CN112801942A - Citrus huanglongbing image identification method based on attention mechanism - Google Patents

Citrus huanglongbing image identification method based on attention mechanism Download PDF

Info

Publication number
CN112801942A
CN112801942A CN202011632531.4A CN202011632531A CN112801942A CN 112801942 A CN112801942 A CN 112801942A CN 202011632531 A CN202011632531 A CN 202011632531A CN 112801942 A CN112801942 A CN 112801942A
Authority
CN
China
Prior art keywords
attention
feature
citrus
greening disease
loss function
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
CN202011632531.4A
Other languages
Chinese (zh)
Other versions
CN112801942B (en
Inventor
苏家仪
韦光亮
王筱东
陈露妃
姚姿娜
韦潇依
关宇晟
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.)
Guangxi Talentcloud Information Technology Co ltd
Original Assignee
Guangxi Talentcloud Information Technology 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 Guangxi Talentcloud Information Technology Co ltd filed Critical Guangxi Talentcloud Information Technology Co ltd
Priority to CN202011632531.4A priority Critical patent/CN112801942B/en
Publication of CN112801942A publication Critical patent/CN112801942A/en
Application granted granted Critical
Publication of CN112801942B publication Critical patent/CN112801942B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30128—Food products

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Multimedia (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the field of image recognition, and particularly discloses an attention mechanism-based citrus huanglongbing image recognition method, which specifically comprises the following steps: collecting diseased leaves and diseased fruit data of citrus greening disease as positive samples and data of other non-greening disease as negative samples, constructing an image classification network based on an attention mechanism, designing a loss function, inputting a training set into the image classification network based on the attention mechanism, carrying out supervision training by adopting the loss function, and inputting a verification set into a trained model for verification in the training process. And loading the trained model parameters to an image classification network based on an attention mechanism, and sequentially inputting the citrus huanglongbing images of the test set to the network for reasoning to obtain a citrus huanglongbing image classification result. By introducing the attention mechanism module, the key features of the citrus greening disease are selected independently, and the citrus greening disease detection method with low cost, high efficiency and high recognition rate is realized based on image recognition.

Description

Citrus huanglongbing image identification method based on attention mechanism
Technical Field
The invention relates to the field of image recognition, in particular to a citrus huanglongbing image recognition method based on an attention mechanism.
Background
The citrus yellow shoot is the most destructive disease in the current global citrus production, is caused by the phlobacterium of the citrus yellow shoot, takes citrus seedlings carrying pathogenic bacteria of the citrus yellow shoot as a transmission way, and takes citrus psyllid and the like as transmission media. The citrus growing areas of nearly 50 countries and regions in the world are infected, the important citrus producing areas such as Florida, Brazilian Saint Paul, Guangdong and Guangxi in China are seriously created, and a method for radically treating citrus yellow shoot is not found yet. Agricultural departments of various countries in the world pay great attention to the prevention and control of citrus greening disease, and 9 months and 15 days in 2020, the citrus greening disease is listed in a crop pest list by the agricultural rural department.
Early diagnosis and early treatment, the detection of citrus greening disease is crucial to prevention and control work, and at present, there are several detection methods: PCR detection, namely detecting whether the DNA sequence of a pathogen in the citrus is consistent with the DNA sequence of the yellow shoot disease by using instrument equipment, wherein a professional institution is required for identification, and the detection efficiency is low; the cost is high by utilizing the hyperspectral remote sensing of the unmanned aerial vehicle to carry out large-area detection; based on the characteristics of low cost and high efficiency of a computer vision scheme, the method is widely used for identifying plant diseases and insect pests in recent years, but the disease characteristics of the citrus huanglongbing belong to fine-grained characteristics, and the identification rate of a general image identification algorithm is low.
Disclosure of Invention
The invention provides an attention-based citrus huanglongbing image recognition method aiming at the defects in the background art, so that the defects that the disease characteristics of citrus huanglongbing belong to fine-grained characteristics and the recognition rate of a general image recognition algorithm is low are overcome.
In order to achieve the purpose, the invention provides an attention mechanism-based citrus greening disease identification method, which comprises the following steps of:
collecting citrus huanglongbing image data as a positive sample and collecting non-huanglongbing image data as a negative sample, and dividing the image data into a training set, a verification set and a test set according to disease categories;
establishing an image classification network, wherein the image classification network comprises an input layer, a feature extraction module, an attention mechanism module, a feature fusion module and an output layer; the input layer is used for receiving the positive samples and the negative samples; the characteristic extraction module is used for automatically extracting a characteristic diagram of the citrus greening disease through a neural learning network; the attention mechanism module is used for generating an attention diagram from the characteristic diagram; the feature fusion module is used for carrying out feature fusion on the feature map and the attention map to obtain a feature matrix; the output layer is used for converting the characteristic matrix into a classified prediction vector to be output;
designing a central loss function and a cross entropy loss function, adopting the central loss function to supervise the process of trying to learn the characteristics of the citrus greening disease, and adopting the cross entropy loss function to supervise the classification process of the classification prediction vector;
inputting the training set into the image classification network, performing supervision training by adopting the central loss function and the cross entropy loss function, and inputting the verification set into the trained image classification network for verification in the training process;
and loading the trained network parameters to the initial image classification network, and sequentially inputting the diseased leaves and the diseased fruit image data of the test set to the image classification network for testing to obtain the citrus huanglongbing image classification result.
Preferably, in the above technical solution, the input layer is configured to scale the citrus greening disease image to 448x448x3 and normalize the image to [ -1, 1 ].
Preferably, in the above technical solution, the feature extraction module adopts an inclusion v3 network structure, performs feature extraction on the citrus greening disease image data of the input layer, and outputs a feature map F with a size of 26x26x768, where 26x26 represents the width and height of the feature map F, and 768 is the number of channels of the feature map F.
Preferably, in the above technical solution, the attention mechanism module is composed of a 1x1 convolution layer, a Batch Normalization layer and a ReLU layer, a characteristic diagram F with a size of 26x26x768 is input, and an attention force diagram a with a size of 26x26x32 is output, where 26x26 represents the width and height of the attention force diagram a, 32 represents the number of channels a, denoted M, of the attention force diagram, and each channel a of the attention force diagram may correspond to a characteristic region of citrus greening disease.
Preferably, in the above technical solution, the inputs of the feature fusion module are a feature map F and an attention map a, as shown in formula (1), each channel a of the attention map a is respectively assignedkMultiplying the corresponding elements of the matrix with the characteristic diagrams F of all the channels to obtain the characteristic diagram F of the attention diagram channelkSize 26x26x768, and for feature map FkPerforming a global average pooling operation, as shown in formula (2), to obtain an attention feature map f with a size of 1x1x768kFinally, 32 attention feature maps fkSplicing is carried out, as shown in formula (3), a feature matrix P of 32x768 is obtained, and the feature matrix P is the final output result of the feature fusion module;
Fk＝Ak⊙F(k＝1，2，...，M) (1)
fk＝g(Fk) (2)
Figure BDA0002880395870000031
preferably, in the above technical solution, the input of the output layer is a feature matrix P, the feature matrix P is normalized by a norm of L2, then subjected to a scatter operation, converted into a 24576-dimensional feature vector, and input to a fully-connected layer with a channel number of 6, and finally a 6-dimensional classification prediction vector is obtained as an output.
Preferably, in the above technical solution, the step of supervising the classification of the classification prediction vector by using the cross entropy Loss function CE Loss specifically includes: firstly, inputting the 6-dimensional prediction vector into a CE Loss function for calculating the cross entropy Loss, wherein the CE Loss function is shown as a formula (4), and xclsPredictor, x, representing true class clsjA predictor representing a jth category;
the process of learning the citrus greening disease features by adopting the Center Loss function Center Loss supervision attention force diagram specifically comprises the following steps: as shown in equation (5), fkFor the kth attention map of 1x1x768 size, c denotes the feature center matrix, initialized to the all-zero matrix of 6x32x768 size,
Figure BDA0002880395870000032
for the feature center of the kth attention feature map in the real class cls, which has a feature center size of 1x1x768, for each attention channel k, f is calculatedkAnd
Figure BDA0002880395870000033
the Euclidean distances of the nodes are summed to obtain a Center Loss;
feature center pair by equation (6)
Figure BDA0002880395870000034
Update is performed, and β controls the update rate, set to 0.05.
CE Loss＝-xcls+log(∑jexp(xj))(j＝1，2，...，N) (4)
Figure BDA0002880395870000035
Figure BDA0002880395870000036
Compared with the prior art, the citrus greening disease image identification method based on the attention mechanism provided by the invention has the advantages that according to the visual characteristics of mottled yellowing of the leaves, coloring near the fruit pedicles and non-coloring at the fruit tops of the citrus greening disease, the key characteristics of the citrus greening disease are autonomously selected by introducing the attention mechanism module, the citrus greening disease detection scheme with low cost, high efficiency and high identification rate is realized based on the image identification, and the classification effect of the citrus greening disease model is improved.
Drawings
Fig. 1 is a flowchart of image classification according to an embodiment of the present invention.
Fig. 2 is a network structure diagram according to an embodiment of the present invention.
Detailed Description
The following detailed description of the present invention is provided in conjunction with the accompanying drawings, but it should be understood that the scope of the present invention is not limited to the specific embodiments.
The main design idea of the citrus huanglongbing image identification method based on the attention mechanism in the embodiment is as follows:
(1) data set preparation: collecting diseased leaves and diseased fruit data of citrus greening disease as positive samples, and other non-greening disease data as negative samples, classifying according to disease categories, and randomly dividing into a training set, a verification set and a test set according to the proportion of 0.8:0.1: 0.1.
(2) Constructing an image classification network based on an attention mechanism: the system consists of an input layer, a feature extraction module, an attention mechanism module, a feature fusion module and an output layer.
(3) Designing a loss function: and (3) adopting a cross entropy Loss CE Loss and Center Loss combined Loss function, wherein the CE Loss is responsible for supervising the classification process.
(4) Model training: and (3) inputting the training set into the image classification network based on the attention mechanism in the step (2), performing supervision training by adopting the loss function in the step (3), and inputting the verification set into the trained model for verification in the training process.
(5) Model reasoning: and (3) loading the model parameters trained in the step (4) to the image classification network based on the attention mechanism in the step (2), and sequentially inputting the citrus greening disease images tested and collected in the step (1) to the network for reasoning to obtain a citrus greening disease image classification result.
More specifically, referring to fig. 1, the citrus huanglongbing image recognition method based on the attention mechanism includes the following steps:
step S1, collecting diseased leaves and diseased fruit data of citrus greening disease as positive samples and other non-greening disease data as negative samples, classifying according to disease categories, and randomly dividing into a training set, a verification set and a test set according to the proportion of 0.8: 0.1;
step S2, constructing an attention-based image classification network, as shown in fig. 2: the system consists of an input layer, a feature extraction module, an attention mechanism module, a feature fusion module and an output layer:
an input layer: and scaling the citrus greening disease image in the step S1 to 448x448x3, normalizing to [ -1, 1], and accelerating the convergence speed of model training.
The feature extraction module is used for automatically extracting the visual feature map of the citrus greening disease through the neural network. More specifically, the feature extraction module adopts an inclusion v3 network structure to perform feature extraction on the image data of the input layer, and outputs a feature map with the size of 26x26x768, which is denoted as F, wherein 26x26 represents the width and height of the feature map F, and 768 is the channel number of the feature map F.
The attention mechanism module is used for generating an attention diagram, which is marked as A, from the feature diagram F output by the feature extraction module. The attention mechanism module consists of a 1x1 convolution layer, a Batch Normalization layer and a ReLU layer, a characteristic diagram F with the size of 26x26x768 is input, an attention diagram A with the size of 26x26x32 is output, wherein 26x26 represents the width and height of the attention diagram A, 32 represents the number of channels of the attention diagram A, and M is a parameter set manually. Each attention map a-channel may correspond to a characteristic region of citrus greening disease, such as a region colored near the citrus fruit base, and key features of citrus greening disease may be autonomously selected through model training.
The feature fusion module is used for performing feature fusion on the feature map F output by the feature extraction module and the attention map A output by the attention mechanism module. The input of the feature fusion module is a feature map F output by the feature extraction module and an attention map A output by the attention mechanism module, and as shown in formula (1), each channel A of the attention map A is respectively subjected to attentionkMultiplying the corresponding elements of the matrix with the characteristic diagrams F of all channels to obtain the characteristic diagram F of the channel A of the attention diagramkSize 26x26x768, and for FkPerforming a global average pooling operation, as shown in formula (2), to obtain an attention feature map f with a size of 1x1x768kFinally, the number of f is 32kAnd (4) splicing, as shown in formula (3), to obtain a feature matrix P of 32x768, which is the final output result of the feature fusion module.
Fk＝Ak⊙F(k＝1，2，...，M) (1)
fk＝g(Fk) (2)
Figure BDA0002880395870000051
The output layer module is used for mapping the feature matrix P of the feature fusion module to the classified prediction vector. Specifically, the input of the output layer is a feature matrix P output by the feature fusion module, L2 norm normalization is performed on P, then a scatter operation is performed, the P is converted into a 24576-dimensional feature vector, the 24576-dimensional feature vector is input to a full-connection layer with a channel number of 6 categories, and finally a 6-dimensional classification prediction vector is obtained as output.
Step S3, designing a loss function: and (3) adopting a combined Loss function of a cross entropy Loss function CE Loss and a central Loss function CenterLoss, wherein the CE Loss function is responsible for supervising the classification process. Inputting the 6-dimensional prediction vector of the output layer in the step S2 into a CE Loss function for calculating the cross entropy Loss, where the CE Loss is shown in formula (4), and xclsPredictor, x, representing true class clsjIndicating the predicted value for the jth category.
The CenterLoss function is responsible for supervising attention machineLearning process of formula (5), fkFor the kth attention feature map of 1x1x768 size in step S2, c denotes a feature center matrix, initialized to an all-zero matrix of 6x32x768 size,
Figure BDA0002880395870000061
for the feature center of the kth attention feature map in the real class cls, which has a feature center size of 1x1x768, for each attention channel k, f is calculatedkAnd
Figure BDA0002880395870000062
and summing the Euclidean distances to obtain a Center Loss function. The feature center can be paired by equation (6)
Figure BDA0002880395870000063
Update is performed, and β controls the update rate, set to 0.05.
CE Loss＝-xcls+log(∑jexp(xj))(j＝1，2，...，N) (4)
Figure BDA0002880395870000064
Figure BDA0002880395870000065
And step S4, inputting the training set into the image classification network based on the attention mechanism in the step S2, performing supervised training by adopting the loss function in the step S3, and inputting the verification set into the trained model for verification in the training process.
And S5, loading the model parameters trained in the step S4 to the image classification network based on the attention mechanism in the step S2, and sequentially inputting the citrus huanglongbing images tested and collected in the step S1 to the network for reasoning to obtain a citrus huanglongbing image classification result.
In summary, compared with the existing citrus greening disease image identification method, the citrus greening disease image identification method based on the attention mechanism in the embodiment has the advantages that the key features of citrus greening disease are automatically selected by introducing the attention mechanism module according to the visual features of mottled yellowing of leaves, coloring near fruit bases and non-coloring top portions of fruit, the citrus greening disease detection scheme with low cost, high efficiency and high identification rate is realized based on image identification, and the classification effect of the citrus greening disease model is improved.
The foregoing descriptions of specific exemplary embodiments of the present invention have been presented for purposes of illustration and description. It is not intended to limit the invention to the precise form disclosed, and obviously many modifications and variations are possible in light of the above teaching. The exemplary embodiments were chosen and described in order to explain certain principles of the invention and its practical application to enable one skilled in the art to make and use various exemplary embodiments of the invention and various alternatives and modifications as are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the claims and their equivalents.

Claims (7)

1. An attention mechanism-based citrus greening disease identification method is characterized by comprising the following steps:
collecting citrus huanglongbing image data as a positive sample and collecting non-huanglongbing image data as a negative sample, and dividing the image data into a training set, a verification set and a test set according to disease categories;
establishing an image classification network, wherein the image classification network comprises an input layer, a feature extraction module, an attention mechanism module, a feature fusion module and an output layer; the input layer is used for receiving the positive samples and the negative samples; the characteristic extraction module is used for automatically extracting a characteristic diagram of the citrus greening disease through a neural learning network; the attention mechanism module is used for generating an attention diagram from the characteristic diagram; the feature fusion module is used for carrying out feature fusion on the feature map and the attention map to obtain a feature matrix; the output layer is used for converting the characteristic matrix into a classified prediction vector to be output;
designing a central loss function and a cross entropy loss function, adopting the central loss function to supervise the process of trying to learn the characteristics of the citrus greening disease, and adopting the cross entropy loss function to supervise the classification process of the classification prediction vector;
inputting the training set into the image classification network, performing supervision training by adopting the central loss function and the cross entropy loss function, and inputting the verification set into the trained image classification network for verification in the training process;
and loading the trained network parameters to the initial image classification network, and sequentially inputting the diseased leaves and the diseased fruit image data of the test set to the image classification network for testing to obtain the citrus huanglongbing image classification result.
2. The attention-based mechanism citrus greening disease identification method according to claim 1, wherein said input layer is used to scale the citrus greening disease image to 448x448x3 and to normalize to [ -1, 1 ].
3. The method for identifying citrus greening disease based on attention mechanism according to claim 2, wherein the feature extraction module adopts an inclusion v3 network structure to perform feature extraction on the citrus greening disease image data of the input layer, and outputs a feature map F with the size of 26x26x768, wherein 26x26 represents the width and height of the feature map F, and 768 is the number of channels of the feature map F.
4. The method for identifying citrus greening disease based on attention mechanism according to claim 3, wherein the attention mechanism module is composed of a 1x1 convolution layer, a Batch Normalization layer and a ReLU layer, a characteristic diagram F with the size of 26x26x768 is input, an attention force diagram A with the size of 26x26x32 is output, wherein 26x26 represents the width and height of the attention force diagram A, 32 represents the number of channels A of the attention force diagram, and is denoted by M, and each channel A of the attention force diagram can correspond to a characteristic region of the citrus greening disease.
5. The attention-based citrus huanglongbing recognition method according to claim 4, wherein the input of the feature fusion module is a feature mapF and attention diagram A, as shown in formula (1), respectively directing attention to each channel A of the attention diagram AkMultiplying the corresponding elements of the matrix with the characteristic diagrams F of all the channels to obtain the characteristic diagram F of the attention diagram channelkSize 26x26x768, and for feature map FkPerforming a global average pooling operation, as shown in formula (2), to obtain an attention feature map f with a size of 1x1x768kFinally, 32 attention feature maps fkAnd (4) splicing, as shown in formula (3), to obtain a feature matrix P of 32x768, which is the final output result of the feature fusion module.
Fk＝Ak⊙F(k＝1，2，...，M) (1)
fk＝g(Fk) (2)
Figure FDA0002880395860000021
6. The method for identifying citrus greening disease based on attention mechanism according to claim 5, wherein the input of the output layer is a feature matrix P, L2 norm normalization is performed on the feature matrix P, then a flatten operation is performed, the feature matrix P is converted into 24576 dimensional feature vectors, the feature vectors are input into a full-connection layer with channel number of 6, and finally 6 dimensional classification prediction vectors are obtained as output.
7. The method for identifying citrus greening disease based on attention mechanism according to claim 6, wherein the step of supervising the classification of the classification prediction vectors by using a cross entropy Loss function CE Loss specifically comprises: firstly, inputting the 6-dimensional prediction vector into a CE Loss function for calculating the cross entropy Loss, wherein the CE Loss function is shown as a formula (4), and xclsPredictor, x, representing true class clsjA predictor representing a jth category;
the process of learning the citrus greening disease features by adopting the Center Loss function Center Loss supervision attention force diagram specifically comprises the following steps: as shown in equation (5), fkFor the kth attention map of 1x1x768 size, c denotes the feature center matrix, initialized to the all-zero matrix of 6x32x768 size,
Figure FDA0002880395860000031
for the feature center of the kth attention feature map in the real class cls, which has a feature center size of 1x1x768, for each attention channel k, f is calculatedkAnd
Figure FDA0002880395860000032
the Euclidean distances of the nodes are summed to obtain a Center Loss;
CE Loss＝-xcls+log(∑jexp(xj))(j＝1，2，...，N) (4)
Figure FDA0002880395860000033
Figure FDA0002880395860000034
feature center pair by equation (6)
Figure FDA0002880395860000035
Update is performed, and β controls the update rate, set to 0.05.
CN202011632531.4A 2020-12-31 2020-12-31 Citrus yellow dragon disease image identification method based on attention mechanism Active CN112801942B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011632531.4A CN112801942B (en) 2020-12-31 2020-12-31 Citrus yellow dragon disease image identification method based on attention mechanism

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011632531.4A CN112801942B (en) 2020-12-31 2020-12-31 Citrus yellow dragon disease image identification method based on attention mechanism

Publications (2)

Publication Number Publication Date
CN112801942A true CN112801942A (en) 2021-05-14
CN112801942B CN112801942B (en) 2023-10-13

Family

ID=75808454

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011632531.4A Active CN112801942B (en) 2020-12-31 2020-12-31 Citrus yellow dragon disease image identification method based on attention mechanism

Country Status (1)

Country Link
CN (1) CN112801942B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113627258A (en) * 2021-07-12 2021-11-09 河南理工大学 Apple leaf pathological detection method
CN114511732A (en) * 2021-12-31 2022-05-17 广西慧云信息技术有限公司 Citrus spotted disease and insect pest fine-grained image identification method
CN114511849A (en) * 2021-12-30 2022-05-17 广西慧云信息技术有限公司 Grape thinning identification method based on graph attention network
CN114677606A (en) * 2022-04-28 2022-06-28 电子科技大学 Citrus fine-grained disease identification method based on attention mechanism and double-branch network
CN115690920A (en) * 2023-01-03 2023-02-03 智慧眼科技股份有限公司 Credible living body detection method for medical identity authentication and related equipment
CN115731189A (en) * 2022-11-22 2023-03-03 河北师范大学 Enteroscopy image lesion identification method and device, electronic equipment and storage medium
CN114511732B (en) * 2021-12-31 2024-05-14 广西慧云信息技术有限公司 Orange spot disease and insect pest fine-granularity image identification method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110110751A (en) * 2019-03-31 2019-08-09 华南理工大学 A kind of Chinese herbal medicine recognition methods of the pyramid network based on attention mechanism
CN111382676A (en) * 2020-02-25 2020-07-07 南京大学 Sand image classification method based on attention mechanism
CN111489334A (en) * 2020-04-02 2020-08-04 暖屋信息科技(苏州)有限公司 Defect workpiece image identification method based on convolution attention neural network
CN111696101A (en) * 2020-06-18 2020-09-22 中国农业大学 Light-weight solanaceae disease identification method based on SE-Inception
AU2020102885A4 (en) * 2020-10-20 2020-12-17 Xijing University Disease recognition method of winter jujube based on deep convolutional neural network and disease image
CN112101424A (en) * 2020-08-24 2020-12-18 深圳大学 Generation method, identification device and equipment of retinopathy identification model

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110110751A (en) * 2019-03-31 2019-08-09 华南理工大学 A kind of Chinese herbal medicine recognition methods of the pyramid network based on attention mechanism
CN111382676A (en) * 2020-02-25 2020-07-07 南京大学 Sand image classification method based on attention mechanism
CN111489334A (en) * 2020-04-02 2020-08-04 暖屋信息科技(苏州)有限公司 Defect workpiece image identification method based on convolution attention neural network
CN111696101A (en) * 2020-06-18 2020-09-22 中国农业大学 Light-weight solanaceae disease identification method based on SE-Inception
CN112101424A (en) * 2020-08-24 2020-12-18 深圳大学 Generation method, identification device and equipment of retinopathy identification model
AU2020102885A4 (en) * 2020-10-20 2020-12-17 Xijing University Disease recognition method of winter jujube based on deep convolutional neural network and disease image

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
苏鸿 等: "基于区域卷积神经网络模型的广西柑橘病虫害识别方法研究", 《西南农业学报》, vol. 33, no. 04, pages 805 - 810 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113627258A (en) * 2021-07-12 2021-11-09 河南理工大学 Apple leaf pathological detection method
CN113627258B (en) * 2021-07-12 2023-09-26 河南理工大学 Apple leaf pathology detection method
CN114511849A (en) * 2021-12-30 2022-05-17 广西慧云信息技术有限公司 Grape thinning identification method based on graph attention network
CN114511849B (en) * 2021-12-30 2024-05-17 广西慧云信息技术有限公司 Grape thinning identification method based on graph attention network
CN114511732A (en) * 2021-12-31 2022-05-17 广西慧云信息技术有限公司 Citrus spotted disease and insect pest fine-grained image identification method
CN114511732B (en) * 2021-12-31 2024-05-14 广西慧云信息技术有限公司 Orange spot disease and insect pest fine-granularity image identification method
CN114677606A (en) * 2022-04-28 2022-06-28 电子科技大学 Citrus fine-grained disease identification method based on attention mechanism and double-branch network
CN114677606B (en) * 2022-04-28 2023-04-18 电子科技大学 Citrus fine-grained disease identification method based on attention mechanism and double-branch network
CN115731189A (en) * 2022-11-22 2023-03-03 河北师范大学 Enteroscopy image lesion identification method and device, electronic equipment and storage medium
CN115731189B (en) * 2022-11-22 2023-06-23 河北师范大学 Enteroscopy image lesion recognition method and device, electronic equipment and storage medium
CN115690920A (en) * 2023-01-03 2023-02-03 智慧眼科技股份有限公司 Credible living body detection method for medical identity authentication and related equipment

Also Published As

Publication number Publication date
CN112801942B (en) 2023-10-13

Similar Documents

Publication Publication Date Title
CN112801942A (en) Citrus huanglongbing image identification method based on attention mechanism
CN110516596B (en) Octave convolution-based spatial spectrum attention hyperspectral image classification method
CN110321862B (en) Pedestrian re-identification method based on compact ternary loss
CN106446942A (en) Crop disease identification method based on incremental learning
Mishra et al. A Deep Learning-Based Novel Approach for Weed Growth Estimation.
CN104217214A (en) Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method
CN107563389A (en) A kind of corps diseases recognition methods based on deep learning
CN106156765A (en) safety detection method based on computer vision
CN106997475A (en) A kind of insect image-recognizing method based on parallel-convolution neutral net
CN111415368B (en) Method for quickly measuring dropping speed of liquid drops of mobile equipment based on deep learning
Rahman et al. Performance comparison of different convolutional neural network architectures for plant seedling classification
CN116994295B (en) Wild animal category identification method based on gray sample self-adaptive selection gate
Seelwal et al. Automatic Detection of Rice Diseases using Deep Convolutional Neural Networks with SGD and ADAM
Miao et al. Crop weed identification system based on convolutional neural network
Menezes et al. Pseudo-label semi-supervised learning for soybean monitoring
Uoc et al. A novel automatic detecting system for cucumber disease based on the convolution neural network algorithm
CN112036470A (en) Cloud transmission-based multi-sensor fusion cucumber bemisia tabaci identification method
Brar et al. A smart approach to coconut leaf spot disease classification using computer vision and deep learning technique
CN116883364A (en) Apple leaf disease identification method based on CNN and Transformer
CN114566216B (en) Attention mechanism-based splice site prediction and interpretation method
CN108846327B (en) Intelligent system and method for distinguishing pigmented nevus and melanoma
CN115631462A (en) AM-YOLOX-based strawberry disease and pest detection method and system
Costales et al. Development of a Prototype Application for Rice Disease Detection Using Convolutional Neural Networks
Li et al. Identification of Crop Diseases Based on Improved Genetic Algorithm and Extreme Learning Machine.
CN114548197A (en) Clustering method based on self-discipline learning SDL model

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