CN113344009A - Light and small network self-adaptive tomato disease feature extraction method - Google Patents
Light and small network self-adaptive tomato disease feature extraction method Download PDFInfo
- Publication number
- CN113344009A CN113344009A CN202110621361.8A CN202110621361A CN113344009A CN 113344009 A CN113344009 A CN 113344009A CN 202110621361 A CN202110621361 A CN 202110621361A CN 113344009 A CN113344009 A CN 113344009A
- Authority
- CN
- China
- Prior art keywords
- tomato
- layer
- tomato leaf
- disease
- light
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- 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
-
- 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
Abstract
The invention discloses a light and small network self-adaptive tomato disease feature extraction method, which comprises the following steps: preprocessing a tomato leaf disease image; orienting all the preprocessed tomato leaf disease images to be uniform through a orthomorphism mechanism; constructing a light and small network model, and extracting the global characteristics and the interesting characteristics of the tomato leaf disease image; the light and small network model comprises a global feature extraction sub-network GFE-Net and an interested feature extraction sub-module FOIE-Block. The method can greatly extract the interesting features in the picture, realize high accuracy and high robustness of blade identification, and greatly reduce the parameter quantity of the network.
Description
Technical Field
The invention relates to the field of precision agriculture and computer vision application, in particular to a light and small network self-adaptive tomato disease feature extraction method.
Background
In recent years, due to the fact that various factors such as climate change, pollinator reduction, plant diseases and the like seriously threaten global grain safety, wherein the plant diseases are important factors which cause serious reduction of quality and quantity of agricultural products, when crops are diseased, physiological functions of the crops are greatly reduced, plants cannot reach an optimal production state due to the fact that the plants are thin, and therefore yield is not high, and economic benefits are low. In the tomato planting process, various diseases severely restrict the tomato production, and the diseases commonly comprise late blight, early blight, leaf mold, mosaic virus disease, spot blight and the like.
In the existing research work, the development process of the detection technology of plant diseases can be roughly divided into three stages. The first stage is manual identification, and the disease type is judged by depending on experience, so that the method is time-consuming and labor-consuming, strong in subjectivity and low in accuracy. The second stage is a traditional machine learning identification method, which extracts features by using a feature engineering and then performs classification identification by a classifier, but the method based on the machine learning still contains a large number of artificial influence factors, and the feature extraction engineering needs to be performed under a specific environment, so that the process is complex. The third stage is a deep learning identification mode, and realizes end-to-end system engineering by using the black box characteristics of the neural network without manually participating in the selection of the characteristics. Although many deep networks have good identification accuracy, the model has low portability due to the characteristics of complex structure, deep layer number, large parameter number and the like.
Disclosure of Invention
In order to meet the requirement of modern agriculture on accuracy, the invention provides a light and small network self-adaptive tomato disease feature extraction method, which can be used for greatly extracting interesting features in pictures, realizing high accuracy and high robustness of leaf identification and greatly reducing the parameter quantity of a network.
In order to achieve the purpose, the technical scheme of the invention is as follows: a light and small network self-adaptive tomato disease feature extraction method comprises the following steps:
preprocessing a tomato leaf disease image;
orienting all the preprocessed tomato leaf disease images to be uniform through a orthomorphism mechanism;
constructing a light and small network model, and extracting the global characteristics and the interesting characteristics of the tomato leaf disease image; the light and small network model comprises a global feature extraction sub-network GFE-Net and an interested feature extraction sub-module FOIE-Block.
Further, all the preprocessed tomato leaf disease images are oriented uniformly through a orthomorphism mechanism, and the method specifically comprises the following steps: unifying the directions of all tomato leaf disease images into that the leaf tips are upward and the leaf stalks are downward, so that the tomato leaves are positioned in the middle of the images, and ensuring that the deviation angle between the tomato leaves and the central line of the images is not more than +/-5 degrees.
Further, the working process of the orthomorphism mechanism is as follows:
acquiring a tomato leaf disease image;
graying the tomato leaf disease image and carrying out Gaussian filtering;
extracting edge characteristics of the tomato leaf disease image through a Canny operator, finding a minimum external rectangle, and obtaining a rotation angle theta;
and carrying out affine transformation on the original tomato leaf disease image by using the rotation angle theta.
Further, the global feature extraction sub-network GFE-Net is used for extracting global features of the tomato leaf disease image, and includes 4 Fire modules, 2 convolution layers, 3 maximum pooling layers, 1 global average pooling layer, 1 drop layer, and 1 softmax layer.
Further, each Fire module includes an Squeeze layer and an Expand layer, each layer consisting of only convolution kernels of 1 × 1 or 3 × 3 size.
Further, the interesting feature extraction submodule FOIE-Block is used for extracting interesting features of the tomato leaf disease image, and comprises 1 global average pooling layer, 2 full-connected layers, 2 activation functions and 1 matrix multiplication operation.
Furthermore, the specific operation of the interested feature extraction submodule FOIE-Block for extracting the interested features of the tomato leaf disease image is as follows:
extracting dimensions of a tomato leaf characteristic graph output by the sub-network GFE-Net convolution layer from the global characteristics, wherein the dimensions are [ W, H, C]Inputting the tomato leaf feature map into an interested feature extraction sub-module FOIE-Block, firstly, summing all values on the tomato leaf feature map with the size of H multiplied by W through a global averaging pooling layer (global averaging pooling), and taking an average value, wherein the output value of each channel is gCThe calculation formula is as follows:
wherein, gCFor the output value of each channel, uc(i, j) is the value in the feature map, the total output of the C channels is G, and G ═ G1,g2,…,gC]Dimension of [1,1, C ]]；
And carrying out operation processing on the total output G through a full connection layer and a nonlinear function ReLU, wherein the full connection layer formula is as follows:
wherein R has a dimension of
Weight parameters of full connectivity layer
The layer reduces the number of channels from C to r by a scaling factor
While using the activation function ReLU to obtain the non-linear relationship between channels, i.e. the characteristics for each channelLearning by weight;
and performing operation processing on the variable R through a full connection layer and a gate function Sigmoid, wherein the weight parameter of the full connection layer is
The full-connection layer operation of the layer restores the number of the channels to the original number, and the gate function sigmoid further captures the nonlinear relation among the channels, and then outputs an interesting characteristic Score FOI Score, the value of which is recorded as FOI _ S, and the operation formula is as follows:
wherein FOI _ S is a scalar comprising C numerical values, and [ FOI _ S1，FOI_S2，...，FOI_SC]The magnitude of (d) represents the magnitude of the score of the feature of interest of the corresponding channel.
Weighting the obtained interesting characteristic scores on the tomato leaf characteristic maps channel by channel, namely weighting each tomato leaf characteristic map [ W, H,1]，[W,H,2]，...，[W,H,C]Respectively with FOI _ S1，FOI_S2，...，FOI_SCMultiplication is carried out according to the following operation formula:
XC＝FOI_SC·uC
the result output by the layer is C H multiplied by W tomato leaf feature maps which are consistent with the input dimension, but the values on the tomato leaf feature maps are all recalibrated at the moment, namely the feature value of the interesting feature becomes larger, and irrelevant feature values become smaller and even suppressed, so that the extraction of the interesting feature is realized.
Due to the adoption of the technical scheme, the invention can obtain the following technical effects:
1. the network designed by the method has high accuracy rate which can reach 97.89%. 2. The network model is small, the size of the model can be 2.64MB, and the portability is stronger. 3. The identification speed is high, reaches 101 ms/sheet, and is easy to use in production. 4. The network has good stability and robustness, the identification accuracy of the disease picture containing Gaussian noise reaches 83.32%, and accurate identification of tomato leaf diseases and insect pests in a complex environment can be realized.
Drawings
FIG. 1 is a disease classification diagram of a new data set PV1 obtained by resampling;
FIG. 2 is a diagram of the operation of the orthomorphism mechanism;
FIG. 3 is a diagram of a Fire module architecture;
FIG. 4 is a diagram of a FOIE-Block structure of an interested feature extraction sub-module;
FIG. 5 is an overall structure diagram of a light small network;
FIG. 6 is a graph of accuracy versus loss for a network training set and a validation set;
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and the detailed description. The following examples are presented to enable one of ordinary skill in the art to more fully understand the present invention and are not intended to limit the scope of the embodiments described herein.
With the introduction of the precise agricultural concept, the accurate, fast and intelligent development becomes the latest target of agricultural development, so that in order to meet the precision requirement of modern agriculture, an efficient disease identification method is provided according to the existing research work progress and by combining the actual environmental background of agriculture, and the designed light and small neural network model can meet the requirements of high precision, high speed, high robustness and low parameter number at the same time.
On the basis of obtaining the global features, the light and small network enhances the expression of interesting Features (FOI) by learning the Feature weight, and simultaneously inhibits irrelevant features to realize the self-adaptive extraction of the features. The network comprises a Global Feature Extraction (GFE) sub-network GFE-Net and an interested Feature extraction (FOIE) sub-module FOIE-Block, wherein the GFE-Net extracts global features such as color, texture and shape of a picture through a series of convolution pooling operations to obtain the overall attribute of the picture; the FOIE-Block performs the Squeeze operation on the feature map containing the global features by utilizing a channel attention mechanism to obtain information descriptors of channel dimensions, generates weights for each feature channel by capturing the nonlinear interaction relation among the channel descriptors, and finally applies the weights to the input feature map, so that the expression of the interesting features is promoted and irrelevant features are suppressed. In conclusion, the light and small network designed by the invention can extract interesting features in pictures to a great extent through the self-adaptive extraction of the features, so that the high accuracy and the high robustness of blade identification are realized, and meanwhile, the parameter quantity of the network is greatly reduced.
Example 1
The embodiment provides a light and small network self-adaptive tomato disease feature extraction method, which comprises the following specific implementation steps:
s1, preprocessing a tomato leaf disease image
Specifically, a tomato leaf disease image is downloaded from a large public Plant disease data set theory altar Plant Village, and 11250 screened data sets are obtained in total of 8 categories. The sample is resampled by averaging 8 classes, randomly oversampling the classes below the average, and generating a new data set PV1 with a more balanced data set, as shown in fig. 1.
S2, all the preprocessed tomato leaf disease images are uniformly and specifically oriented through a positive mechanism, the positive mechanism is adopted to uniformly orient all the tomato leaf disease images to be that the leaf tips are upward and the leaf stalks are downward, so that the tomato leaves are located at the center of the images as far as possible, and the deviation angle between the tomato leaves and the central line of the images is not more than +/-5 degrees. Orientation of all leaf images is unified through a orthomorphism mechanism, so that the convolutional neural network is facilitated to extract more and more detailed disease characteristics, and learning efficiency of the network is greatly improved. The workflow of the orthomorphism mechanism is as shown in fig. 2: firstly, acquiring a tomato leaf disease image; graying the tomato leaf disease image; thirdly, performing Gaussian filtering on the tomato leaf disease image after the graying treatment; fourthly, extracting edge characteristics of the tomato leaf disease image through a Canny operator; finding the minimum external rectangle to obtain a rotation angle theta; and carrying out affine transformation on the original tomato leaf disease image by using the rotation angle theta.
S3, constructing a light and small network model, and extracting global features and interesting features of the tomato leaf disease image as shown in FIG. 5 to realize high accuracy and high robustness of leaf identification; the light and small network model comprises a global feature extraction sub-network GFE-Net and an interested feature extraction sub-module FOIE-Block.
Specifically, the global feature extraction sub-network GFE-Net is used as a basic framework for introducing channel attention and is mainly used for extracting global features of tomato leaf diseases. The basic network comprises 4 Fire modules, 2 convolutional layers, 3 maximum pooling layers, 1 global average pooling layer, 1 drop layer and 1 softmax layer. As shown in fig. 3, each Fire module includes a Squeeze layer and an Expand layer, each layer consisting of only convolution kernels of 1 × 1 or 3 × 3 size, in a way that reduces the amount of parameters of the network by a factor of 9. The 2 convolutional layers are respectively positioned in front of and behind the basic network, so that the basic characteristics and high-level semantic information of the picture can be effectively extracted. The 3 maximum pooling layers can extract feature maps with low resolution and strong semantic information. The global average pooling layer is used for replacing a full connection layer, so that the calculation complexity and parameter quantity of the network can be effectively reduced, and the network training speed is increased. The introduction of the Dropout layer can effectively prevent overfitting. And calculating the probability of each category by using a Softmax function, and outputting a prediction result.
The interesting feature extraction submodule FOIE-Block realizes the extraction of interesting Features (FOI) by means of the working mechanism of channel attention. Channel attention may explicitly model interdependencies between channels, re-adaptively calibrating the characteristic response of the channels. The importance degree of each feature channel is automatically acquired through learning, and then useful features are selectively improved and the features with weak expression ability are inhibited according to the importance degree, so that the expression ability of the whole network is enhanced.
As shown in fig. 4, the interest feature extraction sub-module FOIE-Block includes 1 global average pooling layer, 2 full-connected layers, 2 activation functions and 1 matrix multiplication operation, and the process of extracting the interest feature by the module is as follows:
the first step is as follows: a channel descriptor is obtained. The dimension of the tomato leaf characteristic graph output from the sub-network GFE-Net convolution layer is [ W, H, C ]]Inputting the data into FOIE-Block, firstly passing through global averaging pooling layer (global averaging pooling), summing all values on the tomato leaf characteristic diagram with size of H × W, and averaging, wherein the output value of each channel is gCThe operation formula is as follows:
wherein, gCFor each channel output value, the total output of C channels is G, then G ═ G1,g2,…,gC]Dimension of [1,1, C ]]. Obviously, gCThe larger the channel, the more characteristic information the channel contains.
The second step is that: and reducing the dimension and obtaining the nonlinear relation among channels. And performing operation processing on the output G obtained in the previous step through a full connection layer and a nonlinear function ReLU, wherein the full connection layer formula is as follows:
wherein R has a dimension of
Weight parameters of full connectivity layer
The layer reduces the number of channels from C to r by a scaling factor
Thereby reducing the amount of computation. Meanwhile, the nonlinear relation between the channels is obtained by using the activation function ReLU, namely the characteristic weight of each channel is learned.
The third step: and (5) increasing dimensions and obtaining the interesting characteristic score. The output R obtained in the previous step is put againPerforming operation processing through a full connection layer and a gate function Sigmoid, wherein the weight parameter of the full connection layer is
The full-connection layer operation of the layer restores the number of the channels to the original number, and the gate function Sigmoid further captures the nonlinear relationship between the channels, and then outputs an interesting characteristic Score FOI Score, the value of which is recorded as FOI _ S, and the operation formula is as follows:
wherein FOI _ S is a scalar comprising C numerical values, and [ FOI _ S1，FOI_S2，...，FOI_SC]The value of (a) represents the level of the score of the feature of interest of the corresponding channel;
the fourth step: and extracting the interesting features. Weighting the obtained interested feature scores to the previous tomato leaf feature map channel by channel, namely weighting each feature map [ W, H,1 [ ]]，[W,H,2]，...，[W,H,C]Respectively with FOI _ S1，FOI_S2，...，FOI_SCMultiplication is carried out according to the following operation formula:
XC＝FOI_SC·uC
the result output by the layer is C H multiplied by W tomato leaf feature maps, which are consistent with the input dimension, but the values on the tomato leaf feature maps are all recalibrated at the moment, namely the feature value of the feature of interest becomes larger, and irrelevant feature values become smaller and even suppressed, so that the extraction of the feature of interest is realized.
The specific implementation method of the light and small network model comprises the following steps:
inputting the preprocessed and shaped tomato leaf disease image into a sub-network GFE-Net, extracting simple features of the image through a 7 x 7 convolution layer, and performing convolution operation on the simple features for multiple times by utilizing a large number of 1 x 1 convolution layers and 3 x 3 convolution layers in 4 Fire modules to further extract richer global features. Because the obtained global feature map has higher dimension, 3 maximum pooling layers are adopted in the network to reduce the dimension of the feature map, and the parameter quantity and the calculation quantity of the network are effectively reduced under the condition of hardly influencing the network performance. After the global feature extraction is finished, the feature maps are input into a sub-module FOIE-Block, and the module compresses the feature maps and learns the feature weights from the channel dimension, so that interesting Features (FOI) are found, the expression of the FOI is enhanced, useless features are inhibited, and the network learning efficiency is improved. And (3) the new tomato leaf characteristic diagram after characteristic recalibration passes through a 1 x 1 convolutional layer and a global average pooling layer to obtain characteristic high-level semantic information output, and finally disease categories are output through softmax. The light and small network model provided by the invention can perform self-adaptive extraction on picture features, not only can extract global features, but also can enhance the expression of interesting features, and greatly improves the comprehensive performance of the network.
The accuracy and loss variation curves of the light and small network training set and the verification set are shown in fig. 6, the accuracy of the verification set gradually rises, the loss gradually falls, the network training is finished after 100 rounds of basic convergence are achieved. The confusion matrix on the test set is shown in table 1, and the identification accuracy of each disease category on the test set is 96.58% of early blight, 99.34% of health, 99.45% of late blight, 98.65% of leaf mold, 99.34% of mosaic disease, 99.44% of spot blight, 98.98% of bacterial spot disease and 99.38% of leaf mite disease respectively.
TABLE 1 confusion matrix on test set
The embodiments of the present invention are illustrative, but not restrictive, of the invention in any manner. The technical features or combinations of technical features described in the embodiments of the present invention should not be considered as being isolated, and they may be combined with each other to achieve a better technical effect. The scope of the preferred embodiments of the present invention may also include additional implementations, and this should be understood by those skilled in the art to which the embodiments of the present invention pertain.
Claims (7)
1. A light and small network self-adaptive tomato disease feature extraction method is characterized by comprising the following steps:
preprocessing a tomato leaf disease image;
orienting all the preprocessed tomato leaf disease images to be uniform through a orthomorphism mechanism;
constructing a light and small network model, and extracting the global characteristics and the interesting characteristics of the tomato leaf disease image; the light and small network model comprises a global feature extraction sub-network GFE-Net and an interested feature extraction sub-module FOIE-Block.
2. The method for extracting the disease features of the light and small network self-adaptive tomatoes according to claim 1, wherein all preprocessed tomato leaf disease images are oriented uniformly by a orthomorphism mechanism, and specifically the method comprises the following steps: unifying the directions of all tomato leaf disease images into that the leaf tips are upward and the leaf stalks are downward, so that the tomato leaves are positioned in the middle of the images, and ensuring that the deviation angle between the tomato leaves and the central line of the images is not more than +/-5 degrees.
3. The light small network adaptive tomato disease feature extraction method according to claim 1 or 2, characterized in that the working process of the orthomorphism mechanism is as follows:
acquiring a tomato leaf disease image;
graying the tomato leaf disease image and carrying out Gaussian filtering;
extracting edge characteristics of the tomato leaf disease image through a Canny operator, finding a minimum external rectangle, and obtaining a rotation angle theta;
and carrying out affine transformation on the original tomato leaf disease image by using the rotation angle theta.
4. The method for extracting the disease features of the light and small network self-adaptive tomatoes according to claim 1, wherein the global feature extraction sub-network GFE-Net is used for extracting the global features of the disease images of the leaves of the tomatoes, and comprises 4 Fire modules, 2 convolutional layers, 3 maximum pooling layers, 1 global average pooling layer, 1 dropout layer and 1 softmax layer.
5. The method for extracting the light and small network adaptive tomato disease features as claimed in claim 4, wherein each Fire module comprises a Squeeze layer and an Expand layer, each layer only consisting of convolution kernels of 1 x 1 or 3 x 3 size.
6. The method for extracting the tomato disease features in the light and small network self-adaption mode as claimed in claim 1, wherein the interesting feature extraction submodule FOIE-Block is used for extracting interesting features of tomato leaf disease images and comprises 1 global average pooling layer, 2 full connection layers, 2 activation functions and 1 matrix multiplication operation.
7. The method for extracting the light and small network adaptive tomato disease features according to claim 1 or 6, wherein the interested feature extraction submodule FOIE-Block is used for extracting the interested features of a tomato leaf disease image, and specifically comprises the following steps:
extracting dimensions of a tomato leaf characteristic graph output by the sub-network GFE-Net convolution layer from the global characteristics, wherein the dimensions are [ W, H, C]Inputting the tomato leaf characteristic diagram into an interested characteristic extraction submodule FOIE-Block, firstly passing through a global average pooling layer, summing all values on the tomato leaf characteristic diagram with the size of H multiplied by W, and averaging, wherein the output value of each channel is gCThe calculation formula is as follows:
wherein, gCFor the output value of each channel, uc(i, j) is the value in the feature map, the total output of the C channels is G, and G ═ G1,g2,…,gC]Dimension of [1,1, C ]]；
And carrying out operation processing on the total output G through a full connection layer and a nonlinear function ReLU, wherein the full connection layer formula is as follows:
wherein R has a dimension of
Weight parameters of full connectivity layer
The layer reduces the number of channels from C to r by a scaling factor
Meanwhile, the nonlinear relation between channels is obtained by using an activation function ReLU, namely the characteristic weight of each channel is learned;
and performing operation processing on the variable R through a full connection layer and a gate function Sigmoid, wherein the weight parameter of the full connection layer is
The full-connection layer operation of the layer restores the number of the channels to the original number, and the gate function sigmoid further captures the nonlinear relation among the channels, and then outputs an interesting characteristic Score FOI Score, the value of which is recorded as FOI _ S, and the operation formula is as follows:
wherein FOI _ S is a scalar comprising C numerical values, and [ FOI _ S1，FOI_S2，...，FOI_SC]The value of (a) represents the level of the score of the feature of interest of the corresponding channel;
weighting the obtained interesting characteristic scores on the tomato leaf characteristic maps channel by channel, namely weighting each tomato leaf characteristic map [ W, H,1]，[W,H,2]，...，[W,H,C]Are respectively provided withAnd FOI _ S1，FOI_S2，...，FOI_SCMultiplication is carried out according to the following operation formula:
XC＝FOI_SC·uC
the result output by the layer is C H multiplied by W tomato leaf feature maps which are consistent with the input dimension, but the values on the tomato leaf feature maps are all recalibrated at the moment, namely the feature value of the interesting feature becomes larger, and irrelevant feature values become smaller and even suppressed, so that the extraction of the interesting feature is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110621361.8A CN113344009B (en) | 2021-06-03 | 2021-06-03 | Light and small network self-adaptive tomato disease feature extraction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110621361.8A CN113344009B (en) | 2021-06-03 | 2021-06-03 | Light and small network self-adaptive tomato disease feature extraction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113344009A true CN113344009A (en) | 2021-09-03 |
| CN113344009B CN113344009B (en) | 2023-08-01 |
Family
ID=77473445
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110621361.8A Active CN113344009B (en) | 2021-06-03 | 2021-06-03 | Light and small network self-adaptive tomato disease feature extraction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113344009B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115050014A (en) * | 2022-06-15 | 2022-09-13 | 河北农业大学 | Small sample tomato disease identification system and method based on image text learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180144209A1 (en) * | 2016-11-22 | 2018-05-24 | Lunit Inc. | Object recognition method and apparatus based on weakly supervised learning |
| CN111696101A (en) * | 2020-06-18 | 2020-09-22 | 中国农业大学 | Light-weight solanaceae disease identification method based on SE-Inception |
| CN112183635A (en) * | 2020-09-29 | 2021-01-05 | 南京农业大学 | Method for realizing segmentation and identification of plant leaf lesions by multi-scale deconvolution network |
-
2021
- 2021-06-03 CN CN202110621361.8A patent/CN113344009B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180144209A1 (en) * | 2016-11-22 | 2018-05-24 | Lunit Inc. | Object recognition method and apparatus based on weakly supervised learning |
| CN111696101A (en) * | 2020-06-18 | 2020-09-22 | 中国农业大学 | Light-weight solanaceae disease identification method based on SE-Inception |
| CN112183635A (en) * | 2020-09-29 | 2021-01-05 | 南京农业大学 | Method for realizing segmentation and identification of plant leaf lesions by multi-scale deconvolution network |
Non-Patent Citations (1)
| Title |
|---|
| 李浩;徐航煌;郑恒宇;陈学永;: "基于无人机遥感图像的松材线虫病监测技术研究", 中国农机化学报, no. 09 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115050014A (en) * | 2022-06-15 | 2022-09-13 | 河北农业大学 | Small sample tomato disease identification system and method based on image text learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113344009B (en) | 2023-08-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Hong et al. | Tomato disease detection and classification by deep learning | |
| CN110148120B (en) | Intelligent disease identification method and system based on CNN and transfer learning | |
| CN110009043B (en) | Disease and insect pest detection method based on deep convolutional neural network | |
| Faisal et al. | IHDS: Intelligent harvesting decision system for date fruit based on maturity stage using deep learning and computer vision | |
| Babu et al. | Leaves recognition using back propagation neural network-advice for pest and disease control on crops | |
| CN111860330B (en) | Apple leaf disease identification method based on multi-feature fusion and convolutional neural network | |
| CN107067043A (en) | A kind of diseases and pests of agronomic crop detection method | |
| CN105631415A (en) | Video pedestrian recognition method based on convolution neural network | |
| CN109344699A (en) | Winter jujube disease recognition method based on depth of seam division convolutional neural networks | |
| CN111507319A (en) | Crop disease identification method based on deep fusion convolution network model | |
| CN111652247A (en) | Diptera insect identification method based on deep convolutional neural network | |
| CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
| Selvi et al. | Weed detection in agricultural fields using deep learning process | |
| CN111400536A (en) | Low-cost tomato leaf disease identification method based on lightweight deep neural network | |
| CN112084977A (en) | Image and time characteristic fused apple phenological period automatic identification method | |
| CN113344009A (en) | Light and small network self-adaptive tomato disease feature extraction method | |
| CN113221913A (en) | Agriculture and forestry disease and pest fine-grained identification method and device based on Gaussian probability decision-level fusion | |
| CN113076873A (en) | Crop disease long-tail image identification method based on multi-stage training | |
| Vora et al. | An ensemble of convolutional neural networks to detect foliar diseases in apple plants | |
| Omer et al. | Plant disease diagnosing based on deep learning techniques | |
| Liu et al. | Learning an optical filter for green pepper automatic picking in agriculture | |
| Agarwal et al. | Plant leaf disease classification using deep learning: A survey | |
| CN114937030A (en) | Phenotypic parameter calculation method for intelligent agricultural planting of lettuce | |
| CN115331104A (en) | Crop planting information extraction method based on convolutional neural network | |
| Rony et al. | BottleNet18: Deep Learning-Based Bottle Gourd Leaf Disease Classification |
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 |