CN113298150A - Small sample plant disease identification method based on transfer learning and self-learning - Google Patents
Small sample plant disease identification method based on transfer learning and self-learning Download PDFInfo
- Publication number
- CN113298150A CN113298150A CN202110574163.0A CN202110574163A CN113298150A CN 113298150 A CN113298150 A CN 113298150A CN 202110574163 A CN202110574163 A CN 202110574163A CN 113298150 A CN113298150 A CN 113298150A
- Authority
- CN
- China
- Prior art keywords
- learning
- sample
- model
- plant disease
- data set
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
A small sample plant disease identification method based on transfer learning and self-learning relates to the technical field of artificial intelligence, and aims at the problem that the identification precision of small sample plant diseases in the prior art is low, and comprises the following steps: acquiring a target domain small sample data set; II, secondly: introducing transfer learning, pre-training the deep learning model by using the base data set, and then transferring the pre-trained deep learning model to a target domain which is not crossed with the category in the base data set for learning; thirdly, the method comprises the following steps: introducing self-learning, training the migrated model by using the plant disease sample with the label in the target domain, deducing the category of all plant diseases without labels by using the trained model, giving pseudo labels to the plant diseases, and judging the reliability of the pseudo labels; fourthly, the method comprises the following steps: setting a reliability threshold, adding a pseudo-labeled sample with the reliability higher than the threshold into a labeled plant disease sample to obtain an expanded plant disease labeled sample set, and updating the model by using the expanded plant disease labeled sample set to obtain a final model.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a small sample plant disease identification method based on transfer learning and self-learning.
Background
The development of agriculture and forestry mainly depends on the quality and quantity of agriculture and forestry products, and in the field of development of modern agriculture and forestry, plant diseases and insect pests are serious biological disasters affecting the production safety and ecological safety of agriculture and forestry, and are also one of main factors causing economic, social and environmental changes. Therefore, the types of the diseases are discovered and identified as soon as possible, and the economic loss of the diseases to the agriculture and forestry production can be effectively reduced by taking targeted measures. Various plant disease preventive measures (such as biological and chemical methods) have serious influence on the ecological environment, so that real-time, safe, sustainable and accurate identification and monitoring are important before the preventive measures are implemented on the plant diseases.
Along with the rapid development of artificial intelligence technology, the automatic plant disease identification, monitoring and detection are gradually applied to the field of modern agriculture and forestry, and become a research hotspot of modern agriculture and forestry. In recent years, with the progress of deep learning network methods, researchers at home and abroad have successfully applied the deep learning network which takes a convolutional neural network as a basic characteristic extraction means to the field of plant diseases, and an image identification and classification method which takes deep learning as a means is used for modernization of agricultural development, so that the quality of agricultural products is improved, great contribution is made to solving the world temperature saturation problem, and great research results are obtained.
Although the plant disease identification technology based on deep learning has been developed, it does not mean that the identification rate is reliable under any conditions. Because the deep learning model contains a large number of parameters, the training of the model and the updating of the parameters depend on a large data set, but in practical application, a large amount of data is required to be collected and labeled, and the disease period of certain plant diseases is short, so that the collected data samples are too few. The recognition accuracy rate is low by using too few labeled samples to train the deep learning network model, and the phenomenon that the model is not easy to converge or over-fit easily occurs, so that the generalization capability of the deep learning model is poor, and the recognition accuracy is low.
For example, the citrus canker diseases in citrus crops are difficult to acquire high-quality scab images due to different growing environments, the variable shapes of the scabs, and the influence on the image quality caused by a plurality of reasons such as illumination, shooting angles and shooting skills during image acquisition. For epidemic diseases such as citrus canker and citrus huanglongbing, corresponding measures (burning and burying) must be taken to treat the diseases in time once the diseases are found out in order to prevent the diseases from spreading. At present, domestic epidemic prevention aiming at diseases such as citrus canker and citrus greening disease mainly relies on manual intervention, time and labor are wasted, time and labor are not in time, and the identification effect is still poor no matter how the generalization performance of the model is improved when the identification task of the small sample plant diseases is processed by deep learning.
Disclosure of Invention
The purpose of the invention is: aiming at the problem of low identification precision of small sample plant diseases in the prior art, a small sample plant disease identification method based on transfer learning and self-learning is provided.
The technical scheme adopted by the invention to solve the technical problems is as follows:
a small sample plant disease identification method based on transfer learning and self-learning comprises the following steps:
the method comprises the following steps: acquiring a target domain small sample data set;
step two: introducing transfer learning, pre-training the deep learning model by using the base data set, and then transferring the pre-trained deep learning model to a target domain which is not crossed with the category in the base data set for learning;
step three: introducing self-learning, training the migrated model by using the plant disease sample with the label in the target domain, deducing the category of all plant diseases without labels by using the trained model, giving pseudo labels to the plant diseases, and judging the reliability of the pseudo labels;
step four: setting a reliability threshold, adding a pseudo-labeled sample with the reliability higher than the threshold into a labeled plant disease sample to obtain an expanded plant disease labeled sample set, and updating the model by using the expanded plant disease labeled sample set to obtain a final model;
step five: and identifying the small sample plant diseases by using the final model.
Further, the deep learning model is ResNet-12.
Further, in the second step, the base data set adopts a public data set plantvillage.
Further, the reliability of the pseudo label is judged by an ICI algorithm.
Further, the target domain small sample dataset comprises 5 labeled plant disease sample sets per class and a plurality of unlabeled sample sets per class.
The invention has the beneficial effects that:
aiming at the problem that the identification precision is too low by adopting a deep learning method when the quantity of plant disease marking samples is too small, the small sample disease identification model based on transfer learning and self-learning is provided, the model can reflect the real characteristic distribution of unmarked samples, the characteristic information is learned robustly from the small sample disease identification model, credible pseudo samples are selected and added into an improved classifier of training iteration, the performance of the classifier is improved, the high identification precision can be finally achieved, a platform and technical support are provided for promoting the development of modern agriculture and forestry and the high quality and yield of agricultural products, and a new solution is provided for small sample plant leaf disease and insect identification.
Drawings
FIG. 1 is a schematic diagram of a sample of 10 types of target area diseases 1;
FIG. 2 is a schematic diagram of a sample of 10 types of target area diseases 2;
FIG. 3 is a schematic diagram of a sample of 10 types of target area diseases 3;
FIG. 4 is a sample schematic view of a class 10 target domain disease sample 4;
FIG. 5 is a sample schematic view of a class 10 target domain disease sample 5;
FIG. 6 is a sample schematic view of a class 10 target domain disease sample 6;
FIG. 7 is a sample schematic view of a class 10 target domain disease sample 7;
FIG. 8 is a sample schematic view 8 of a class 10 target domain disease;
FIG. 9 is a sample schematic view of a class 10 target domain disease sample 9;
FIG. 10 is a sample schematic 10 of a class 10 target area disease;
FIG. 11 is a flow chart of the present application;
FIG. 12 is a diagram of a ResNet-12 network architecture;
FIG. 13 is a comparison graph I of training loss and verification accuracy of four optimizers SGD, Adam, RMSprop and Amsgrad;
FIG. 14 is a graph II comparing training loss and validation accuracy of four optimizers SGD, Adam, RMSprop and Amsgrad;
fig. 15 is a schematic diagram of a confusion matrix for predicting a sample with 10 types of diseases.
Detailed Description
It should be noted that, in the present invention, the embodiments disclosed in the present application may be combined with each other without conflict.
The first embodiment is as follows: specifically, the present embodiment is described with reference to fig. 1, and the method for identifying a small sample plant disease based on migration learning and self-learning according to the present embodiment is characterized by including the following steps:
the method comprises the following steps: acquiring a target domain small sample data set;
step two: introducing transfer learning, pre-training the deep learning model by using the base data set, and then transferring the pre-trained deep learning model to a target domain which is not crossed with the category in the base data set for learning;
step three: introducing self-learning, training the migrated model by using the plant disease sample with the label in the target domain, deducing the category of all plant diseases without labels by using the trained model, giving pseudo labels to the plant diseases, and judging the reliability of the pseudo labels;
step four: setting a reliability threshold, adding a pseudo-labeled sample with the reliability higher than the threshold into a labeled plant disease sample to obtain an expanded plant disease labeled sample set, and updating the model by using the expanded plant disease labeled sample set to obtain a final model;
step five: and identifying the small sample plant diseases by using the final model.
The second embodiment is as follows: the present embodiment is a further description of the first embodiment, and the difference between the present embodiment and the first embodiment is that the deep learning model is ResNet-12.
The third concrete implementation mode: the present embodiment is further described with respect to the second embodiment, and the difference between the present embodiment and the second embodiment is that a public data set plantvillage is adopted as the small sample data set in the first step.
The fourth concrete implementation mode: the third embodiment is a further description of the third embodiment, and the difference between the third embodiment and the fourth embodiment is that the confidence level of the pseudo label is determined by an ICI algorithm.
The fifth concrete implementation mode: the present embodiment is a further description of a fourth embodiment, and the difference between the present embodiment and the fourth embodiment is that the target domain small sample dataset includes 5 labeled plant disease sample sets per class and a plurality of unlabeled sample sets per class.
Example (b):
the method comprises the following steps: a small sample dataset is constructed. The data set contains various plants and various diseases of one plant as much as possible, so that the authority of the data set is ensured, and the generalization performance of the model is fully verified.
Step two: ResNet-12 was introduced. Some lightweight networks such as ConvNet and MobileNet have shallow network structures, but the recognition accuracy is often unsatisfactory, and Wide-ResNet and DenseNet which have deep network structures have high recognition accuracy, but have many limitations in practical application due to the complex network, so ResNet-12 which has shallow network and reliable recognition accuracy is introduced as a feature extractor.
Step three: training a pre-training model. Based on the inspiration that human beings have the ability to quickly learn new concepts based on existing knowledge accumulation and a small number of instances, a small sample learning hypothesis model can be pre-trained on a data set (a base data set) with a large number of samples to obtain feature extraction ability or meta-learning ability and the like, and then the model needs to learn on a new data set (a target domain) without intersecting with the base data set.
Step four: self-learning is introduced. In a general small data set, for example, in the case of about 100 pieces of label data, it is possible to deal with the problem well by using only the migration learning. However, in some extreme cases, the obtained labeled samples are very few, for example, 1-5 labeled samples, and it is difficult to obtain a good identification accuracy even if the migration learning is used as a premise, so the invention combines a self-learning method to process the identification in the case of the extremely few labeled samples.
As a preferred embodiment of the present invention, in the first step, the data set uses a common data set plantvillage. The data set contains 38 types of health and disease samples of 14 plants in total, and 54302 sample pictures in total.
And dividing a small sample data set, wherein target domain data for testing during the division covers different plants and different diseases of the same plant as much as possible, so as to fully show the effectiveness and high generalization performance of the small sample plant disease identification model based on the migration learning and the self-learning provided by the invention under the condition of extremely few labeled samples. The pre-training model data finally used for training transfer learning are 14 types; the data used for verifying the loss and the recognition accuracy of each epoch in the training of the model are 14 types; and (3) 10 classes of small sample data (target domain), wherein each class has no more than 5 labeled samples for learning features, and the rest samples are not labeled.
All samples are stored in the same folder, a CSV file is created to label each sample, wherein the first column of the CSV file represents the name of the sample image, and the second column represents the category label corresponding to the sample.
As a preferred technical scheme of the invention, ResNet-12 has 4 residual blocks, each block containing three convolution layers of 3 x 3. The convolutional neural network has the powerful performances of lighter network layer number (only 12 layers), high training speed and high test precision, and is very suitable for practical application.
As a preferred technical scheme of the invention, the step three is to compare four optimizers commonly used for deep learning, namely SGD, Adam, RMSprop and Amsgrad, preferably select an optimizer with the highest convergence rate and the highest verification precision as the optimizer used for training the pre-training model, and obtain the optimal pre-training model by adjusting other parameters.
As a preferred technical solution of the present invention, the step of four self-learning refers to: for all 10 types of diseases, firstly, 5 labeled samples are used for learning each type to train a classifier, then 30 unlabeled samples are added into each type, and the classifier deduces the category of the unlabeled samples and assigns pseudo labels.
And deducing the credibility of the pseudo-labeled samples by an ICI algorithm, selecting credible pseudo-labeled samples to be added into a training and updating classifier, improving the classification performance of the classifier, returning the incredible pseudo-labeled samples, and deducing the category by the updated classifier again.
This process is repeated to ensure that all unlabeled samples are selected until a maximum of 5 pseudo-labeled samples of each class are excluded (the least reliable pseudo-labeled samples are removed) and the correct pseudo-labeled samples are used to update the classifier.
And at the moment, the performance of the classifier is optimal, and finally, classification prediction is carried out on the test set.
The hardware and software conditions for training the convolutional neural network of this embodiment are: all experiments were based on ubuntu 14.04; a CPU: ES-2650v4, 2.20GHz X13; GPU: NVIDIA TITAN Xp; python 3.6.12; pythrch 1.4.0; CUDA 10.0.130; cudnn 7.6.5.
Firstly, a small sample data set is constructed and used as the number of samples for training a pre-training model, the number of samples of a verification set and the number of samples (target domains) of a test set are respectively 35186 in 14 classes, 7962 in 14 classes and 500 in 10 classes, and the schematic diagram of the samples of the target domains in 10 classes is shown in fig. 1 to 10. Wherein, the 10 types of target domain diseases are cherry powdery mildew (figure 1), healthy corn (figure 2), corn rust (figure 3), grape black rot (figure 4), grape black measles (figure 5), grape leaf spot (figure 6), potato early blight (figure 7), potato late blight (figure 8), strawberry leaf scorch (figure 9) and tomato early blight (figure 10), the types are marked as 0-9, all pictures are cut into 168 x 168 before the test, each type of target domain contains 5 marked samples, 30 unmarked samples and 15 tested samples.
As shown in the flowchart of fig. 11, in the small sample plant disease identification method based on migration learning and self-learning implemented in this embodiment, a pre-training model is first trained on 35186 data sets of 14 classes to obtain feature extraction capability, and then the pre-training model is migrated to a target domain data set of 10 classes. The classifier is trained by learning 5 labeled samples in each class, then 30 unlabeled samples are added in each class, and the classifier infers the class of the unlabeled samples and assigns pseudo labels.
And deducing the credibility of the pseudo-labeled samples by an ICI algorithm, selecting credible pseudo-labeled samples to be added into a training and updating classifier, improving the classification performance of the classifier, returning the incredible pseudo-labeled samples, deducing the category by the updated classifier again and endowing corresponding pseudo labels.
The ICI algorithm judges the credibility of the pseudo label sample, which is the most important link in the invention, if the classifier is not endowed with the credibility of the pseudo label of the label-free sample for judgment, the classification performance of the classifier is reduced by adding the wrong pseudo label sample into training.
The following ICI algorithm is introduced to determine the confidence level of the pseudo-labeled sample:
in the statistical and financial fields, researchers are studying the problem of sporadic parameters in machine learning, which is believed to exist in addition to the conventional structural parameters in the model, for example, for a linear regression model expressed as:
wherein eiObeying a normal distribution with a mean value of 0, beta being the regression coefficient, gammai(x) is a parameter related to the samplei，yi) Features corresponding to sample pointsThe eigenvector, the linear regression model, is usually best when the value of the regression coefficient β is what.
Based on the thought, a linear regression model containing accidental parameters is constructed for the labeled samples and the pseudo-labeled samples, and the related structural parameters and accidental parameters are solved, so that the model can be optimized as follows:
unlike previous linear regression models, the ICI model focuses on the estimation of β
In the case of a weak strength, the strength,iis estimated by
The degree of fit of the sample points can be reflected. For example, γ may beiConsidered as a pair linear regression model
Correction of (1) | gammaiThe larger |, the worse the fit of the regression model to the sample point. When the label of the sample is a pseudo label, the confidence level of the sample point can be reflected.
To solve the above problem, the model loss function is:
wherein
Is Frobenius norm, gamma is incidental parameter, Y, X refers to finger label and characteristic input respectively,
as a penalty term, λ is a penalty term coefficient. The partial derivative of the equation for beta is found and made 0To obtain
(+ is the generalized inverse matrix). Based on previous analysis, this network is not concerned with
While solving, at the same time
Also relying on an estimate of γ, i.e., using γ to measure the confidence of each sample along its regularization path, the problem can be categorized as:
wherein, H ═ X (X)TX)+XT. Further define the
Then it can be simplified to:
in this case, the problem becomes a linear regression model using γ as a regression coefficient. For this model, all γ are nonzero when λ is 0, γ gradually decreases as λ increases, and all terms of γ are 0 when λ is ∞. Based on the initial assumptions about γ, when a certain term γ is in the processiAnd first becomes 0, which means that we can fit the sample point than a weaker linear regression model, and the sample point is more reliable. Therefore, a penalty term is set
Such that gamma is attenuated by each row, i.e. each sample point, according to gammaiAll the pseudo label samples are sorted according to the lambda value changed into 0, and the nearest credible subset is selected from the lambda values to be combined with the labeled sample setAnd 5, hoisting a new training linear classifier.
Fig. 12 is a feature extraction network ResNet-12 adopted by the present invention, which has a 12-layer network structure, and includes 4 residual blocks, each block includes three 3 × 3 convolutional layers, each convolutional layer is followed by a BatchNorm layer and adopts a Relu activation function, and a 2 × 2 maximal pooling layer is adopted after each block to reduce output, and the final output is a 512-dimensional feature vector.
For the characteristics of each disease sample, the invention adopts embedding to map the high-dimensional original data of the sample to the low dimension, adopts PCA dimension reduction to reduce the characteristic tensor of d 512 to d 10, thereby accelerating the training speed of the model and improving the final identification precision.
Comparing ConvNet, MobileNet, ResNet-12, Wide-ResNet and DenseNet-121 in the selection of the feature extraction network, comparing the recognition average accuracy of 5 networks and the detection time of a single picture on the basis of 5 marked samples and 30 unmarked samples of each type, and testing results are as follows:
as can be seen from the table, although detection time of a single picture is short, recognition accuracy is low, and the requirement of practical application cannot be met, while DenseNet-121 and WideResNet have high recognition accuracy, the DenseNet-121 network has deep layers, the WideResNet network has Wide structure, parameters are exponentially multiplied, time consumed by training of the single picture is long, and although training time of the single picture is not long, time difference is very large when tens of thousands of samples and hundreds of epochs are involved, and the practical application has many limitations, so ResNet-12 which gives consideration to light network and high recognition accuracy is adopted as a feature extraction network.
For the training of the migration model, as shown in fig. 13 and fig. 14, the present invention compares four optimizers Adam, Amsgrad, SGD of carry-over amount, and RMSprop commonly used for deep learning in the selection of the optimizers, and the momentum factor is set to 0.9 and all adopt a cross-entropy loss function. The training loss and the verification precision of each epoch are respectively subjected to new visual comparison, and according to the graph, the SGD optimizer is completely converged after the 32 th epoch, Adam and RMSgrad are respectively converged around the 61 th epoch, and Amsgrad is converged after the 90 epochs. Therefore, the SGD optimizer is superior in convergence speed, and therefore the SGD optimizer is selected to train the transfer learning amount pre-training model. The verification precision is not improved any more after the 33 rd epoch, so that the pre-training model corresponding to the 33 th epoch is the optimal model, and the pre-training model of the epoch is selected as the transfer learning model.
As shown in FIG. 15, the prediction results of 15 test samples of each type of all 10 plant diseases are visualized by a confusion matrix, and the overall classification precision is 88.67 percent
It should be noted that the detailed description is only for explaining and explaining the technical solution of the present invention, and the scope of protection of the claims is not limited thereby. It is intended that all such modifications and variations be included within the scope of the invention as defined in the following claims and the description.
Claims (5)
1. A small sample plant disease identification method based on transfer learning and self-learning is characterized by comprising the following steps:
the method comprises the following steps: acquiring a target domain small sample data set;
step two: introducing transfer learning, pre-training the deep learning model by using the base data set, and then transferring the pre-trained deep learning model to a target domain which is not crossed with the category in the base data set for learning;
step three: introducing self-learning, training the migrated model by using the plant disease sample with the label in the target domain, deducing the category of all plant diseases without labels by using the trained model, giving pseudo labels to the plant diseases, and judging the reliability of the pseudo labels;
step four: setting a reliability threshold, adding a pseudo-labeled sample with the reliability higher than the threshold into a labeled plant disease sample to obtain an expanded plant disease labeled sample set, and updating the model by using the expanded plant disease labeled sample set to obtain a final model;
step five: and identifying the small sample plant diseases by using the final model.
2. The small sample plant disease identification method based on transfer learning and self-learning as claimed in claim 1, wherein the deep learning model is ResNet-12.
3. The method for identifying diseases of small sample plants based on migratory learning and self-learning as claimed in claim 2, wherein the base data set in step two is a public data set plantvillage.
4. The small sample plant disease identification method based on transfer learning and self-learning as claimed in claim 3, wherein the confidence level of the judgment pseudo label is obtained by ICI algorithm.
5. The method for identifying small sample plant diseases based on migratory learning and self-learning as claimed in claim 4, wherein the target domain small sample data set comprises 5 labeled plant disease sample sets per class and a plurality of unlabeled sample sets per class.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110574163.0A CN113298150A (en) | 2021-05-25 | 2021-05-25 | Small sample plant disease identification method based on transfer learning and self-learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110574163.0A CN113298150A (en) | 2021-05-25 | 2021-05-25 | Small sample plant disease identification method based on transfer learning and self-learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113298150A true CN113298150A (en) | 2021-08-24 |
Family
ID=77324849
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110574163.0A Pending CN113298150A (en) | 2021-05-25 | 2021-05-25 | Small sample plant disease identification method based on transfer learning and self-learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113298150A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117237814A (en) * | 2023-11-14 | 2023-12-15 | 四川农业大学 | Large-scale orchard insect condition monitoring method based on attention mechanism optimization |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010036304A1 (en) * | 2000-01-22 | 2001-11-01 | Yang Mary M. | Visualization and processing of multidimensional data using prefiltering and sorting criteria |
| US20150102216A1 (en) * | 2013-09-16 | 2015-04-16 | Biodesix, Inc. | Classification Generation Method Using Combination of Mini-Classifiers with Regularization and Uses Thereof |
| CN104616031A (en) * | 2015-01-22 | 2015-05-13 | 哈尔滨工业大学深圳研究生院 | Transfer learning method and device |
| CN110717554A (en) * | 2019-12-05 | 2020-01-21 | 广东虚拟现实科技有限公司 | Image recognition method, electronic device, and storage medium |
| CN112069921A (en) * | 2020-08-18 | 2020-12-11 | 浙江大学 | Small sample visual target identification method based on self-supervision knowledge migration |
-
2021
- 2021-05-25 CN CN202110574163.0A patent/CN113298150A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20010036304A1 (en) * | 2000-01-22 | 2001-11-01 | Yang Mary M. | Visualization and processing of multidimensional data using prefiltering and sorting criteria |
| US20150102216A1 (en) * | 2013-09-16 | 2015-04-16 | Biodesix, Inc. | Classification Generation Method Using Combination of Mini-Classifiers with Regularization and Uses Thereof |
| CN104616031A (en) * | 2015-01-22 | 2015-05-13 | 哈尔滨工业大学深圳研究生院 | Transfer learning method and device |
| CN110717554A (en) * | 2019-12-05 | 2020-01-21 | 广东虚拟现实科技有限公司 | Image recognition method, electronic device, and storage medium |
| CN112069921A (en) * | 2020-08-18 | 2020-12-11 | 浙江大学 | Small sample visual target identification method based on self-supervision knowledge migration |
Non-Patent Citations (1)
| Title |
|---|
| 黄建平，陈镜旭等: "基于神经结构搜索的多种植物叶片病害识别", 《农业工程学报》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117237814A (en) * | 2023-11-14 | 2023-12-15 | 四川农业大学 | Large-scale orchard insect condition monitoring method based on attention mechanism optimization |
| CN117237814B (en) * | 2023-11-14 | 2024-02-20 | 四川农业大学 | Large-scale orchard insect condition monitoring method based on attention mechanism optimization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Thapa et al. | The plant pathology 2020 challenge dataset to classify foliar disease of apples | |
| Wilson et al. | Towards a generic artificial neural network model for dynamic predictions of algal abundance in freshwater lakes | |
| US10887188B2 (en) | System and methods for data evaluation through network sensitivity analysis | |
| Carbonell et al. | Biological invasion modifies the co‐occurrence patterns of insects along a stress gradient | |
| Matthiopoulos et al. | State-space modelling reveals proximate causes of harbour seal population declines | |
| CN111105045A (en) | Method for constructing prediction model based on improved locust optimization algorithm | |
| CN108038517A (en) | Based on the maize leaf disease recognition method for improving convolutional neural networks MODEL C ifar10 | |
| Nigam et al. | Wheat disease severity estimation: a deep learning approach | |
| CN115602337A (en) | Cryptocaryon irritans disease early warning method and system based on machine learning | |
| CN113298150A (en) | Small sample plant disease identification method based on transfer learning and self-learning | |
| CN108053423A (en) | A kind of multiple target animal tracking method and device | |
| Tirpak et al. | Assessing ecoregional‐scale habitat suitability index models for priority landbirds | |
| Peng et al. | Modified dragonfly algorithm based multilevel thresholding method for color images segmentation | |
| CN115510991A (en) | Sunlight greenhouse aphid early warning method based on 1DCNN-LSTM multi-source information fusion model | |
| CN115311657B (en) | Multi-source algae image target detection method, system, electronic equipment and storage medium | |
| Villa et al. | Understanding the role of chronic environmental disturbance in the context of island biogeographic theory | |
| Piyaratne et al. | A model to analyze weather impact on aphid population dynamics: an application on swallowtail catastrophe model | |
| Montejo‐Kovacevich et al. | Exploring context dependency in eco‐evolutionary patterns with the stick insect Timema cristinae | |
| Farah et al. | A Deep Learning-Based Approach for the Detection of Infested Soybean Leaves | |
| Venu et al. | Disease Identification in Plant Leaf Using Deep Convolutional Neural Networks | |
| Alshammari et al. | Optimized recurrent neural network mechanism for olive leaf disease diagnosis based on wavelet transform | |
| CN108875774A (en) | Butterfly category identification method, electronic equipment, storage medium and system | |
| Panchami et al. | Rice Leaf Disease Detection and Diagnosis Using Convolution Neural Network | |
| Lawson | Reducing Uncertainties in Conservation Decision-Making for American Alligators | |
| Fowler | The role of microbial mutualists in hosts' resilience to environmental variation across space and time |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210824 |