CN114676769A - Visual transform-based small sample insect image identification method - Google Patents
Visual transform-based small sample insect image identification method Download PDFInfo
- Publication number
- CN114676769A CN114676769A CN202210282073.9A CN202210282073A CN114676769A CN 114676769 A CN114676769 A CN 114676769A CN 202210282073 A CN202210282073 A CN 202210282073A CN 114676769 A CN114676769 A CN 114676769A
- Authority
- CN
- China
- Prior art keywords
- image
- insect
- layer
- insects
- type
- 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.)
- Withdrawn
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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a visual transform-based small sample insect image identification method, which comprises the steps of firstly searching images of various insects by utilizing a search engine, and manually labeling labels on the images; then constructing a pre-training model taking a visual Transformer as a core, and performing optimization training on parameters in the pre-training model by using a training set; removing a classifier in the pre-training model, randomly extracting a small number of image samples of each type of insects in a training set and a testing set, inputting the image samples into a visual transform to extract image characteristics, and calculating the average value of each type of samples to be used as the image prototype characteristics of each type of insects for storage; and finally, acquiring an insect image on line, inputting the insect image into a Transformer to extract image characteristics, calculating the distance between the image characteristics and the representation of each type of insect image, and outputting the insect type with the nearest distance as the type of the image. The invention uses a small amount of training samples to finish the classification identification of the insects, and can overcome the technical problem that a large amount of image samples are needed when the convolutional neural network used in the current insect identification is trained.
Description
Technical Field
The invention relates to a small sample insect image identification method based on a visual Transformer, and belongs to the field of computer vision.
Background
Insects are the most numerous animal groups on earth, and they are of various kinds and shapes. From the viewpoint of human interest, these insects can be generally classified into two major categories, pest and beneficial insect. Pests can harm the growth of crops and ornamental flowers, and serious economic loss can be brought to human beings, while beneficial insects can bring nutritious food or abundant industrial materials to human beings. Therefore, effective identification of insects is of great significance for protecting the ecological environment and promoting the development of social production. Compared with the traditional manual identification method, the automatic insect identification method based on the computer vision technology has the advantages that the labor cost is greatly reduced, the identification efficiency is high, and the subjectivity of the identification result is small.
In recent years, with the rapid development of deep learning technology, computer vision tasks such as image classification, target detection, semantic segmentation, and the like have been developed in a breakthrough manner. The convolutional neural network with strong characteristic representation capability is also applied to the field of insect identification, for example, the leather sweet and the like (leather sweet, Xin, Penmingjie, Wushiyu. the dragonflies insect identification method based on the regional suggestion network, application No. 202110480792.7 discloses a dragonflies insect identification method based on the regional suggestion network, which uses a large-capacity convolutional neural network ResNet50 to extract characteristics to classify and identify the dragonflies insects; the catadioptric insect is a diptera insect on the basis of deep convolution nerve network, application number 202010471036.3, which adopts the RetinaNet target detection model and uses the ResNeXt network as the feature extraction network, and adds the improved attention module in the feature extraction network to classify and recognize the diptera insect. However, the technical problems still remain in the above methods as follows: (1) the category for identifying insects is limited to insects of the dragonflies or diptera, and image datasets containing four orders of insects are still lacking; (2) parameters contained in convolutional neural networks such as ResNet50 or ResNeXt are large in capacity, and a large number of image samples are needed for optimization training to ensure good recognition performance. Therefore, in order to overcome the above problems in the field of insect identification, a small sample image identification method based on visual Transformer is disclosed to segment an insect image so as to complete the classification and identification of insects.
Disclosure of Invention
In view of the problems in the prior art, the present invention provides a method for identifying small sample insect images based on visual transform, so as to solve the above technical problems.
In order to achieve the purpose, the invention adopts the technical scheme that: a small sample insect image identification method based on visual Transformer comprises the following steps;
s1, constructing an insect data set, namely searching images of various insects in seven major categories of insects by using an image search engine such as a Baidu search engine, and manually labeling the searched images to finish the construction of the insect data set;
s2, constructing a pre-training model, wherein the pre-training model mainly comprises a linear mapping layer of image small blocks, a visual Transformer, a full connection layer and a Softmax layer;
s3, optimizing the pre-training model, randomly dividing the insect data set into a training set, a verification set and a test set, and performing optimization training on the parameters in the pre-training model by using the training set;
s4, constructing a small sample insect recognition model, removing a classifier in a pre-training model, randomly extracting a small number of image samples of each type of insects in a training set and a testing set, inputting the image samples into a visual Transformer to extract image characteristics, and calculating the average value of each type of samples to be used as the prototype characteristics of each type of insects for storage;
And S5, online testing of the insect images, acquiring the insect images online, inputting the insect images into a Transformer to extract image characteristics, calculating the distance between the insect images and the representation of each type of insect image, and outputting the insect type with the nearest distance as the type of the image.
Further, the specific steps of S1 are as follows;
s11: taking coleoptera, lepidoptera, diptera, hymenoptera, hemiptera, orthoptera and dragonflies as the meta categories in the insect data set, and searching 100 images for 20 small categories in each meta category by using an image search engine such as a Baidu image search engine;
s12: all images were uniformly normalized to 224 x 224 dimensions and their categories were manually labeled.
Further, the specific steps of S2 are as follows;
s21: the parameter of the linear mapping layer of the image small block is a projection coefficient matrixWherein, P is the size of the image small block, C is 3 to represent the channel number of the original image, D represents the dimension of the image small block after linear mapping;
s22: visual Transformer is denoted TθWhere θ is a parameter, it is composed of L layers of transform encoders, each layer of encoder is composed of a normalization layer, a multi-head attention, and a multi-layer perceptron.
Further, the specific steps of S3 are as follows;
S31: randomly dividing an insect data set into a training set, a verification set and a test set, wherein the training set is used for optimizing model parameters, the verification set is used for setting hyper-parameters, and the test set is used for constructing an insect identification model of a small sample;
s32: randomly extracting a batch of images in a training set, wherein the total number of the images in the batch is T, and the ith image is represented as xiWith a corresponding artificial label value of yiDividing an image with a resolution of H × W × C into a number of image patches of P × P size with a sampling step of S pixels, assuming that the number of image patches isM, conversion thereof to P2The C-dimensional vectors are spliced to obtain the size of MxP2C, the ith image is represented asThe corresponding coordinate vector is represented as
S33: using projection matricesConvert it into a fixed length vector of D dimension, at which time the ith image is represented asInsert with a P2The vector of C is used as the global characteristic representation of the image, and the representation of the ith image is converted into the global characteristic representation of the imageObtaining a first layer input of a Transformer encoder;
s34: for the l-th layer of the encoder, the input is recorded asOutput is asThe calculation process is:
wherein MSA is Multi-Head Self-Attention (MSA), MLP is Multi-Layer Perceptron (MLP), LN is normalized Layer (Layer Norm, LN);
S35: output of L-th layerThe 1 st vector in (1)Inputting the data into a full connection layer and a Softmax layer to obtain a class probability output value of the ith image
S36: the other images in the batch are also processed in sequence according to steps S32, S33, S34, and S35, the total cross entropy loss function for the batch of images is calculated, and the parameters in the model are optimized using the cross entropy loss function.
Further, the specific steps of S4 are as follows;
s41: removing a full connection layer and a Softmax layer in the pre-training model;
s42: randomly extracting a small number of image samples of each type of insects in a training set, a verification set and a testing set, wherein the number of the image samples of each type of insects is less than or equal to 5, inputting the image samples into a pre-training model, the total number of the types of the insects in the image data set is N, the total number of the extracted samples of each type is K less than or equal to 5, and the kth image in the nth insect type is represented as xkDividing the image into a plurality of image small blocks with the size of P multiplied by P by taking the sampling step length as S pixels, and then obtaining the image small blocks through a linear mapping layerInput into a Transformer encoder of an L layer to obtainThe 1 st vectorAs a feature representation of the image;
s43: calculate KThe average value of the image sample characteristics is stored as the prototype characteristic of each type of insects, namely the prototype characteristic R of the nth type of insects nThe calculation formula of (c) is:
further, the specific steps of S5 are as follows;
s51: collecting insect image x on line, dividing the insect image into a plurality of image small blocks with the sampling step length of S pixels, and obtaining the insect image x through a linear mapping layerInput into a Transformer encoder of an L layer to obtainThe 1 st vectorAs a feature representation of the image;
s52: and (3) calculating the Euclidean distance between the features of the image x and the stored prototype features of each type of insects, wherein the calculation formula of the Euclidean distance and the nth prototype feature is as follows:
the insect class closest to the image is output as the class of the image.
The invention has the beneficial effects that: the invention uses a small amount of training samples to finish the classification identification of the insects, and can overcome the technical problem that a large amount of image samples are needed when the convolutional neural network used in the current insect identification is trained; the invention uses visual Transform as a feature extractor of the image, and can overcome the problem that insects are blocked in natural images.
Drawings
FIG. 1 is a flow chart of a visual transform-based small sample insect image recognition method according to the present invention;
FIG. 2 is an exemplary image of an insect data set of the present invention;
FIG. 3 is a schematic structural diagram of a visual transducer according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood, however, that the detailed description herein of specific embodiments is intended to illustrate the invention and not to limit the scope of the invention.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs, and the terms used herein in the specification of the present invention are for the purpose of describing particular embodiments only and are not intended to limit the present invention.
As shown in FIG. 1, a visual Transformer-based small sample insect image identification method comprises the following steps;
s1 construction of an insect data set,
s11, taking coleoptera, lepidoptera, diptera, hymenoptera, hemiptera, orthoptera and dragonfly as the meta-categories in the insect data set, and searching 100 images for 20 small categories in each meta-category by using an image search engine such as Baidu and the like; an example image in each meta category in the dataset is shown in FIG. 2;
s12: all the images are unified and normalized to the size of 224 multiplied by 224, and the categories of the images are manually marked;
S2 construction of a pre-trained model,
s21: the parameter of the linear mapping layer of the image small block is a projection coefficient matrixWherein, P is the size of the image small block, C is 3 to represent the channel number of the original image, D represents the dimension of the image small block after linear mapping;
s22: visual Transformer is represented asTθWherein θ is a parameter, and is composed of L layers of transform encoders, each layer of encoder is composed of a normalization layer, a multi-head attention and a multi-layer perceptron, and the structure diagram is shown in fig. 3;
s3, optimizing a pre-training model, namely randomly dividing an insect data set into a training set, a verification set and a test set, and performing optimization training on parameters in the pre-training model by using the training set;
s31: randomly dividing an insect data set into a training set, a verification set and a test set, wherein the training set is used for optimizing model parameters, the verification set is used for setting hyper-parameters, and the test set is used for constructing an insect identification model of a small sample;
s32: randomly extracting a batch of images in a training set, wherein the total number of the images in the batch is T, and the ith image is represented as xiWith a corresponding artificial label value of yiDividing the image with the resolution of H multiplied by W multiplied by C into a plurality of image small blocks with the size of P multiplied by P by taking the sampling step size as S pixels, and converting the image small blocks into P multiplied by P by assuming that the number of the image small blocks is M 2The C-dimensional vectors are spliced to obtain the size of MxP2C, then the ith image is represented asThe corresponding coordinate vector is represented as
S33: using projection matricesConvert it into a fixed length vector in D dimension, at which time the ith image is represented asInsert with a P2The vector of C is used as the global characteristic representation of the image, and the representation of the ith image is converted into the global characteristic representation of the imageObtaining a first layer input of a Transformer encoder;
s34: for the l-th layer of the encoder, the input is recorded asOutput is asThe calculation process is:
wherein MSA is Multi-Head Self-Attention (MSA), MLP is Multi-Layer Perceptron (MLP), LN is normalized Layer (Layer Norm, LN);
s35: output of L-th layerThe 1 st vector ofInputting the data into a full connection layer and a Softmax layer to obtain a class probability output value of the ith image
S36: processing other images in the batch in sequence according to steps S32, S33, S34 and S35, calculating a total cross entropy loss function of the batch of images, and optimizing parameters in the model by using the cross entropy loss function;
s4 construction of a small sample insect recognition model,
s41: removing a full connection layer and a Softmax layer in the pre-training model;
S42: in training set, verification set andrandomly extracting a small number of image samples of each type of insects in the test set, inputting the image samples of each type of insects into a pre-training model, wherein the total number of the categories of the insects in the image data set is N, the total number of the extracted samples of each type is K less than or equal to 5, and the kth image in the nth insect category is represented as xkThe image is divided into a plurality of image small blocks with the size of P multiplied by P by taking the sampling step length as S pixels, and then the image small blocks are obtained by a linear mapping layerInput into Transformer encoder of L layer to obtainThe 1 st vector of the vectorAs a feature representation of the image;
s43: calculating the average value of the K image sample characteristics as the prototype characteristic of each type of insects for storage, namely the prototype characteristic R of the nth type of insectsnThe calculation formula of (2) is as follows:
and S5, online testing of the insect images, acquiring the insect images online, inputting the insect images into a Transformer to extract image characteristics, calculating the distance between the insect images and the representation of each type of insect image, and outputting the insect type with the nearest distance as the type of the image.
S51: collecting insect image x on line, dividing the insect image into a plurality of image small blocks with the sampling step length of S pixels, and obtaining the insect image x through a linear mapping layer Input into Transformer encoder of L layer to obtainIt is put intoMiddle 1 st vectorAs a feature representation of this image;
s52: and calculating Euclidean distance between the features of the image x and the stored prototype features of each type of insects, wherein the Euclidean distance and the nth prototype feature are calculated according to the formula:
the insect category closest to the insect is output as the category of the image.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents or improvements made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (6)
1. A visual transform-based small sample insect image identification method is characterized by comprising the following steps;
s1, constructing an insect data set, namely searching images of various insects in seven major categories of the insects by utilizing an image search engine such as Baidu and the like, and manually labeling the searched images so as to finish the construction of the insect data set;
s2, constructing a pre-training model, wherein the pre-training model mainly comprises a linear mapping layer of image small blocks, a visual Transformer, a full connection layer and a Softmax layer;
s3, optimizing a pre-training model, namely randomly dividing an insect data set into a training set, a verification set and a test set, and performing optimization training on parameters in the pre-training model by using the training set;
S4, constructing a small sample insect recognition model, removing a classifier in a pre-training model, randomly extracting a small number of image samples of each type of insects in a training set and a testing set, inputting the image samples into a visual Transformer to extract image characteristics, and calculating the average value of each type of samples to be used as the prototype characteristics of each type of insects for storage;
and S5, carrying out online test on the insect images, collecting the insect images online, inputting the insect images into a Transformer to extract image characteristics, calculating the distance between the insect images and the representation of each type of insect images, and outputting the insect type with the closest distance as the type of the image.
2. The visual Transformer-based small sample insect image recognition method according to claim 1, wherein the specific steps of S1 are as follows;
s11: taking coleoptera, lepidoptera, diptera, hymenoptera, hemiptera, orthoptera and dragonflies as the meta categories in the insect data set, and searching 100 images for 20 small categories in each meta category by using an image search engine such as a Baidu image search engine;
s12: all images were uniformly normalized to 224 x 224 dimensions and their categories were manually labeled.
3. The visual Transformer-based small sample insect image recognition method according to claim 1, wherein the specific steps of S2 are as follows;
S21: the parameter of the linear mapping layer of the image small block is a projection coefficient matrixWherein, P is the size of the image small block, C is 3 to represent the channel number of the original image, D represents the dimension of the image small block after linear mapping;
s22: visual Transformer is denoted TθWhere θ is a parameter, it is composed of L layers of transform encoders, each layer of encoder is composed of a normalization layer, a multi-head attention, and a multi-layer perceptron.
4. The visual Transformer-based small sample insect image recognition method according to claim 1, wherein the specific steps of S3 are as follows;
s31: randomly dividing an insect data set into a training set, a verification set and a test set, wherein the training set is used for optimizing model parameters, the verification set is used for setting hyper-parameters, and the test set is used for constructing an insect identification model of a small sample;
s32: randomly extracting a batch of images in a training set, wherein the total number of the images in the batch is T, and the ith image is represented as xiWith a corresponding artificial label value of yiDividing the image with the resolution of H multiplied by W multiplied by C into a plurality of image small blocks with the size of P multiplied by P by taking the sampling step size as S pixels, and converting the image small blocks into P multiplied by P by assuming that the number of the image small blocks is M 2The C-dimensional vectors are spliced to obtain the size of MxP2C, then the ith image is represented asThe corresponding coordinate vector is represented as
S33: using projection matricesConvert it into a fixed length vector of D dimension, at which time the ith image is represented asInsert with a P2The vector of C is used as the global characteristic representation of the image, and the representation of the ith image is converted into the global characteristic representation of the imageObtaining a first layer input of a Transformer encoder;
s34: for the l-th layer of the encoder, the input is recorded asOutput is asThe calculation process is:
wherein MSA is Multi-Head Self-Attention (MSA), MLP is Multi-Layer Perceptron (MLP), LN is normalized Layer (Layer Norm, LN);
s35: output of L-th layerThe 1 st vector ofInputting the data into a full connection layer and a Softmax layer to obtain a class probability output value of the ith image
S36: the other images in the batch are processed in sequence according to steps S32, S33, S34 and S35, the total cross entropy loss function of the batch of images is calculated, and the parameters in the model are optimized by the cross entropy loss function.
5. The visual Transformer-based small sample insect image recognition method according to claim 1, wherein the specific steps of S4 are as follows;
S41: removing a full connection layer and a Softmax layer in the pre-training model;
s42: randomly extracting a small number of image samples of each type of insects in a training set, a verification set and a testing set, wherein the number of the image samples of each type of insects is less than or equal to 5, inputting the image samples into a pre-training model, and the total number of types of insects in an image data setN, the total number of samples taken for each class K ≦ 5, and the kth image in the nth insect class denoted xkDividing the image into a plurality of image small blocks with the size of P multiplied by P by taking the sampling step length as S pixels, and then obtaining the image small blocks through a linear mapping layerInput into a Transformer encoder of an L layer to obtainThe 1 st vectorAs a feature representation of the image;
s43: calculating the average value of the K image sample characteristics as the prototype characteristic of each type of insects for storage, namely the prototype characteristic R of the nth type of insectsnThe calculation formula of (2) is as follows:
6. the visual Transformer-based small sample insect image recognition method according to claim 1, wherein the specific steps of S5 are as follows;
s51: collecting insect image x on line, dividing the insect image into a plurality of image small blocks with the sampling step length of S pixels, and obtaining the insect image x through a linear mapping layer Input into Transformer encoder of L layer to obtainThe 1 st vector of the vectorAs a feature representation of this image;
s52: and (3) calculating the Euclidean distance between the features of the image x and the stored prototype features of each type of insects, wherein the calculation formula of the Euclidean distance and the nth prototype feature is as follows:
the insect class closest to the image is output as the class of the image.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210282073.9A CN114676769A (en) | 2022-03-22 | 2022-03-22 | Visual transform-based small sample insect image identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210282073.9A CN114676769A (en) | 2022-03-22 | 2022-03-22 | Visual transform-based small sample insect image identification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114676769A true CN114676769A (en) | 2022-06-28 |
Family
ID=82075143
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210282073.9A Withdrawn CN114676769A (en) | 2022-03-22 | 2022-03-22 | Visual transform-based small sample insect image identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114676769A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115879404A (en) * | 2022-11-07 | 2023-03-31 | 华南理工大学 | Circuit netlist simulation acceleration optimization method based on visual Transform |
CN115879109A (en) * | 2023-02-06 | 2023-03-31 | 山东省计算中心(国家超级计算济南中心) | Malicious software identification method based on visual transform |
CN115953665A (en) * | 2023-03-09 | 2023-04-11 | 武汉人工智能研究院 | Target detection method, device, equipment and storage medium |
CN116597384A (en) * | 2023-06-02 | 2023-08-15 | 中国人民解放军国防科技大学 | Space target identification method and device based on small sample training and computer equipment |
CN115879404B (en) * | 2022-11-07 | 2024-10-22 | 华南理工大学 | Simulation acceleration optimization method based on visual transformation circuit netlist |
-
2022
- 2022-03-22 CN CN202210282073.9A patent/CN114676769A/en not_active Withdrawn
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115879404A (en) * | 2022-11-07 | 2023-03-31 | 华南理工大学 | Circuit netlist simulation acceleration optimization method based on visual Transform |
CN115879404B (en) * | 2022-11-07 | 2024-10-22 | 华南理工大学 | Simulation acceleration optimization method based on visual transformation circuit netlist |
CN115879109A (en) * | 2023-02-06 | 2023-03-31 | 山东省计算中心(国家超级计算济南中心) | Malicious software identification method based on visual transform |
CN115953665A (en) * | 2023-03-09 | 2023-04-11 | 武汉人工智能研究院 | Target detection method, device, equipment and storage medium |
CN116597384A (en) * | 2023-06-02 | 2023-08-15 | 中国人民解放军国防科技大学 | Space target identification method and device based on small sample training and computer equipment |
CN116597384B (en) * | 2023-06-02 | 2024-03-05 | 中国人民解放军国防科技大学 | Space target identification method and device based on small sample training and computer equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110442684B (en) | Class case recommendation method based on text content | |
CN105631479B (en) | Depth convolutional network image labeling method and device based on non-equilibrium study | |
CN114676769A (en) | Visual transform-based small sample insect image identification method | |
CN110516095B (en) | Semantic migration-based weak supervision deep hash social image retrieval method and system | |
CN111611924B (en) | Mushroom identification method based on deep migration learning model | |
CN107480261A (en) | One kind is based on deep learning fine granularity facial image method for quickly retrieving | |
CN107451565B (en) | Semi-supervised small sample deep learning image mode classification and identification method | |
CN111198959A (en) | Two-stage image retrieval method based on convolutional neural network | |
CN110942091B (en) | Semi-supervised few-sample image classification method for searching reliable abnormal data center | |
CN109241995B (en) | Image identification method based on improved ArcFace loss function | |
CN114332621B (en) | Disease and pest identification method and system based on multi-model feature fusion | |
Zeng et al. | Identification of maize leaf diseases by using the SKPSNet-50 convolutional neural network model | |
CN109858557B (en) | Novel semi-supervised classification method for hyperspectral image data | |
CN115019104A (en) | Small sample remote sensing image classification method and system based on multi-source domain self-attention | |
CN115359353A (en) | Flower identification and classification method and device | |
CN115019103A (en) | Small sample target detection method based on coordinate attention group optimization | |
CN114329031A (en) | Fine-grained bird image retrieval method based on graph neural network and deep hash | |
CN114022688A (en) | Real-time livestock identity recognition method | |
CN108052981B (en) | Image classification method based on nonsubsampled Contourlet transformation and convolutional neural network | |
CN115661739A (en) | Vineyard pest fine-grained identification method based on attribute characteristic knowledge graph | |
CN114663791A (en) | Branch recognition method for pruning robot in unstructured environment | |
CN111401519B (en) | Deep neural network unsupervised learning method based on similarity distance in object and between objects | |
Li et al. | Early drought plant stress detection with bi-directional long-term memory networks | |
CN114610941A (en) | Cultural relic image retrieval system based on comparison learning | |
CN113516209A (en) | Comparison task adaptive learning method for few-sample intention recognition |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220628 |