf picture and a silkworm picture shot by a silkworm farmer is higher, the accuracy of ambient light compensation data is facilitated, and the silkworm disease identification accuracy is finally improved.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Detailed Description
The present invention is described in further detail so that those skilled in the art can practice the invention with reference to the description.
It is to be noted that the experimental methods described in the following embodiments are all conventional methods unless otherwise specified, and the reagents and materials are commercially available unless otherwise specified.
The invention provides a silkworm disease identification method, which comprises the following steps:
a, acquiring a plurality of silkworm pictures, and carrying out classification and labeling on the silkworm pictures, wherein the classification and labeling comprise silkworm disease names corresponding to the silkworm pictures; the more the number and the types of the silkworm pictures are, the more the accuracy of the identification method is facilitated. The classification label includes whether the silkworm is ill or not in the picture, the type of the disease, namely the name, the corresponding symptoms, the causes and the hazards of each silkworm disease, and the common treatment method. Through adopting artifical mode of marking, mark, although when the system is found in earlier stage, work load is big, nevertheless the system found the back of accomplishing, just can be fast, convenient, accurately discern whether sick the silkworm falls into illness in the silkworm picture, what kind of illness has.
B, constructing a TensorFlow frame, taking a plurality of classified silkworm pictures as a training set, extracting image characteristic vectors of the silkworm pictures in the training set, and training the TensorFlow frame by using the image characteristic vectors of the silkworm pictures in the training set and corresponding classification labels to obtain a silkworm disease identification model; representing and storing each silkworm picture in the training set through a 1024-dimensional feature vector, and establishing a mapping relation between the image feature vector and the classification label of each silkworm picture in the training set, namely training a TensorFlow frame to obtain a silkworm disease identification model;
and c, extracting the image characteristic vector of the silkworm picture to be identified, inputting the image characteristic vector into the silkworm disease identification model, and outputting to obtain the classification label corresponding to the silkworm picture to be identified. The silkworm disease image to be identified is expressed by 1024-dimensional characteristic vectors, the silkworm disease image to be identified is input into a silkworm disease identification model, the silkworm disease image characteristic vectors are compared with all stored training set silkworm disease image characteristic vectors through the Euclidean distance, one or more silkworm pictures with the highest similarity value are selected, and the corresponding classification labels are output for displaying.
In the technical scheme, a large number of silkworm pictures and silkworm diseases are classified and labeled, intelligent training is carried out through a TensorFlow frame, the mapping relation between the silkworm pictures and the silkworm diseases is obtained, only the diseased silkworm needs to be photographed in the later period, the photographed picture is subjected to image feature vector representation, and then the image feature vector representation is input into a silkworm disease recognition model, so that whether the silkworm is diseased or not, what kind of disease is diseased, and symptoms, etiology and harm of the disease and a common treatment method can be quickly known. And a silkworm disease expert does not need to visit the silkworm farmer on the spot, so that a large amount of time can be saved.
In another technical scheme, the method further comprises identification verification, and the method comprises the following steps:
d, randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles, and classifying and labeling the silkworm pictures; the same silkworm picture is rotated in multiple angles, and the accuracy of the identification model can be improved.
E, constructing a graph variation self-encoder, selecting a plurality of silkworm pictures classified and labeled as the same class as a training set, extracting image characteristic vectors of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vectors of the silkworm pictures of the training set to obtain the graph variation self-encoder of the classification label; the variational self-encoder is an unsupervised learning depth generation model. In the variational self-encoder, an inference network and a generation network based on a deep neural network are constructed. The inference network is used for the variation inference of original data to generate the variation probability distribution of hidden variables; and the generation network restores and generates the approximate probability distribution of the original data according to the generated hidden variable variation probability distribution. Therefore, intelligent clustering can be performed on each type of classified labels, and clustering accuracy is improved.
F, obtaining a plurality of classified and labeled graph variation autocoders according to the classified and labeled types and the method of the step e;
step g, randomly rotating the silkworm pictures to be identified to obtain a plurality of silkworm pictures with different rotation angles, extracting image characteristic vectors of the silkworm pictures, respectively inputting the image characteristic vectors into the graph variation self-encoders corresponding to the classification mark types output in the step c, and outputting to obtain loss values of a plurality of image characteristics;
and h, comparing with a preset loss value threshold, and outputting the classification label obtained in the step c if the loss value threshold is within the range of the loss value threshold. And (c) verifying the classification labels output in the step (c), if the classification labels are in the loss value threshold range, the classification labels are reliable, outputting and displaying, and if the classification labels are not in the loss value threshold range, the classification result is uncertain, for example, the silkworm diseases are difficult to diagnose.
In the technical scheme, in order to further improve the accuracy of silkworm disease identification, particularly for some silkworm disease types with high morbidity, high fatality rate or difficult rash, the graph variational self-encoder can be independently established, so that further identification verification is carried out before final output of classification labels, and the loss of silkworm farmers due to misdiagnosis is avoided.
In another technical scheme, the method further comprises the step of preprocessing the silkworm picture before extracting the characteristic vector of the silkworm picture, wherein the preprocessing comprises unifying the background of the silkworm picture, unifying the pixels of the silkworm picture and unifying the size of the silkworm picture. The influence of background, pixel inconsistency and noise can be reduced, and therefore the accuracy of feature vector representation is improved.
In another technical scheme, in the step c, comparing the image characteristic vector of the silkworm picture to be identified with the image characteristic vector of the silkworm disease picture in the training set through the Euclidean distance, performing similarity matching retrieval, arranging the similarity values in a high-to-low sequence, selecting and arranging the classification labels of the silkworm disease pictures in the first N training sets, and outputting the classification labels, wherein N is an integer larger than 0. When the classification labeling result of the integer larger than 1 is output, more references can be provided for the user, and the method is helpful for assisting in artificial judgment. For example, the silkworm raiser can select according to a plurality of output results and by combining the specific conditions in the self-breeding process.
In another technical scheme, the loss value is obtained by calculating the sum of the reconstruction loss and the KL divergence. In order to ensure the generalization capability of the graph variation self-encoder, the loss value of the variation self-encoder is calculated by adopting the sum of reconstruction loss and KL divergence (Kullback-Leibler divergence), and the variation self-encoder is trained and optimized on the basis of the loss value.
In another technical scheme, the method also comprises the step of associating the silkworm disease name with the information of symptoms, causes and treatment methods corresponding to the silkworm disease name. Thereby providing more silkworm disease information and helping silkworm farmers to process sick silkworms.
In another technical scheme, the method for unifying the background of silkworm pictures specifically comprises the following steps: selecting a silkworm leaf picture as a standard picture, calculating the average value of the brightness values of all pixel points of the standard picture to obtain first illumination intensity information, and calculating the average value of the gray scale values of all pixel points of the standard picture to obtain second illumination intensity information;
determining ambient light compensation data according to the first illumination intensity information and the second illumination intensity information;
and carrying out exposure compensation processing on the silkworm disease picture according to the ambient light compensation data to obtain the silkworm picture with a uniform background.
In the technical scheme, silkworms are usually cultivated indoors and are cultivated layer by layer at intervals from top to bottom, and the shot silkworm pictures often have insufficient light or uneven light. Therefore, a mode of selecting a standard picture is adopted, the silkworm picture is compensated according to the illumination intensity of the standard picture, the standard picture is preferably a silkworm leaf picture, the similarity of the silkworm leaf picture and a silkworm picture shot by a silkworm farmer is higher, the accuracy of ambient light compensation data is facilitated, and the silkworm disease identification accuracy is finally improved.
Provided is a silkworm disease recognition system including:
the silkworm image acquisition module is used for acquiring a silkworm image;
the image processing module is used for preprocessing the silkworm image and extracting an image characteristic vector from the silkworm image;
the silkworm image identification module is used for constructing a TensorFlow frame, training the TensorFlow frame by taking a plurality of silkworm images which are classified and labeled as a training set and taking the image characteristic vector of the silkworm images of the training set and the corresponding classification label to obtain a silkworm disease identification model, and outputting the image characteristic vector of the silkworm image to be identified as an input value to obtain the classification label corresponding to the silkworm image to be identified;
and the display module is used for displaying the silkworm pictures and the classification labels.
In the technical scheme, a large number of silkworm pictures and silkworm diseases are classified and labeled, intelligent training is carried out through a TensorFlow frame, the mapping relation between the silkworm pictures and the silkworm diseases is obtained, only the diseased silkworm needs to be photographed in the later period, the photographed picture is subjected to image feature vector representation, and then the image feature vector representation is input into a silkworm disease recognition model, so that whether the silkworm is diseased or not, what kind of disease is diseased, and symptoms, etiology and harm of the disease and a common treatment method can be quickly known. And a silkworm disease expert does not need to visit the silkworm farmer on the spot, so that a large amount of time can be saved.
In another technical scheme, the silkworm diseases verification system further comprises a silkworm diseases verification module, and the silkworm diseases verification module comprises:
the processing unit is used for randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles;
the extraction unit is used for selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set and extracting the image feature vector of the silkworm pictures;
the judging unit is used for constructing the graph variation self-encoder, selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set, extracting the image characteristic vector of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vector of the silkworm pictures of the training set to obtain the graph variation self-encoder which is classified and labeled;
and the verification unit is used for respectively inputting the image characteristic vectors of a plurality of silkworm pictures to be identified with different rotation angles to the graph variation self-encoder corresponding to the classification marking types output by the silkworm picture identification module, outputting loss values of a plurality of image characteristics, comparing the loss values with a preset loss value threshold value, and outputting the classification marking obtained by the silkworm picture identification module if the loss values are within the loss value threshold value range.
In the technical scheme, in order to further improve the accuracy of silkworm disease identification, particularly for some silkworm disease types with high morbidity, high fatality rate or difficult rash, the graph variational self-encoder can be independently established, so that further identification verification is carried out before final output of classification labels, and the loss of silkworm farmers due to misdiagnosis is avoided. The picture processing module is used for unifying the background of the silkworm pictures.
In the technical scheme, silkworms are usually cultivated indoors and are cultivated layer by layer at intervals from top to bottom, and the shot silkworm pictures often have insufficient light or uneven light. Therefore, a mode of selecting a standard picture is adopted, the silkworm picture is compensated according to the illumination intensity of the standard picture, the standard picture is preferably a silkworm leaf picture, the similarity of the silkworm leaf picture and a silkworm picture shot by a silkworm farmer is higher, the accuracy of ambient light compensation data is facilitated, and the silkworm disease identification accuracy is finally improved.
< example 1>
The general architecture of the silkworm disease identification system is divided into two parts: client side and server side.
The silkworm disease image acquisition system of the client acquires silkworm disease images, uploads the silkworm disease images to the server through a wireless network, and realizes remote communication with the server. Silkworm disease image acquisition at the client can be realized by taking pictures on site or selecting mobile phone album pictures, so that a good interactive interface is provided for a user; the server side expresses the silkworm disease image to be identified by using 1024-dimensional characteristic vectors, inputs the silkworm disease image to the silkworm disease identification model, compares the silkworm disease image characteristic vectors with all stored training set silkworm disease image characteristic vectors by using the Euclidean distance, selects one or more silkworm pictures with the highest similarity value, outputs the silkworm pictures with corresponding classification labels, and then pushes the silkworm disease image identification result to the client side. Through the hierarchical and modular design, the efficient cooperative work of the whole system is ensured.
The working process of the mobile phone client comprises the following steps: the corresponding App can be established at the mobile phone client, a user enters the App software system, the sick silkworm is photographed on site (or the sick silkworm picture in a mobile phone photo album is selected), the silkworm image photographed in real time (or selected in a library) is uploaded to the server through a wireless network, the photographed silkworm image photographed in real time is stored in the local photo album, and the silkworm identification image-text result pushed by the server is displayed on a mobile phone screen.
The working process of the server side comprises the following steps: after receiving a silkworm disease image request to be identified transmitted by the mobile phone client, the back-end server extracts the characteristics of the silkworm disease image to be identified by calling the silkworm disease identification model in the server, calculates the characteristic vector value of the silkworm disease image to be identified, compares the characteristic vector value with the characteristic vector value of each silkworm disease image in the library, obtains three silkworm disease images with the highest similarity through indexing, associates the corresponding silkworm disease characteristic character description, pushes a silkworm disease image-text result to the mobile phone client, and ends a user access request.
While embodiments of the invention have been disclosed above, it is not limited to the applications listed in the description and the embodiments, which are fully applicable in all kinds of fields of application of the invention, and further modifications may readily be effected by those skilled in the art, so that the invention is not limited to the specific details without departing from the general concept defined by the claims and the scope of equivalents.
Claims (8)
1. The silkworm disease identification method is characterized by comprising the following steps of:
a, acquiring a plurality of silkworm pictures, and carrying out classification and labeling on the silkworm pictures, wherein the classification and labeling comprise silkworm disease names of the silkworm pictures;
b, constructing a TensorFlow frame, taking a plurality of classified silkworm pictures as a training set, extracting image characteristic vectors of the silkworm pictures in the training set, and training the TensorFlow frame by using the image characteristic vectors of the silkworm pictures in the training set and corresponding classification labels to obtain a silkworm disease identification model;
c, extracting the image characteristic vector of the silkworm picture to be identified, inputting the image characteristic vector into a silkworm disease identification model, and outputting to obtain a classification label corresponding to the silkworm picture to be identified;
also included is identification verification comprising the steps of:
d, randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles, and classifying and labeling the silkworm pictures;
e, constructing a graph variation self-encoder, selecting a plurality of silkworm pictures classified and labeled as the same class as a training set, extracting image characteristic vectors of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vectors of the silkworm pictures of the training set to obtain the graph variation self-encoder of the classification label;
f, obtaining a plurality of classified and labeled graph variation autocoders according to the classified and labeled types and the method of the step e;
step g, randomly rotating the silkworm pictures to be identified to obtain a plurality of silkworm pictures with different rotation angles, extracting image characteristic vectors of the silkworm pictures, respectively inputting the image characteristic vectors into the graph variation self-encoders corresponding to the classification mark types output in the step c, and outputting to obtain loss values of a plurality of image characteristics;
and h, comparing with a preset loss value threshold, and outputting the classification label obtained in the step c if the loss value threshold is within the range of the loss value threshold.
2. The silkworm disease identification method according to claim 1, further comprising preprocessing the silkworm pictures before extracting the feature vectors from the silkworm pictures, wherein the preprocessing includes unifying backgrounds of the silkworm pictures, unifying pixels of the silkworm pictures, and unifying sizes of the silkworm pictures.
3. The silkworm disease recognition method according to claim 1, wherein in the step c, the image feature vector of the silkworm picture to be recognized and the image feature vector of the silkworm disease picture in the training set are compared by the euclidean distance, similarity matching search is performed, the similarity values are arranged in the order from high to low, and the classification labels of the silkworm disease pictures in the first N training sets are selected and output, wherein N is an integer greater than 0.
4. The silkworm disease identification method of claim 1, wherein the loss value is calculated by using the sum of the reconstruction loss and the KL dispersion.
5. The method for identifying silkworm diseases according to claim 1, further comprising associating the silkworm disease name with information on symptoms, causes, and treatment methods corresponding to the silkworm disease name.
6. The silkworm disease recognition method according to claim 2, wherein the method of unifying the background of silkworm pictures is specifically: selecting a silkworm leaf picture as a standard picture, calculating the average value of the brightness values of all pixel points of the standard picture to obtain first illumination intensity information, and calculating the average value of the gray scale values of all pixel points of the standard picture to obtain second illumination intensity information;
determining ambient light compensation data according to the first illumination intensity information and the second illumination intensity information;
and carrying out exposure compensation processing on the silkworm disease picture according to the ambient light compensation data to obtain the silkworm picture with a uniform background.
7. Silkworm disease identification system, characterized by, includes:
the silkworm image acquisition module is used for acquiring a silkworm image;
the image processing module is used for preprocessing the silkworm image and extracting an image characteristic vector from the silkworm image;
the silkworm image identification module is used for constructing a TensorFlow frame, training the TensorFlow frame by taking a plurality of silkworm images which are classified and labeled as a training set and taking the image characteristic vector of the silkworm images of the training set and the corresponding classification label to obtain a silkworm disease identification model, and outputting the image characteristic vector of the silkworm image to be identified as an input value to obtain the classification label corresponding to the silkworm image to be identified;
the display module is used for displaying silkworm pictures and classification labels;
still include silkworm disease verification module, it includes:
the processing unit is used for randomly rotating each silkworm picture to obtain a plurality of silkworm pictures with different rotation angles;
the extraction unit is used for selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set and extracting the image feature vector of the silkworm pictures;
the judging unit is used for constructing the graph variation self-encoder, selecting a plurality of silkworm pictures which are classified and labeled as the same class as a training set, extracting the image characteristic vector of the silkworm pictures of the training set, and training the graph variation self-encoder by using the image characteristic vector of the silkworm pictures of the training set to obtain the graph variation self-encoder which is classified and labeled;
and the verification unit is used for respectively inputting the image characteristic vectors of a plurality of silkworm pictures to be identified with different rotation angles to the graph variation self-encoder corresponding to the classification marking types output by the silkworm picture identification module, outputting loss values of a plurality of image characteristics, comparing the loss values with a preset loss value threshold value, and outputting the classification marking obtained by the silkworm picture identification module if the loss values are within the loss value threshold value range.
8. The system for identifying silkworm diseases of claim 7, wherein the picture processing module is used for unifying the background of the pictures of silkworms.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011509810.1A CN112613393B (en) | 2020-12-18 | 2020-12-18 | Silkworm disease identification system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011509810.1A CN112613393B (en) | 2020-12-18 | 2020-12-18 | Silkworm disease identification system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112613393A CN112613393A (en) | 2021-04-06 |
| CN112613393B true CN112613393B (en) | 2022-08-12 |
Family
ID=75240729
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011509810.1A Active CN112613393B (en) | 2020-12-18 | 2020-12-18 | Silkworm disease identification system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112613393B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409440A (en) * | 2018-11-09 | 2019-03-01 | 中国计量大学 | A kind of silk cocoon classification method based on color characteristic and support vector machines |
| CN110321830A (en) * | 2019-06-28 | 2019-10-11 | 北京邮电大学 | A kind of Chinese character string picture OCR recognition methods neural network based |
| CN111126325A (en) * | 2019-12-30 | 2020-05-08 | 哈尔滨工程大学 | Intelligent personnel security identification statistical method based on video |
| CN111199233A (en) * | 2019-12-30 | 2020-05-26 | 四川大学 | Improved deep learning pornographic image identification method |
| CN111950415A (en) * | 2020-07-31 | 2020-11-17 | 北京捷通华声科技股份有限公司 | Image detection method and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3029606A3 (en) * | 2014-11-14 | 2016-09-14 | Thomson Licensing | Method and apparatus for image classification with joint feature adaptation and classifier learning |
-
2020
- 2020-12-18 CN CN202011509810.1A patent/CN112613393B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409440A (en) * | 2018-11-09 | 2019-03-01 | 中国计量大学 | A kind of silk cocoon classification method based on color characteristic and support vector machines |
| CN110321830A (en) * | 2019-06-28 | 2019-10-11 | 北京邮电大学 | A kind of Chinese character string picture OCR recognition methods neural network based |
| CN111126325A (en) * | 2019-12-30 | 2020-05-08 | 哈尔滨工程大学 | Intelligent personnel security identification statistical method based on video |
| CN111199233A (en) * | 2019-12-30 | 2020-05-26 | 四川大学 | Improved deep learning pornographic image identification method |
| CN111950415A (en) * | 2020-07-31 | 2020-11-17 | 北京捷通华声科技股份有限公司 | Image detection method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112613393A (en) | 2021-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110148120B (en) | Intelligent disease identification method and system based on CNN and transfer learning | |
| Kan | Machine learning applications in cell image analysis | |
| CN110210542B (en) | Picture character recognition model training method and device and character recognition system | |
| CN109993102B (en) | Similar face retrieval method, device and storage medium | |
| JP2022551683A (en) | Methods and systems for non-invasive genetic testing using artificial intelligence (AI) models | |
| CN112581438B (en) | Slice image recognition method and device, storage medium and electronic equipment | |
| CN111178120A (en) | Pest image detection method based on crop identification cascade technology | |
| CN114359727A (en) | Tea disease identification method and system based on lightweight optimization Yolo v4 | |
| CN113378706B (en) | Drawing system for assisting children in observing plants and learning biological diversity | |
| CN109948429A (en) | Image analysis method, device, electronic equipment and computer-readable medium | |
| CN116012721B (en) | Deep learning-based rice leaf spot detection method | |
| CN112149690A (en) | Tracing method and tracing system based on biological image feature recognition | |
| CN111967527A (en) | Peony variety identification method and system based on artificial intelligence | |
| Zhang et al. | Deep learning based rapid diagnosis system for identifying tomato nutrition disorders | |
| CN113128522B (en) | Target identification method, device, computer equipment and storage medium | |
| CN112613393B (en) | Silkworm disease identification system | |
| CN112598013A (en) | Computer vision processing method based on neural network | |
| CN112446421A (en) | Silkworm cocoon counting and identifying method based on machine vision | |
| US20230297646A1 (en) | System and methods for classifying magnetic resonance imaging (mri) image characteristics | |
| CN114550210B (en) | Pedestrian re-identification method based on modal self-adaptive mixing and invariance convolution decomposition | |
| CN107563327B (en) | Pedestrian re-identification method and system based on self-walking feedback | |
| CN116071653A (en) | Automatic extraction method for multi-stage branch structure of tree based on natural image | |
| CN111401519B (en) | Deep neural network unsupervised learning method based on similarity distance in object and between objects | |
| CN113011506A (en) | Texture image classification method based on depth re-fractal spectrum network | |
| Rangappa et al. | Automatic counting and classification of silkworm eggs using deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |