CN116363460A - High-resolution remote sensing sample labeling method based on topic model - Google Patents
High-resolution remote sensing sample labeling method based on topic model Download PDFInfo
- Publication number
- CN116363460A CN116363460A CN202310355409.4A CN202310355409A CN116363460A CN 116363460 A CN116363460 A CN 116363460A CN 202310355409 A CN202310355409 A CN 202310355409A CN 116363460 A CN116363460 A CN 116363460A
- Authority
- CN
- China
- Prior art keywords
- sample set
- training sample
- training
- model
- features
- 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
- 238000002372 labelling Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 130
- 230000000007 visual effect Effects 0.000 claims abstract description 51
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000013139 quantization Methods 0.000 claims abstract description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 238000013135 deep learning Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 238000007637 random forest analysis Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013398 bayesian method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V10/7753—Incorporation of unlabelled data, e.g. multiple instance learning [MIL]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/54—Extraction of image or video features relating to texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a high-resolution remote sensing sample labeling method based on a topic model, which comprises the following steps: acquiring a training sample set and a sample set to be marked; extracting traditional features and deep features of a training sample set; carrying out feature quantization on the traditional features and the deep features of the training sample set to obtain a word bag representation of the training sample set; constructing a visual theme model, and inputting a training sample set expressed by a word bag into the visual theme model to obtain theme distribution of the training sample set; constructing and training a weak classifier model by using the topic distribution and the labeling information of the training sample set, wherein the weak classifier model comprises at least two weak classifiers; and labeling the sample set to be labeled by using the weak classifier model. According to the method, a small quantity of training sample sets with labels are used for assisting in iterative training of a plurality of weak classifiers, the accuracy of the weak classifiers is improved, the generalization capability of the finally obtained weak classifier model is strong, and the remote sensing samples and the remote sensing images can be accurately classified.
Description
Technical Field
The invention relates to a remote sensing image labeling method, in particular to a high-resolution remote sensing sample labeling method based on a topic model, and belongs to the field of remote sensing image classification.
Background
The remote sensing technology is an emerging comprehensive technology developed in the sixties of the century, is closely related to the scientific technologies such as space, electron optics, computer, geography and the like, and is one of the most powerful technical means for researching the earth resource environment. The semantic automatic labeling of the remote sensing image is the basis of the organization and the index of the remote sensing image, and particularly, with the explosive increase of the remote sensing data in recent years, the data volume of the high-resolution remote sensing image is extremely large, the difficulty in directly searching and acquiring the image is larger, and the semantic labeling meaning of the high-resolution remote sensing image is more obvious. However, the existing manual direct labeling is time-consuming and labor-consuming, the feasibility is not strong, and an efficient labeling method from the image semantic hierarchy is urgently needed.
The current automatic labeling model is mainly classified into a labeling model based on discriminant and a labeling model based on a generating formula. The labeling model based on the discriminant is a supervised classification model, a classifier is trained for each type of keywords, and the keywords of the labeled images are transmitted to the unlabeled images according to the similarity between the visual contents of the images. Based on the generated annotation model, a probability association model is constructed between the image and the annotation words, and the probability of annotating the keyword sequences with similar semantics is higher for the image with similar visual characteristics.
Many existing algorithms still rely on the number of tagged images, i.e., the tagging model typically requires a large amount of tagged data to train, which is labor intensive. In the case where the number of the marked data is insufficient, the generalization performance of many models is not high. In a specific task, the marked data is difficult to obtain, and the unmarked data is easy to obtain. Therefore, how to fully utilize a large amount of unmarked data and limited priori knowledge and combine the self characteristics of the remote sensing image is a main way for improving the semantic annotation of the remote sensing image.
Disclosure of Invention
Based on the technical problems, the invention provides a high-resolution remote sensing sample labeling method based on a topic model, which adopts semi-supervised learning, only needs a small part of samples with labels to train and learn parts, assists a weak classifier to train out a weak classifier model with stronger generalization capability, can label samples to be labeled more accurately,
the invention provides a high-resolution remote sensing sample labeling method based on a topic model, which comprises the following steps:
s1, acquiring a high-resolution remote sensing sample set, wherein the high-resolution remote sensing sample set comprises a training sample set and a sample set to be marked, and the training sample set contains marking information;
s2, extracting features of the training sample set by using a traditional method and a deep learning method to obtain traditional features and deep features of the training sample set;
s3, carrying out feature quantization on the traditional features and the deep features of the training sample set to obtain a word bag representation of the training sample set;
s4, constructing a visual theme model, and inputting a training sample set expressed by a word bag into the visual theme model to obtain theme distribution of the training sample set;
s5, constructing and training a weak classifier model by using the topic distribution and the labeling information of the training sample set, wherein the weak classifier model comprises at least two weak classifiers;
s6, inputting the sample set to be marked into a weak classifier model to obtain marking information of the sample set to be marked.
Further, step S2 includes:
constructing a deep convolutional neural network, and extracting features of a training sample set by using the deep convolutional neural network to obtain deep features of the training sample set, wherein the deep convolutional neural network is VGG, resNet or AlexNet;
and carrying out feature extraction on the training sample set according to a traditional algorithm to obtain traditional features of the training sample set, wherein the traditional algorithm is a SIFT algorithm, a gray level co-occurrence matrix algorithm or a HOG feature extraction algorithm.
Further, step S3 includes:
clustering the traditional features and the deep features of the training sample set by adopting a clustering method, wherein each clustering center is a visual word, and all the visual words form a visual dictionary;
and mapping the traditional features and the deep features of the training sample set to a visual dictionary, and carrying out histogram statistics on visual words of the training sample set to obtain the word bag representation of the training sample set.
Further, step S4 includes:
and constructing a visual theme model, inputting a training sample set into the visual theme model, and learning the training sample set expressed by the word bags to obtain visual word theme distribution and potential theme distribution of the training sample set.
Further, the weak classifier model includes T weak classifiers, and step S5 includes:
training a plurality of weak classifiers according to potential theme distribution and labeling information of each training sample, and solving the classification error rate of each weak classifier on the training sample set to obtain the optimal weak classifier of the round of training;
and updating the sample weight, and obtaining T weak classifiers after T iterations.
Further, step S6 includes:
S2-S4 are executed on the sample set to be marked, and the theme distribution of the sample set to be marked is obtained;
learning potential topic distribution of the sample set to be annotated according to the visual word topic distribution of the training sample set;
inputting the potential theme distribution of the sample set to be annotated into a trained weak classifier model to obtain a word sequence to be annotated of the sample set to be annotated;
and selecting the word to be annotated with the highest probability to construct an annotation word set of the sample set to be annotated, and obtaining the annotation information of the sample set to be annotated.
The beneficial effects of the invention are as follows: the invention provides a high-resolution remote sensing sample labeling method based on a topic model, which comprises the steps of firstly extracting features of a training sample set with a label to obtain traditional features and deep features, and jointly representing images by the traditional features and the deep features with good properties to avoid the problem of image information deletion caused by single features; carrying out feature quantization on the traditional features and the deep features to obtain word bag representations of a training sample set; learning a training sample set represented by a word bag according to the topics by utilizing a visual topic model to obtain topic distribution of the training sample set; training a plurality of weak classifiers in an assisted manner according to the subject distribution of the training sample set, wherein the weak classifiers form a weak classifier model; and after the training process is finished, labeling the sample set to be labeled by using the trained weak classifier model. According to the method, a semi-supervised learning mode is utilized, a small number of training sample sets with labels are used for assisting in training a plurality of weak classifiers, in the iterative training process, the accuracy of the weak classifiers is improved, the generalization capability of the finally obtained weak classifier model is high, and the remote sensing samples and the remote sensing images can be accurately classified.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
fig. 2 is a schematic representation of an LDA diagram according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which are derived by a person skilled in the art based on the embodiments of the invention, fall within the scope of protection of the invention.
Referring to fig. 1, the present invention provides a high-resolution remote sensing sample labeling method based on a topic model, which includes:
s1, acquiring a high-resolution remote sensing sample set, wherein the high-resolution remote sensing sample set comprises a training sample set and a sample set to be marked, and the training sample set contains marking information;
s2, extracting features of the training sample set by using a traditional method and a deep learning method to obtain traditional features and deep features of the training sample set;
s3, carrying out feature quantization on the traditional features and the deep features of the training sample set to obtain a word bag representation of the training sample set;
s4, constructing a visual theme model, and inputting a training sample set expressed by a word bag into the visual theme model to obtain theme distribution of the training sample set;
s5, constructing and training a weak classifier model by using the topic distribution and the labeling information of the training sample set, wherein the weak classifier model comprises at least two weak classifiers;
s6, inputting the sample set to be marked into a weak classifier model to obtain marking information of the sample set to be marked.
The method comprises the steps of obtaining a high-resolution remote sensing sample set, and dividing the high-resolution remote sensing sample set into a training sample set comprising a small number of samples and a sample set to be marked comprising a large number of samples, wherein the training sample set is provided with marking information, and the sample set to be marked does not contain the marking information.
And respectively extracting features of the training sample set by using a traditional algorithm and a deep learning method:
(1) The feature points of the training sample set are extracted by using a traditional algorithm, the obtained features are traditional features, the traditional algorithm can be a SIFT algorithm, a gray level co-occurrence matrix algorithm or a HOG feature extraction algorithm, the obtained features are SIFT features, texture features or HOG features respectively, and the traditional features of the invention can be one or more of the features.
(2) Constructing a deep convolutional neural network, and extracting features of a training sample set by using the deep convolutional neural network to obtain deep features of the training sample set, wherein the deep convolutional neural network is VGG, resNet, alexNet or other deep learning networks;
and clustering the traditional features and the deep features of the training sample set by adopting a clustering method, wherein the clustering method can be a K-means clustering method, a plurality of clustering centers are obtained by clustering, each clustering center represents one visual word, and all the visual words form a visual dictionary.
And carrying out feature quantization on the traditional features and the deep features of the training sample sets, and carrying out histogram statistics on the visual words of each training sample set. The specific process of characteristic quantification is as follows: the distance between two feature vectors is calculated, then the cluster center closest to the feature vector is searched in the visual dictionary, and the feature vector is mapped to the visual word. By counting the occurrence times of the visual words of each training sample, a visual word histogram of a single training sample corresponding to the feature can be generated, and the histograms of the traditional feature and the depth feature are connected to obtain the visual histogram of each training sample, namely the word bag representation of the training sample set.
And constructing a visual theme model, and inputting a training sample set expressed by a word bag into the visual theme model to obtain theme distribution of the training sample set. The visual topic model may be an LDA model assuming a training sample consisting of N visual wordsWherein->Is the nth visual word in the training sample. The training sample set contains M training samples, denoted +.>. Each training sample is generated by mixing K potential topics, and each topic is formed by the parameters ofThe generated one is probability distribution on the visual dictionary. Parameter->And parameters->Subject parameters of +.>Dirichlet distribution of>Representing the mixing ratio of the distribution of the image subjects, +.>Representing the distribution of visual words under each topic, w representing the visual words of the image. The LDA diagram model is shown in fig. 2.
Inputting the training sample set into the LDA model to obtain the topic distribution of the training sample set. Given parametersAnd->Mixed topic distribution->The joint probability distribution for topic Z and training sample W is as follows:
for a pair ofBy integrating and summing over z, the edge distribution of the training samples is obtained as follows:
finally, multiplying the edge probability of each training sample to obtain the probability distribution of the training sample set:
parameters in the LDA model can be estimated by empirical Bayesian methods, using a variational distribution to approximate likelihood of failure to calculate, the method can be used to estimate the likelihood of failure to calculateAnd->Maximization.
The LDA model learns the training sample set to obtain theme distribution, including potential theme distribution of the training sample set and visual word theme distribution of the training sample set. The LDA model comprises an independence assumption that the visual word topic distribution of the training sample set is independent of the specific training sample, so that the visual word topic distribution is also applicable to the sample set to be annotated.
Constructing a weak classifier model according to potential theme distribution and labeling information of a training sample set, wherein the weak classifier model comprises at least two weak classifiers, the weak classifiers can be random forest classifiers, decision tree classifiers or other weak classifiers, and the weak classifier model can be composed of the same weak classifier or different weak classifiers. In this embodiment, T random forest classifiers are used, which specifically includes: (1) And performing first iteration, training a plurality of random forest classifiers according to potential theme distribution and labeling information of each training sample, and then solving the classification error rate of each random forest classifier on the training sample set to obtain the optimal random forest classifier for the round of training. (2) And updating the sample weight, and obtaining T random forest classifiers after T iterations. T random forest classifiers form a weak classifier model. The sample weight is the probability distribution of the training sample.
Extracting features of a sample set to be marked to obtain traditional features and deep features of the sample set to be marked, carrying out feature quantization on the traditional features and the deep features of the sample set to be marked to obtain word bag representation of the sample set to be marked, learning potential theme distribution of the sample set to be marked according to visual word theme distribution of the training sample set, inputting the potential theme distribution of the sample set to be marked into a trained weak classifier model to obtain a word sequence to be marked of the sample set to be marked, and selecting the word to be marked with the highest probability to construct the word set to be marked of the sample set to be marked to obtain marking information of the sample set to be marked.
The beneficial effects of the invention are as follows: the invention provides a high-resolution remote sensing sample labeling method based on a topic model, which comprises the steps of firstly extracting features of a training sample set with a label to obtain traditional features and deep features, and jointly representing images by the traditional features and the deep features with good properties to avoid the problem of image information deletion caused by single features; carrying out feature quantization on the traditional features and the deep features to obtain word bag representations of a training sample set; learning a training sample set represented by a word bag according to the topics by utilizing a visual topic model to obtain topic distribution of the training sample set; training a plurality of weak classifiers in an assisted manner according to the subject distribution of the training sample set, wherein the weak classifiers form a weak classifier model; and after the training process is finished, labeling the sample set to be labeled by using the trained weak classifier model. According to the method, a semi-supervised learning mode is utilized, a small number of training sample sets with labels are used for assisting in training a plurality of weak classifiers, in the iterative training process, the accuracy of the weak classifiers is improved, the generalization capability of the finally obtained weak classifier model is high, and the remote sensing samples and the remote sensing images can be accurately classified.
Claims (6)
1. A high-resolution remote sensing sample labeling method based on a topic model is characterized by comprising the following steps:
s1, acquiring a high-resolution remote sensing sample set, wherein the high-resolution remote sensing sample set comprises a training sample set and a sample set to be marked, and the training sample set contains marking information;
s2, extracting features of the training sample set by using a traditional method and a deep learning method to obtain traditional features and deep features of the training sample set;
s3, carrying out feature quantization on the traditional features and the deep features of the training sample set to obtain a word bag representation of the training sample set;
s4, constructing a visual theme model, and inputting a training sample set expressed by a word bag into the visual theme model to obtain theme distribution of the training sample set;
s5, constructing and training a weak classifier model by using the topic distribution and the labeling information of the training sample set, wherein the weak classifier model comprises at least two weak classifiers;
s6, inputting the sample set to be marked into a weak classifier model to obtain marking information of the sample set to be marked.
2. The method for labeling a high-resolution remote sensing sample based on a topic model as claimed in claim 1, wherein the step S2 comprises:
constructing a deep convolutional neural network, and extracting features of a training sample set by using the deep convolutional neural network to obtain deep features of the training sample set, wherein the deep convolutional neural network is VGG, resNet or AlexNet;
and carrying out feature extraction on the training sample set according to a traditional algorithm to obtain traditional features of the training sample set, wherein the traditional algorithm is a SIFT algorithm, a gray level co-occurrence matrix algorithm or a HOG feature extraction algorithm.
3. The method for labeling a high-resolution remote sensing sample based on a topic model according to claim 2, wherein step S3 comprises:
clustering the traditional features and the deep features of the training sample set by adopting a clustering method, wherein each clustering center is a visual word, and all the visual words form a visual dictionary;
and mapping the traditional features and the deep features of the training sample set to a visual dictionary, and carrying out histogram statistics on visual words of the training sample set to obtain the word bag representation of the training sample set.
4. The method for labeling a high-resolution remote sensing sample based on a topic model as claimed in claim 3, wherein the step S4 comprises:
and constructing a visual theme model, inputting a training sample set into the visual theme model, and learning the training sample set expressed by the word bags to obtain visual word theme distribution and potential theme distribution of the training sample set.
5. The method for labeling high-resolution remote sensing samples based on the topic model according to claim 4, wherein the weak classifier model includes T weak classifiers, and step S5 includes:
training a plurality of weak classifiers according to potential theme distribution and labeling information of each training sample, and solving the classification error rate of each weak classifier on the training sample set to obtain the optimal weak classifier of the round of training;
and updating the sample weight, and obtaining T weak classifiers after T iterations.
6. The method for labeling a high-resolution remote sensing sample based on a topic model according to claim 5, wherein step S6 comprises:
S2-S4 are executed on the sample set to be marked, and the theme distribution of the sample set to be marked is obtained;
learning potential topic distribution of the sample set to be annotated according to the visual word topic distribution of the training sample set;
inputting the potential theme distribution of the sample set to be annotated into a trained weak classifier model to obtain a word sequence to be annotated of the sample set to be annotated;
and selecting the word to be annotated with the highest probability to construct an annotation word set of the sample set to be annotated, and obtaining the annotation information of the sample set to be annotated.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310355409.4A CN116363460A (en) | 2023-04-06 | 2023-04-06 | High-resolution remote sensing sample labeling method based on topic model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310355409.4A CN116363460A (en) | 2023-04-06 | 2023-04-06 | High-resolution remote sensing sample labeling method based on topic model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116363460A true CN116363460A (en) | 2023-06-30 |
Family
ID=86915651
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310355409.4A Pending CN116363460A (en) | 2023-04-06 | 2023-04-06 | High-resolution remote sensing sample labeling method based on topic model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116363460A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117218396A (en) * | 2023-10-08 | 2023-12-12 | 深圳海角兰科技有限公司 | Automatic visual sample data labeling method and system based on large model |
-
2023
- 2023-04-06 CN CN202310355409.4A patent/CN116363460A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117218396A (en) * | 2023-10-08 | 2023-12-12 | 深圳海角兰科技有限公司 | Automatic visual sample data labeling method and system based on large model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112115995B (en) | Image multi-label classification method based on semi-supervised learning | |
Yang et al. | Visual sentiment prediction based on automatic discovery of affective regions | |
CN111914558B (en) | Course knowledge relation extraction method and system based on sentence bag attention remote supervision | |
CN111966917B (en) | Event detection and summarization method based on pre-training language model | |
CN105760507B (en) | Cross-module state topic relativity modeling method based on deep learning | |
CN112819023B (en) | Sample set acquisition method, device, computer equipment and storage medium | |
CN110222218B (en) | Image retrieval method based on multi-scale NetVLAD and depth hash | |
CN108509521B (en) | Image retrieval method for automatically generating text index | |
Guan et al. | Tag-based Weakly-supervised Hashing for Image Retrieval. | |
Niu et al. | Knowledge-based topic model for unsupervised object discovery and localization | |
CN110598022B (en) | Image retrieval system and method based on robust deep hash network | |
CN113946677B (en) | Event identification and classification method based on bidirectional cyclic neural network and attention mechanism | |
CN114896434B (en) | Hash code generation method and device based on center similarity learning | |
CN112199532A (en) | Zero sample image retrieval method and device based on Hash coding and graph attention machine mechanism | |
CN110647907A (en) | Multi-label image classification algorithm using multi-layer classification and dictionary learning | |
Su et al. | Semi-supervised knowledge distillation for cross-modal hashing | |
CN116363460A (en) | High-resolution remote sensing sample labeling method based on topic model | |
CN113987175A (en) | Text multi-label classification method based on enhanced representation of medical topic word list | |
Lu et al. | Web multimedia object classification using cross-domain correlation knowledge | |
Perdana et al. | Instance-based deep transfer learning on cross-domain image captioning | |
Foumani et al. | A probabilistic topic model using deep visual word representation for simultaneous image classification and annotation | |
Papapanagiotou et al. | Improving concept-based image retrieval with training weights computed from tags | |
Chu et al. | Social-guided representation learning for images via deep heterogeneous hypergraph embedding | |
CN115994239A (en) | Prototype comparison learning-based semi-supervised remote sensing image retrieval method and system | |
CN113434668B (en) | Deep learning text classification method and system based on model fusion |
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 |