CN110188635B - Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics - Google Patents
Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics Download PDFInfo
- Publication number
- CN110188635B CN110188635B CN201910404278.8A CN201910404278A CN110188635B CN 110188635 B CN110188635 B CN 110188635B CN 201910404278 A CN201910404278 A CN 201910404278A CN 110188635 B CN110188635 B CN 110188635B
- Authority
- CN
- China
- Prior art keywords
- plant
- insect
- plant disease
- network model
- insect pest
- 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
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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Catching Or Destruction (AREA)
Abstract
A plant disease and insect pest identification method based on an attention mechanism and multi-level convolution characteristics. The method aims to classify plant disease and insect pest images in a natural scene by combining an attention mechanism and multi-level convolution characteristics, and is characterized by solving the problem of interference of large-area complex backgrounds in natural scene images on plant disease and insect pest identification and fully utilizing all convolution layer characteristics in a network model. The method designs a deep convolution neural network comprising an attention mechanism, generates a local position area mask covering a target space position in an image by using the attention mechanism, is used for weakening the complex background interference information under a natural scene image, combines multi-level convolution characteristics, fully utilizes semantic information and detail information, and performs characteristic representation with more distinctiveness on plant diseases and insect pests. And finally, the obtained model is used for classifying plant diseases and insect pests in a natural scene, and the obtained local area mask is used for obtaining the position information of the plant diseases and insect pests.
Description
Technical Field
The invention belongs to the technical field of image processing, relates to a natural scene plant disease and insect pest image identification method, and more particularly relates to a plant disease and insect pest identification method based on an attention mechanism and multi-level convolution characteristics.
Background
Plant pests are one of the major causes of damage to commercial agricultural products. Plant pest identification plays a crucial role in agricultural plant pest prediction, grain safety and agricultural economy stabilization. Because plant diseases and insect pests are various and have slight differences, the identification of the plant diseases and insect pests depends on the professional knowledge of agricultural experts to a large extent, which means high cost, time consumption and labor consumption. With the development of machine learning and computer vision technology, automatic identification of plant diseases and insect pests is receiving more and more attention from researchers.
The traditional plant disease and insect pest identification work is mostly described by adopting a traditional machine learning classification framework, wherein the classification framework consists of two modules, namely 1) the characteristic representation of a plant disease and insect pest image is realized by adopting a series of manually-made characteristics such as a color histogram, scale space invariance and the like to represent the whole image. 2) The machine learning classifier includes a support vector machine and a k-nearest neighbor classifier. The paper "Tea insects classification based on identification neural networks" published by Samanta et al in International Journal of Computer Engineering Science (2(6):336) in 2012 diagnoses 8 Tea pests based on a data set of 609 samples using correlation-based feature selection and artificial neural networks. The paper "Early detection of pest on leaves using a support vector machine classifier" published by Manoja et al 2014 in International Journal of electric and Electronics Research (2(4):187:194) classified white flies, aphids and thrips in leaf images using a support vector machine classifier. However, the pest species are various in real life, and designing a feature extractor for recognizing various pests is not only inefficient but also time-consuming. Furthermore, the hand-made features lack the ability to represent high-level semantic information and, depending on the careful selection of features, if incomplete or erroneous features are extracted from the pest image, it will be difficult for subsequent classifiers to distinguish pest species in a complex background of natural scenes.
In recent years, deep learning has made feature learning more robust, achieving the most advanced performance in various image classification tasks. Several efforts have also been successful in applying neural networks to solve the pest identification problem. Paddy field pests are classified by training a deep convolutional neural network as in the paper "Localization and classification of behavior field pests using a saliency map and deep convolutional neural network" published by Liu et al 2016 in Scientific Reports (6: 20410). However, most of the existing pest images in the data set verified by the methods are collected in a controlled laboratory environment, most of plant pests and diseases in practical application appear in natural scenes and generally have complex background information, so that the method provides requirements on the capability of weakening the interference of the complex background information in the natural scene images of the plant pest and disease identification method and is a great difficulty in applying a deep learning technology to the plant pest and disease identification task.
Attention mechanisms have been applied to various computer vision problems, and early studies have utilized cyclic neural network architectures for attention modeling, which address classification tasks by sequentially selecting attention regions from images and then learning feature representations for each part, using target component detection. In addition to the recurrent neural network method, Bolei Zhou et al 2016 proposed CAM (class Activation mapping) in "Learning Deep Features for cognitive localization" published by CVPR (2921-2929), and created a weight distribution on the convolution characteristics by introducing a global mean pooling and a fully-connected neural network into the original convolutional neural network structure. Mnih et al 2014, in "Current models of visual attribution" published by Advances in neural information processing systems (2204-2212), proposed a gaze strategy for learning a chaotic number classification task based on a cyclic attention model that can focus high-resolution attention to the most discriminative areas without the need for bounding boxes or component positions.
Some latest achievements in the field stimulate the inspiration of us, and also provide a solid technical foundation for developing a plant disease and insect pest identification method based on an attention mechanism and multi-level convolution characteristics.
Disclosure of Invention
The invention aims to solve the technical problems that a user inputs a plant disease and insect pest image with any size, a network model can predict the category of the plant disease and insect pest and output the position mask information of the plant disease and insect pest in the image.
The technical scheme of the invention is as follows:
a plant disease and insect pest identification method based on an attention mechanism and multi-level convolution characteristics comprises the following steps:
a. inputting a plant disease and insect pest image with any size into a deep convolutional neural network by a user, calculating by using a network model, and extracting the convolutional characteristic of each layer;
the network model comprises two branches after the convolutional layer, wherein the first branch is an attention mechanism branch and comprises a full connection layer for learning weight distribution, and a local position area mask of plant diseases and insect pests is obtained by setting a weight threshold; the second branch is used for classifying plant diseases and insect pests;
b. adding an attention mechanism branch after the last layer of convolution feature map of the network model, generating weight distribution on the feature map through a full connection layer and calculating classification loss and gradient to represent the attention of the network model to the local area where the plant disease and insect target is located, obtaining a local position area mask of the plant disease and insect in the image through a preset weight threshold value, and using the local position area mask as a basis for shielding complex background information in the natural scene image;
c. b, performing average pooling operation on each layer of convolution features extracted by the network model in the step a to obtain convolution features with the same size, connecting the convolution features on a channel in a series connection mode, and fully utilizing the high-layer semantic information and the middle-bottom layer detail information of the network model to realize the combination of multi-layer convolution features;
d. the network model is multiplied by the mask of the local position area of the plant diseases and insect pests obtained in the step b element by element on a channel with multilayer convolution characteristics, so that the network notices the local area containing the plant diseases and insect pests, and the interference of a complex background in the image in a natural scene on plant disease and insect pest identification is weakened; then, local features are extracted, global average pooling operation is carried out to obtain feature representation of plant diseases and insect pests, prediction of plant disease and insect pest categories is carried out through plant disease and insect pest classification branches of a network model, semantic information and detail information are combined to carry out more sufficient feature representation on the plant diseases and insect pests, and meanwhile, the local area mask of the plant diseases and insect pests can reflect spatial position information of the plant diseases and insect pests in the image;
e. and (3) combining the prediction result of the attention branch of the network model to assist the plant disease and insect pest classification branch, respectively calculating the losses of the attention branch and the plant disease and insect pest classification branch of the model by taking the real plant disease and insect pest category as supervision information, then taking the average as the total loss of the model, and optimizing the whole network model by a small-batch random gradient descent method until convergence.
In practical application, the network model with the training convergence is used for predicting plant disease and insect pest images, and the model outputs a final classification recognition result.
The invention has the advantages and beneficial effects that: the method can be simply transferred to similar object recognition application, and only the class labels of the objects in the images need to be provided in the training process of the network model. The attention mechanism branch can realize the detection of the position of the object in the image under the condition of not needing the supervision information of the object coordinate, and effectively reduces the interference of a large amount of background information in the natural scene image in practical application. In addition, the method uses all the convolution layer characteristics in the network model, makes full use of the semantic information of the high layer and the detail information of the middle and bottom layers of the model, and can perform more distinctive characteristic representation on plant diseases and insect pests. In summary, the present invention provides a novel solution for rapidly and accurately identifying plant pests, and is believed to be well suited for many other computer vision related tasks.
Drawings
FIG. 1 is an architecture diagram of a plant pest identification method based on attention mechanism and multi-level convolution characteristics.
FIG. 2 is a flow chart of a plant pest identification method based on an attention mechanism and a multi-level convolution characteristic.
FIG. 3 is a graph of the effect of a plant pest identification method based on an attention mechanism and multi-level convolution characteristics.
Detailed Description
The invention designs a novel deep convolutional neural network architecture, namely, the deep convolutional neural network model comprises two branches after a convolutional layer, wherein the first branch is an attention mechanism branch, and the second branch is a branch for plant disease and pest classification. The mask of the local position area of the plant diseases and insect pests is generated through the attention mechanism branch, and the mask is combined with the multilayer convolution characteristics, so that the plant diseases and insect pests are identified under the complex background of the natural scene. The following describes in further detail embodiments of the present invention with reference to the accompanying drawings. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
Referring to the deep convolutional neural network architecture of the present invention shown in fig. 1, the following describes the implementation process in detail:
firstly, the neural network model designed by the invention needs a training sample set for model parameter learning, for example, a large-scale plant disease and insect pest data set IP102 is used, and the data set comprises 102 plant disease and insect pest categories and 75,222 images. In the model training process, a small batch of random gradient descent algorithm is used as an optimizer, the cross entropy function is a loss function, and the learning rate is set to be 0.01. And stopping training when the average loss of the model on the training set approaches 1E-2 level, and finally obtaining the deep convolutional neural network model for plant disease and insect pest classification.
The main architecture of the present invention can be any Deep neural network model, such as the common AlexNet, VGGNet, etc., as shown in fig. 1, the present invention uses the ResNet-50 model mentioned in the article "Deep residual learning for image recognition" published by He Kaiming et al 2016 in CVPR (770-: the attention mechanism branch module and the plant disease and insect pest classification branch module enable the plant disease and insect pest identification to be more efficient and accurate by sensing the physical space position of the plant disease and insect pest and applying the physical space position to convolution characteristics of all layers of the network model and extracting and fusing multi-layer characteristics.
The first branch module is an attention mechanism, the structure of which is essentially a classification process, a global average pooling layer is added after the last convolution layer with the size of 7x7 of a main body network architecture, the convolution feature map of the last convolution layer is pooled into a 2048-dimensional feature vector, the specific dimension can be changed according to the selection of the main body network (for example, the VGGNet network is 4096-dimensional), then the feature vector is mapped to the specific category of plant diseases and insect pests through a full connection layer, the parameter of the full connection layer is used as a weight to carry out weighted average on the convolution feature map of the last convolution layer from the main body network, the obtained weight distribution with the value range of [0,1] is passed through a threshold (0.4) to obtain a local position area mask of the plant diseases and insect pests, namely, the area larger than the threshold is the area mask contains the plant diseases and insect pests, otherwise, the mask can reflect the position information of the plant diseases and insect pests in the map, the size of the mask is 7x 7;
the second branch firstly fuses multilevel convolution characteristics, the ResNet-50 network comprises 4 residual blocks to gradually extract the convolution characteristics, wherein the convolution characteristics extracted by the later residual blocks contain richer detail information, and otherwise, the convolution characteristics contain more semantic information. However, the sizes of convolution features extracted from these 4 residual blocks are different, and are respectively: 56x56, 28x28, 14x14, 7x7, so that the 4 convolution features are respectively passed with step sizes of: 8. 4, 2 and 1, so as to obtain 4 convolution characteristics with the size of 7x7, then the 4 convolution characteristics are connected in series on a channel and multiplied element by a local position area mask to obtain local multi-level convolution characteristics, and a 2048-dimensional characteristic vector for describing plant disease and insect pest characteristics is obtained through a global average pooling layer, and the characteristic vector predicts the category of the plant disease and insect pest through a full connection layer.
Referring to fig. 2, a flow chart of the method of the present invention and a detailed effect chart of fig. 3:
when a user actually uses the method to identify the plant disease and insect pest image, firstly, as shown in the first line (a) of fig. 3, a plant disease and insect pest image containing the green moth wax cicada is input into a ResNet-50 network, and the image is gradually subjected to convolution feature extraction through 4 residual blocks of the ResNet-50 network.
The convolution features of size 7 × 7 extracted from the last convolution layer of the network are sent to the attention mechanism branch to generate a weight distribution on the feature map, as shown in fig. 3(b), the attention mechanism branch perceives a local region where a target exists on the original image (a), and the brighter region indicates that the possibility of plant diseases and insect pests is higher. By trying different weight thresholds in the [0,1] interval at intervals of 0.1, a mask of the plant pest local position area of fig. 3(c) is finally generated when the weight threshold is 0.4, and at this time, the mask is most suitable for the size of the coverage area of the plant pest, and the size of the mask is 7 × 7.
And simultaneously fusing multi-level convolution features extracted by 4 residual blocks in sequence, obtaining 4 convolution features with the size of 7x7 through average pooling layers with different step lengths, and connecting the 4 convolution features with the same size in series on a channel, thereby obtaining the multi-level convolution features simultaneously containing rich detail information and semantic information. The local area position mask generated by the attention mechanism and the multilayer convolution characteristic are multiplied element by element to generate the local multilayer convolution characteristic, and at the moment, an image area covered by the multilayer convolution characteristic only contains the local position of plant diseases and insect pests as shown in figure 3(d), so that complex background information is effectively shielded.
The local multilevel convolution characteristics are subjected to global average pooling to obtain 2048-dimensional characteristic vectors, and the characteristic vectors are subjected to a softmax classifier, as shown in fig. 3(e), so that the classifier finally identifies that the input plant disease and pest image is the cabbage moth and wax cicada.
The second and third lines in fig. 3 are also examples of plant pest identification based on attention mechanism and multi-level convolution features. The method comprises the steps of respectively representing different plant disease and insect pest image inputs, obtaining a local position area mask (c) of the insect pest through a preset threshold value by means of an attention mechanism on an original image (a) for sensing a local area (b) where a target exists, wherein the lighter the color is, the higher the possibility that the target exists in the area is, and shielding complex background information by the model through the local position area mask, so that only a local foreground area (d) is focused to generate a more accurate recognition result (e), and the effectiveness of the method is maintained when the method is applied to different examples.
Claims (3)
1. A plant disease and insect pest identification method based on an attention mechanism and multi-level convolution characteristics is characterized by comprising the following steps:
a. inputting a plant disease and insect pest image with any size into a deep convolutional neural network model by a user, calculating by the network model and extracting the convolutional characteristic of each layer;
b. adding an attention mechanism branch after the last layer of convolution feature map of the network model, generating weight distribution on the feature map through a full connection layer and calculating classification loss and gradient to represent the attention of the network model to the local area where the plant disease and insect target is located, obtaining a local position area mask of the plant disease and insect in the image through a preset weight threshold value, and using the local position area mask as a basis for shielding complex background information in the natural scene image;
c. b, performing average pooling operation on each layer of convolution features extracted by the network model in the step a to obtain convolution features with the same size, connecting the convolution features on a channel in a series connection mode, and fully utilizing the high-layer semantic information and the middle-bottom layer detail information of the network model to realize the combination of multi-layer convolution features;
d. the network model is multiplied by the mask of the local position area of the plant diseases and insect pests obtained in the step b element by element on a channel with multilayer convolution characteristics, so that the network notices the local area containing the plant diseases and insect pests, and the interference of a complex background in the image in a natural scene on plant disease and insect pest identification is weakened; then, local features are extracted, global average pooling operation is carried out to obtain feature representation of plant diseases and insect pests, prediction of plant disease and insect pest categories is carried out through plant disease and insect pest classification branches of a network model, semantic information and detail information are combined to carry out more sufficient feature representation on the plant diseases and insect pests, and meanwhile, the local area mask of the plant diseases and insect pests can reflect spatial position information in an image where the local area mask is located;
e. the prediction results of the attention branches of the network model are combined to assist plant disease and insect pest classification branches, the losses of the attention mechanism branches and the plant disease and insect pest classification branches of the model are respectively calculated by taking the real plant disease and insect pest categories as supervision information, then the average is taken as the total loss of the model, and the whole network model is optimized by a small-batch random gradient descent method until convergence;
in practical application, the network model with the training convergence is used for predicting plant disease and insect pest images, and the model outputs a final classification recognition result.
2. The plant pest identification method based on attention mechanism and multi-level convolution features according to claim 1, characterized in that: the network model in the step a comprises two branches after the convolutional layer, wherein the first branch is an attention mechanism branch and comprises a full connection layer for learning weight distribution, and a local position area mask of plant diseases and insect pests is obtained by setting a weight threshold; and the second branch is used for classifying plant diseases and insect pests, element-by-element multiplication operation is carried out on each channel by connecting multi-level convolution characteristics and local position area masks output in the step b, a characteristic vector is obtained through average pooling, and finally a new full-connection layer is used for mapping each plant disease and insect pest type.
3. The plant pest identification method based on attention mechanism and multi-level convolution features according to claim 1, characterized in that: in the step d, the network model is multiplied element by element with the local position area mask of the plant diseases and insect pests obtained in the step b on each channel of the multilayer convolution characteristics to obtain local multilayer convolution characteristics with background information shielded, the local characteristics are extracted and subjected to average pooling operation to obtain characteristic representation of the plant diseases and insect pests, the characteristics are mapped into prediction results of each type of plant diseases and insect pests through a full connection layer, and prediction probability is obtained through a softmax layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910404278.8A CN110188635B (en) | 2019-05-16 | 2019-05-16 | Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910404278.8A CN110188635B (en) | 2019-05-16 | 2019-05-16 | Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110188635A CN110188635A (en) | 2019-08-30 |
| CN110188635B true CN110188635B (en) | 2021-04-30 |
Family
ID=67716262
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910404278.8A Active CN110188635B (en) | 2019-05-16 | 2019-05-16 | Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110188635B (en) |
Families Citing this family (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110515130B (en) * | 2019-09-03 | 2021-08-06 | 河南工业大学 | Stored grain pest detection method and device based on channel state information |
| CN110648311B (en) * | 2019-09-03 | 2023-04-18 | 南开大学 | Acne image focus segmentation and counting network model based on multitask learning |
| CN111881705B (en) * | 2019-09-29 | 2023-12-12 | 深圳数字生命研究院 | Data processing, training and identifying method, device and storage medium |
| MX2022005392A (en) * | 2019-11-05 | 2022-06-08 | Apeel Tech Inc | Prediction of infection in plant products. |
| CN111027597B (en) * | 2019-11-22 | 2023-07-21 | 江苏农林职业技术学院 | Quick classification method and device for strawberry diseases and insect pests |
| CN110689093B (en) * | 2019-12-10 | 2020-04-21 | 北京同方软件有限公司 | Image target fine classification method under complex scene |
| CN111046793B (en) * | 2019-12-11 | 2023-05-02 | 北京工业大学 | Tomato disease identification method based on deep convolutional neural network |
| CN111104898B (en) * | 2019-12-18 | 2022-03-25 | 武汉大学 | Image scene classification method and device based on target semantics and attention mechanism |
| CN113096023B (en) * | 2020-01-08 | 2023-10-27 | 字节跳动有限公司 | Training method, image processing method and device for neural network and storage medium |
| CN111429447A (en) * | 2020-04-03 | 2020-07-17 | 深圳前海微众银行股份有限公司 | Focal region detection method, device, equipment and storage medium |
| CN112464971A (en) * | 2020-04-09 | 2021-03-09 | 丰疆智能软件科技(南京)有限公司 | Method for constructing pest detection model |
| CN111626969B (en) * | 2020-05-22 | 2023-05-30 | 张卫东 | Corn disease image processing method based on attention mechanism |
| CN111985552B (en) * | 2020-08-17 | 2022-07-29 | 中国民航大学 | Method for detecting diseases of thin strip-shaped structure of airport pavement under complex background |
| CN111914951A (en) * | 2020-08-21 | 2020-11-10 | 安徽省农业科学院农业经济与信息研究所 | Crop pest intelligent diagnosis system and method based on image real-time identification |
| CN112232343B (en) * | 2020-09-03 | 2023-11-21 | 国家粮食和物资储备局科学研究院 | Grain mildew grain identification neural network and identification method |
| CN112101265B (en) * | 2020-09-22 | 2023-04-25 | 四川大学 | Robust crop disease diagnosis system |
| CN112001365A (en) * | 2020-09-22 | 2020-11-27 | 四川大学 | High-precision crop disease and insect pest identification method |
| CN112149682B (en) * | 2020-09-29 | 2023-08-15 | 青海大学 | Method, storage medium and device for automatically identifying raspberry plants based on deep learning |
| CN112241762B (en) * | 2020-10-19 | 2022-07-01 | 吉林大学 | Fine-grained identification method for pest and disease damage image classification |
| CN112465038A (en) * | 2020-11-30 | 2021-03-09 | 深圳市识农智能科技有限公司 | Method and system for identifying disease and insect pest types of fruit trees |
| CN112668404B (en) * | 2020-12-10 | 2022-08-30 | 哈尔滨师范大学 | Effective identification method for soybean diseases and insect pests |
| CN112841154A (en) * | 2020-12-29 | 2021-05-28 | 长沙湘丰智能装备股份有限公司 | Disease and pest control system based on artificial intelligence |
| CN116738296B (en) * | 2023-08-14 | 2024-04-02 | 大有期货有限公司 | Comprehensive intelligent monitoring system for machine room conditions |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647684A (en) * | 2018-05-02 | 2018-10-12 | 深圳市唯特视科技有限公司 | A kind of Weakly supervised semantic segmentation method based on guiding attention inference network |
| CN108961350A (en) * | 2018-07-17 | 2018-12-07 | 北京工业大学 | One kind being based on the matched painting style moving method of significance |
| CN109271878A (en) * | 2018-08-24 | 2019-01-25 | 北京地平线机器人技术研发有限公司 | Image-recognizing method, pattern recognition device and electronic equipment |
| CN109446923A (en) * | 2018-10-10 | 2019-03-08 | 北京理工大学 | Depth based on training characteristics fusion supervises convolutional neural networks Activity recognition method |
| CN109447115A (en) * | 2018-09-25 | 2019-03-08 | 天津大学 | Zero sample classification method of fine granularity based on multilayer semanteme supervised attention model |
| CN109508663A (en) * | 2018-10-31 | 2019-03-22 | 上海交通大学 | A kind of pedestrian's recognition methods again based on multi-level supervision network |
| CN109753959A (en) * | 2018-12-21 | 2019-05-14 | 西北工业大学 | Road traffic sign detection method based on self-adaptive multi-scale feature fusion |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10679085B2 (en) * | 2017-10-31 | 2020-06-09 | University Of Florida Research Foundation, Incorporated | Apparatus and method for detecting scene text in an image |
-
2019
- 2019-05-16 CN CN201910404278.8A patent/CN110188635B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108647684A (en) * | 2018-05-02 | 2018-10-12 | 深圳市唯特视科技有限公司 | A kind of Weakly supervised semantic segmentation method based on guiding attention inference network |
| CN108961350A (en) * | 2018-07-17 | 2018-12-07 | 北京工业大学 | One kind being based on the matched painting style moving method of significance |
| CN109271878A (en) * | 2018-08-24 | 2019-01-25 | 北京地平线机器人技术研发有限公司 | Image-recognizing method, pattern recognition device and electronic equipment |
| CN109447115A (en) * | 2018-09-25 | 2019-03-08 | 天津大学 | Zero sample classification method of fine granularity based on multilayer semanteme supervised attention model |
| CN109446923A (en) * | 2018-10-10 | 2019-03-08 | 北京理工大学 | Depth based on training characteristics fusion supervises convolutional neural networks Activity recognition method |
| CN109508663A (en) * | 2018-10-31 | 2019-03-22 | 上海交通大学 | A kind of pedestrian's recognition methods again based on multi-level supervision network |
| CN109753959A (en) * | 2018-12-21 | 2019-05-14 | 西北工业大学 | Road traffic sign detection method based on self-adaptive multi-scale feature fusion |
Non-Patent Citations (5)
| Title |
|---|
| Fine-Grained Recognition of Vegetable Images Based on Multi-scale Convolution Neural Network;Xiu-Hong Yang et al.;《International Conference on Intelligent Computing》;20180706;第67-76页 * |
| HashGAN:Attention-aware Deep Adversarial Hashing for Cross Modal Retrieval;Xi Zhang et al.;《arXiv》;20171126;第1-10页 * |
| Recurrent Models of Visual Attention;Volodymyr Mnih et al.;《arXiv》;20140624;第1-12页 * |
| Residual Attention Network for Image Classfication;Fei Wang et al.;《2017 IEEE Conference on Computer Vision and Pattern Recognition》;20171109;第6450-6456页 * |
| 基于多注意力多尺度特征融合的图像描述生成算法;陈龙杰 等;《计算机应用》;20180928;第1-8页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110188635A (en) | 2019-08-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110188635B (en) | Plant disease and insect pest identification method based on attention mechanism and multi-level convolution characteristics | |
| WO2021043193A1 (en) | Neural network structure search method and image processing method and device | |
| Chiang et al. | Deep learning-based automated forest health diagnosis from aerial images | |
| US8379994B2 (en) | Digital image analysis utilizing multiple human labels | |
| CN112541508A (en) | Fruit segmentation and recognition method and system and fruit picking robot | |
| CN114821014B (en) | Multi-mode and countermeasure learning-based multi-task target detection and identification method and device | |
| CN109002755B (en) | Age estimation model construction method and estimation method based on face image | |
| CN110619059B (en) | Building marking method based on transfer learning | |
| Bueno et al. | Hierarchical object detection with deep reinforcement learning | |
| CN110633708A (en) | Deep network significance detection method based on global model and local optimization | |
| CN111291809A (en) | Processing device, method and storage medium | |
| CN114998220B (en) | Tongue image detection and positioning method based on improved Tiny-YOLO v4 natural environment | |
| CN111709468B (en) | Training method and device for directional artificial intelligence and storage medium | |
| Su et al. | LodgeNet: Improved rice lodging recognition using semantic segmentation of UAV high-resolution remote sensing images | |
| CN114387499A (en) | Island coastal wetland waterfowl identification method, distribution query system and medium | |
| Yasir et al. | Two-handed hand gesture recognition for Bangla sign language using LDA and ANN | |
| CN113159067A (en) | Fine-grained image identification method and device based on multi-grained local feature soft association aggregation | |
| Naseer et al. | Pixels to precision: features fusion and random forests over labelled-based segmentation | |
| CN116612386A (en) | Pepper disease and pest identification method and system based on hierarchical detection double-task model | |
| Pratiwi et al. | Early detection of deforestation through satellite land geospatial images based on CNN architecture | |
| Ouadiay et al. | Simultaneous object detection and localization using convolutional neural networks | |
| CN111352926B (en) | Method, device, equipment and readable storage medium for data processing | |
| CN113128308A (en) | Pedestrian detection method, device, equipment and medium in port scene | |
| Singhi et al. | Integrated YOLOv4 deep learning pretrained model for accurate estimation of wheat rust disease severity | |
| CN112668643A (en) | Semi-supervised significance detection method based on lattice tower rule |
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 |