CN116310463A - Remote sensing target classification method for unsupervised learning - Google Patents
Remote sensing target classification method for unsupervised learning Download PDFInfo
- Publication number
- CN116310463A CN116310463A CN202310594028.1A CN202310594028A CN116310463A CN 116310463 A CN116310463 A CN 116310463A CN 202310594028 A CN202310594028 A CN 202310594028A CN 116310463 A CN116310463 A CN 116310463A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- training
- model
- sensing target
- weight
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 95
- 238000002372 labelling Methods 0.000 claims abstract description 15
- 230000008859 change Effects 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 abstract description 2
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000008569 process Effects 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
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- G06V10/77—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
-
- 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)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a remote sensing target classification method for unsupervised learning, which relates to the technical field of remote sensing image processing, and comprises the following steps: s1, extracting characteristics of a remote sensing target without labels through a ResNet model to generate training data; s2, performing coarse classification on all remote sensing targets in the training data by using a clustering algorithm, and giving an initial annotation to all unlabeled remote sensing target images according to a clustering result to generate an initial annotation set Y 0 The method comprises the steps of carrying out a first treatment on the surface of the S3, calculating a feature center for the formed feature clusters respectively; s4, generating an initial weight set P for all remote sensing target images based on the feature center 0 The method comprises the steps of carrying out a first treatment on the surface of the S5, the initial annotation set Y is subjected to a pre-training model 0 And an initial weight set P 0 Performing first training; s6, carrying out iterative updating on the labels; and S7, repeating iterative updating, and stopping training until the change of the labeling set is smaller than a threshold value, so that the classification of the remote sensing targets is completed.
Description
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a remote sensing target classification method for unsupervised learning.
Background
The classification and identification of remote sensing targets plays a very important role in many practical applications such as city planning, land use analysis, disaster relief, automatic mapping, and modern agriculture. With the continuous development of the aerospace remote sensing technology in recent years, the acquired remote sensing target images are greatly increased.
Interpretation of remote sensing targets by deep learning technology is widely applied, however, remote sensing target classification based on deep learning depends on large-scale marked remote sensing targets for training, marking of remote sensing images requires huge labor cost, and cumbersome marking process prevents the application of the completely supervised methods in remote sensing image classification.
In order to overcome the limitation, the invention provides a remote sensing target classification method for unsupervised learning.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a remote sensing target classification method for unsupervised learning, which aims to solve the technical problems.
The technical method adopted for solving the technical problems is as follows: in a remote sensing target classification method for unsupervised learning, the improvement comprising: the method comprises the following steps of:
s1, extracting characteristics of a remote sensing target without labels through a ResNet model to generate training data;
s2, using a clustering algorithm to perform clustering on training dataCoarse classification is carried out on all remote sensing targets of the model, and an initial label is given to all unlabeled remote sensing target images according to the clustering result, so as to generate an initial label set Y 0 ;
S3, calculating a feature center for the formed feature clusters respectively
wherein ,representing categoriescIs the first of (2)iThe image of the object to be sensed is a single image,N c representing categoriescIs used to determine the total number of remote sensing target images,u c representing categoriescIs characterized by a remote sensing target image feature center;
s4, generating an initial weight set P for all remote sensing target images based on the feature center 0
wherein ,representing computing categoriescIs the first of (2)iThe feature of each remote sensing target image and the corresponding feature center thereof>L2 distance of->Feature extractor representing a pre-trained model +.>Is of the categorycIs the first of (2)iWeights of the remote sensing target images;
s5, passing through a pre-training modelC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 Training for the first time to obtain a modelC 1 (f 1 (·));
S6, performing iterative updating on the labels, in the first stepkIn the training, the last model is usedC k-1 (f k-1 (·)) update of the firstkLabeling set during secondary model training;
Iteratively updating the weights, in the firstkIn the training, the output result Pre of the last training model is used k-1 As the known information, a weight generated based on the known information is usedAnd the firstkWeight of remote sensing target image in model training 1 time>Generate the firstkWeight set of remote sensing target image during secondary model training>;
S7, repeating iterative updating until the change of the labeling set is smaller than the threshold value, stopping training, and completing classification of the remote sensing targets
In the above method, the step S1 includes the following steps:
s11, pre-training a ResNet model on an ImageNet;
s12, carrying out feature extraction on a large number of unlabeled remote sensing targets by the ResNet model to generate training data;
feature extractionIs a feature extractor through ResNet modelf(. Cndot.) and full connection classifierC(. Cndot.) for any unlabeled remote sensing target imagexExtracting the last layer of featuresConv5_x。
In the above method, in the step S5, the model is trained in advanceC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 The model training objective function for the first training is
Wherein the labeling set is Y 0 ={},/>True labels representing category c, weight set P 0 ={/>},|c|The number of feature clusters, i.e. the number of categories, +.>Cross entropy loss between the prediction result of the input sample by the calculation model and the corresponding class label is calculated.
In the above method, in the step S6, the label is iteratively updated, in the first stepkIn the training, the last model is usedC k-1 (f k-1 (·)) update of the firstkLabeling set during secondary model trainingComprising the following steps:
iteratively updating the labels, in the firstkIn the training, the last model is usedC k-1 (f k-1 (-)) obtaining a set Pre of recognition results of all current remote sensing target images k-1= {C k-1 (f k-1 ())/>;
Then, iteratively updating the set of labels
In the above method, in the step S6, the weights are iteratively updated, in the first stepkIn the training, the output result Pre of the last training model is used k-1 As the known information, a weight generated based on the known information is usedAnd the firstkWeight of remote sensing target image in model training 1 time>Generate the firstkWeight set of remote sensing target image during secondary model training>Comprising the following steps:
iteratively updating the weights, in the firstkIn the training, the output result Pre of the last training model is used k-1 Generating weights as known information
wherein ,representing calculation numberk-1-time dieOutput->With labeling->The larger the distance is, which indicates that the misclassification possibility of the remote sensing target image is larger, namely, the smaller weight is given in the subsequent training;
the L2 distance is a euclidean distance between two vectors, the distance is used to measure the difference between the two vectors, in this model, the difference between the prediction result of the model on the sample and the real label of the sample is used to measure, and aN N-dimensional vector a= [ a1, a2, a3, & gt, aN is set], B=[b1,b2.b3,...,bN]Wherein a is j Representing the j-th dimension of vector A, the L2 distance between the two vectors is defined as
Using the generated weightsAnd the firstkWeight of remote sensing target image in model training 1 time>Iteratively updating the weight set to generate the firstkWeight set of remote sensing target image during secondary model training>
In the first placekTraining for the second time, and iteratively updating the weight set,is the original weight set +.>Is->Is the currently generated weight set +.>Is used for controlling the duty ratio of the two part weight sets when the iteration is updated respectively.
The beneficial effects of the invention are as follows: the method has the advantages that the characteristic extraction and the labeling are carried out on a large number of remote sensing targets without labels through the model, so that the weight generation of remote sensing target images is realized, the classification and identification model training of the remote sensing targets is realized, and the classification and identification of a large number of remote sensing targets under the condition of no manual labeling of experts are realized.
Drawings
Fig. 1 is a schematic diagram of a remote sensing target classification method without supervised learning according to the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
The conception, specific structure, and technical effects produced by the present invention will be clearly and completely described below with reference to the embodiments and the drawings to fully understand the objects, features, and effects of the present invention. It is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments, and that other embodiments obtained by those skilled in the art without inventive effort are within the scope of the present invention based on the embodiments of the present invention. In addition, all the coupling/connection relationships referred to in the patent are not direct connection of the single-finger members, but rather, it means that a better coupling structure can be formed by adding or subtracting coupling aids depending on the specific implementation. The technical features in the invention can be interactively combined on the premise of no contradiction and conflict.
Referring to fig. 1, the invention provides a remote sensing target classification method for unsupervised learning, which comprises the following steps:
s1, extracting characteristics of a remote sensing target without labels through a ResNet model to generate training data;
specifically, the step S1 includes the following steps:
s11, pre-training a ResNet model on an ImageNet;
s12, extracting features of a large number of untagged remote sensing targets by a pre-trained ResNet model to generate training data; feature extraction is a feature extractor through ResNet modelf(. Cndot.) and full connection classifierC(. Cndot.) for any unlabeled remote sensing target imagexExtracting the last layer of featuresConv5_x。
S2, extracting image features of all unlabeled remote sensing target training data through a pre-trained ResNet model, roughly classifying all remote sensing targets in the training data by using a clustering algorithm, and giving an initial label to all unlabeled remote sensing target images according to a clustering result to generate an initial label set Y 0 ;
S3, after the clustering of the remote sensing target image features is completed, calculating a feature center for the formed feature clusters respectively
wherein ,representing categoriescIs the first of (2)iThe image of the object to be sensed is a single image,N c representing categoriescIs used to determine the total number of remote sensing target images,u c representing categoriescIs characterized by a remote sensing target image feature center;
s4, generating an initial weight set P for all remote sensing target images based on the feature center 0
wherein ,representing computing categoriescIs the first of (2)iThe feature of each remote sensing target image and the corresponding feature center thereof>L2 distance of->Feature extractor representing a pre-trained model +.>Is of the categorycIs the first of (2)iAnd weighting the remote sensing target images. By calculating the distances of all images of the class from the feature center and forming a one-dimensional vector, weights are then generated for them by the softmax function. Since the farther from the feature center the image has a greater probability of misclassification, its weight should be smaller. The final weight is thus obtained by subtracting the value generated by the softmax function from 1, the weight generated at this time being called the initial weight, the weight set being P 0 And (3) representing.
The influence of inaccurate labels on the model performance is reduced by calculating a weight for each remote sensing target training sample.
After the initial labeling and initial weight generation of all the remote sensing target images are completed, the next training of the remote sensing target classification recognition model can be carried out.
S5, passing through a pre-training modelC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 Training for the first time to obtain a modelC 1 (f 1 (·));
Specifically, in the step S5, a model is trained in advanceC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 The model training objective function for the first training is
Wherein the labeling set is Y 0 ={},/>True labels representing category c, weight set P 0 ={/>},|c|The number of feature clusters, i.e. the number of categories, +.>Representing the cross entropy loss between the prediction result of the calculation model on the input sample and the real label of the corresponding category.
Because the labels and weights generated on the remote sensing target images are inaccurate in the first training, the invention updates the labels and weights through subsequent iteration, timely adjusts the training direction of the model according to the feedback of the training result, and avoids the influence of inaccurate labels and weights on the performance of the remote sensing target classification recognition model.
S6, performing iterative updating on the labels, in the first stepkIn the training, the last model is usedC k-1 (f k-1 (·)) update of the firstkLabeling set during secondary model training;
Specifically, the labels are iteratively updated, in the following stepkIn the training, the last model is usedC k-1 (f k-1 (·)) update of the firstkLabeling set during secondary model trainingComprising the following steps:
iteratively updating the labels, inFirst, thekIn the training, the last model is usedC k-1 (f k-1 (-)) obtaining a set Pre of recognition results of all current remote sensing target images k-1= {C k-1 (f k-1 ())/>;
Then, iteratively updating the set of labels
iteratively updating the weights, in the firstkIn the training, the output result of the model after the last training is usedAs the known information, use is made of the weight generated based on the known information +.>And the firstkWeight of remote sensing target image in model training 1 time>Generate the firstkWeight set of remote sensing target image during secondary model training>;
Specifically, the method comprises the following steps:
iteratively updating the weights, in the firstkIn the training, the output result Pre of the last training model is used k-1 Generating weights as known information
wherein ,representing calculation numberkOutput of the-1 order model->With labeling->The larger the distance is, the larger the possibility of misclassification of the remote sensing target image is, namely, the smaller weight is given in the subsequent training, so that the influence on model training is reduced;
the L2 distance is a euclidean distance between two vectors, the distance is used to measure the difference between the two vectors, in this model, the difference between the prediction result of the model on the sample and the real label of the sample is used to measure, and aN N-dimensional vector a= [ a1, a2, a3, & gt, aN is set], B=[b1,b2.b3,...,bN]Let a be j Representing the j-th dimension of vector A, the L2 distance between the two vectors is defined as
Then, the generated weights are utilizedAnd the firstkWeighting of remote sensing target image during model training 1 timeIteratively updating the weight set to generate the firstkWeight set of remote sensing target image during secondary model training>
During the kth training and the iterative updating of the weight set,is the original weight set +.>Is->Is the currently generated weight set +.>Is used for controlling the duty ratio of the two part weight sets when the iteration is updated respectively.
S7, repeating iterative updating until the change of the labeling set is smaller than the threshold value, stopping training, and completing classification of the remote sensing targets
According to the method, the characteristics of a large number of remote sensing targets without labels are extracted and labeled through the model, so that the weight generation of the remote sensing target images is realized, the classification recognition model training of the remote sensing targets is realized, and the classification recognition of a large number of remote sensing targets under the condition of no manual labeling of an expert is realized.
While the preferred embodiment of the present invention has been described in detail, the present invention is not limited to the embodiments, and those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention, and these equivalent modifications or substitutions are included in the scope of the present invention as defined in the appended claims.
Claims (5)
1. A remote sensing target classification method for unsupervised learning is characterized in that: the method comprises the following steps of:
s1, extracting characteristics of a remote sensing target without labels through a ResNet model to generate training data;
s2, performing coarse classification on all remote sensing targets in the training data by using a clustering algorithm, and giving an initial annotation to all unlabeled remote sensing target images according to a clustering result to generate an initial annotation set Y 0 ;
S3, calculating a feature center for the formed feature clusters respectively
wherein ,representing categoriescIs the first of (2)iThe image of the object to be sensed is a single image,N c representing categoriescIs used to determine the total number of remote sensing target images,u c representing categoriescIs characterized by a remote sensing target image feature center;
s4, generating an initial weight set P for all remote sensing target images based on the feature center 0
wherein ,the ith remote sensing target image feature representing the calculation class c and the corresponding feature center +.>L2 distance of->Feature extractor representing a pre-trained model +.>Is of the categorycIs the first of (2)iWeights of the remote sensing target images;
s5, passing through a pre-training modelC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 Training for the first time to obtain a modelC 1 (f 1 (·));
S6, performing iterative updating on the labels, and using the last model in the kth trainingC k-1 (f k-1 (·)) updating the set of labels at the time of model training at the kth time;
Iteratively updating the weight, and using the output result Pre of the model after the last training in the kth training k-1 As the known information, a weight generated based on the known information is usedWeights of remote sensing target image in k-1 model training>Generating a weight set of the remote sensing target image during kth model training>;
S7, repeating iterative updating until the change of the labeling set is smaller than the threshold value, stopping training, and completing classification of the remote sensing targets
2. The method for classifying remote sensing targets for unsupervised learning of claim 1, further comprising: the step S1 comprises the following steps:
s11, pre-training a ResNet model on an ImageNet;
s12, carrying out feature extraction on a large number of unlabeled remote sensing targets by the ResNet model to generate training data;
feature extraction is a feature extractor through ResNet modelf(. Cndot.) and full connection classifierC(. Cndot.) for any unlabeled remote sensing target imagexExtracting the last layer of featuresConv5_x。
3. The method for classifying remote sensing targets for unsupervised learning of claim 1, further comprising: in the step S5, the model is trained in advanceC 0 (f 0 (. Cndot.)) for the initial set of labels Y 0 And an initial weight set P 0 The model training objective function for the first training is
Wherein the labeling set is Y 0 ={},/>True labels representing category c, weight set P 0 ={/>},|c|The number of feature clusters, i.e. the number of categories, being clustersQuantity (S)>And representing cross entropy loss between the prediction result of the calculation model on the input sample and the corresponding class label.
4. The method for classifying remote sensing targets for unsupervised learning of claim 1, further comprising: in the step S6, the label is iteratively updated, and the last model is used during the kth trainingC k-1 (f k-1 (·)) updating the set of labels at the time of model training at the kth timeComprising the following steps:
iteratively updating the labels, in the firstkIn the training, the last model is usedC k-1 (f k-1 (-)) obtaining a set Pre of recognition results of all current remote sensing target images k-1= {C k-1 (f k-1 ())/>;
Then, iteratively updating the set of labels
5. The method for classifying remote sensing targets for unsupervised learning of claim 1, further comprising: said step S6, iteratively updating the weights, in the firstkIn the training, the output result Pre of the last training model is used k-1 As the known information, a weight generated based on the known information is usedWeights of remote sensing target image in k-1 model training>Generating a weight set of the remote sensing target image during kth model training>Comprising the following steps:
iteratively updating the weights, in the firstkIn the training, the output result Pre of the last training model is used k-1 Generating weights as known information
wherein ,representing calculation numberkOutput of the-1 order model->With labeling->The larger the distance is, which indicates that the misclassification possibility of the remote sensing target image is larger, namely, the smaller weight is given in the subsequent training;
the L2 distance is a euclidean distance between two vectors, the distance is used to measure the difference between the two vectors, in this model, the difference between the prediction result of the model on the sample and the real label of the sample is used to measure, and aN N-dimensional vector a= [ a1, a2, a3, & gt, aN is set], B=[b1,b2.b3,...,bN]Let a be j Representing the j-th dimension of vector A, the L2 distance between the two vectors is defined as
Using the generated weightsAnd the firstkWeight of remote sensing target image in model training 1 time>Iteratively updating the weight set to generate the firstkWeight set of remote sensing target image during secondary model training>
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310594028.1A CN116310463B (en) | 2023-05-25 | 2023-05-25 | Remote sensing target classification method for unsupervised learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310594028.1A CN116310463B (en) | 2023-05-25 | 2023-05-25 | Remote sensing target classification method for unsupervised learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116310463A true CN116310463A (en) | 2023-06-23 |
CN116310463B CN116310463B (en) | 2024-01-26 |
Family
ID=86783665
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310594028.1A Active CN116310463B (en) | 2023-05-25 | 2023-05-25 | Remote sensing target classification method for unsupervised learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116310463B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096652A (en) * | 2016-06-12 | 2016-11-09 | 西安电子科技大学 | Based on sparse coding and the Classification of Polarimetric SAR Image method of small echo own coding device |
CN111753874A (en) * | 2020-05-15 | 2020-10-09 | 江苏大学 | Image scene classification method and system combined with semi-supervised clustering |
CN114120063A (en) * | 2021-11-29 | 2022-03-01 | 中国人民解放军陆军工程大学 | Unsupervised fine-grained image classification model training method and classification method based on clustering |
CN115761359A (en) * | 2022-11-24 | 2023-03-07 | 浙江大学 | Photovoltaic image defect classification method based on transfer learning and unsupervised learning method |
-
2023
- 2023-05-25 CN CN202310594028.1A patent/CN116310463B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106096652A (en) * | 2016-06-12 | 2016-11-09 | 西安电子科技大学 | Based on sparse coding and the Classification of Polarimetric SAR Image method of small echo own coding device |
CN111753874A (en) * | 2020-05-15 | 2020-10-09 | 江苏大学 | Image scene classification method and system combined with semi-supervised clustering |
CN114120063A (en) * | 2021-11-29 | 2022-03-01 | 中国人民解放军陆军工程大学 | Unsupervised fine-grained image classification model training method and classification method based on clustering |
CN115761359A (en) * | 2022-11-24 | 2023-03-07 | 浙江大学 | Photovoltaic image defect classification method based on transfer learning and unsupervised learning method |
Non-Patent Citations (1)
Title |
---|
张小娟;汪西莉: "完全残差连接与多尺度特征融合遥感图像分割", 遥感学报, no. 09 * |
Also Published As
Publication number | Publication date |
---|---|
CN116310463B (en) | 2024-01-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11960568B2 (en) | Model and method for multi-source domain adaptation by aligning partial features | |
CN110414462B (en) | Unsupervised cross-domain pedestrian re-identification method and system | |
CN111583263B (en) | Point cloud segmentation method based on joint dynamic graph convolution | |
WO2022037233A1 (en) | Small sample visual target identification method based on self-supervised knowledge transfer | |
CN114492574A (en) | Pseudo label loss unsupervised countermeasure domain adaptive picture classification method based on Gaussian uniform mixing model | |
CN113326731A (en) | Cross-domain pedestrian re-identification algorithm based on momentum network guidance | |
CN109492750B (en) | Zero sample image classification method based on convolutional neural network and factor space | |
CN107169117B (en) | Hand-drawn human motion retrieval method based on automatic encoder and DTW | |
CN111079847B (en) | Remote sensing image automatic labeling method based on deep learning | |
CN111274958B (en) | Pedestrian re-identification method and system with network parameter self-correction function | |
CN108345866B (en) | Pedestrian re-identification method based on deep feature learning | |
CN110188827B (en) | Scene recognition method based on convolutional neural network and recursive automatic encoder model | |
CN109840518B (en) | Visual tracking method combining classification and domain adaptation | |
CN111126361A (en) | SAR target identification method based on semi-supervised learning and feature constraint | |
CN113837238A (en) | Long-tail image identification method based on self-supervision and self-distillation | |
CN113222072A (en) | Lung X-ray image classification method based on K-means clustering and GAN | |
CN111239137B (en) | Grain quality detection method based on transfer learning and adaptive deep convolution neural network | |
CN116310647A (en) | Labor insurance object target detection method and system based on incremental learning | |
CN116910571B (en) | Open-domain adaptation method and system based on prototype comparison learning | |
CN112465016A (en) | Partial multi-mark learning method based on optimal distance between two adjacent marks | |
CN116310463B (en) | Remote sensing target classification method for unsupervised learning | |
CN116911377A (en) | Radiation source individual identification method, equipment and medium based on transfer learning | |
CN114417938B (en) | Electromagnetic target classification method embedded by using knowledge vector | |
CN113158878B (en) | Heterogeneous migration fault diagnosis method, system and model based on subspace | |
CN115661539A (en) | Less-sample image identification method embedded with uncertainty information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |